1

Prostate cancer detection and HIFU therapy monitoring using elastography

Table of contents

Prostate cancer detection and HIFU therapy monitoring using elastography 1 Introduction (Français)___________________________________________ 4

1. Détection du Cancer de la Prostate____________________________________________ 4

2. La thérapie du cancer de la prostate __________________________________________ 6

3. Imagerie de l’élasticité des tissus _____________________________________________ 7

4. Objectifs ________________________________________________________________ 10

Introduction (English)___________________________________________ 12 1. Prostate Cancer Detection ____________________________________________________ 12

2. Prostate Cancer Therapy _____________________________________________________ 14

3. Tissue Elasticity Imaging _____________________________________________________ 15

4. Objectives__________________________________________________________________ 17

1 Theory (Review) ____________________________________________ 19 1.1 Mechanics _______________________________________________________________ 20

1.1.1 Stress ______________________________________________________________________ 20 1.1.2 Strain ______________________________________________________________________ 22 1.1.3 Hooke’s law ________________________________________________________________ 23 1.1.4 Non-linearity ________________________________________________________________ 26 1.1.5 Visco-elasticity ______________________________________________________________ 26 1.1.6 Anisotropy__________________________________________________________________ 28

1.2 Basic Ultrasound Physics & Acoustic Imaging _________________________________ 28 1.2.1 Sound waves and the wave equation______________________________________________ 29 1.2.2 Intensity____________________________________________________________________ 31 1.2.3 Reflection __________________________________________________________________ 31 1.2.4 Scattering __________________________________________________________________ 32 1.2.5 Refraction __________________________________________________________________ 32 1.2.6 Diffraction__________________________________________________________________ 33 1.2.7 Interference and Speckle _______________________________________________________ 33 1.2.8 Absorption and Thermal Index (TI) ______________________________________________ 33 1.2.9 Attenuation _________________________________________________________________ 34 1.2.10 Cavitation and Mechanical Index (MI) ____________________________________________ 35 1.2.11 Echo ranging ________________________________________________________________ 35 1.2.12 Basic ultrasound instrumentation ________________________________________________ 36

1.3 Ultrasonic Elastography ___________________________________________________ 38 1.3.1 Principle ___________________________________________________________________ 38

2

1.3.2 Methods____________________________________________________________________ 39 1.3.3 System characterization________________________________________________________ 41

2 System Development & Characterization_________________________ 45 2.1 Data acquisition __________________________________________________________ 46

2.2 Data processing___________________________________________________________ 47 2.2.1 Displacement estimation using correlation at zero lag ________________________________ 49 2.2.2 Staggered strain estimates (interleaved gradient) ____________________________________ 51 2.2.3 Correction for lateral displacements ______________________________________________ 52 2.2.4 Adaptive stretching ___________________________________________________________ 52 2.2.5 Multi-compression ___________________________________________________________ 52

2.3 System validation on phantoms______________________________________________ 53 2.3.1 Homogeneous phantom________________________________________________________ 53 2.3.2 Phantom with a stiff inclusion___________________________________________________ 54

2.4 Experimental characterization ______________________________________________ 55 2.4.1 Materials and methods ________________________________________________________ 55 2.4.2 Simulation and experimental results ______________________________________________ 55 2.4.3 Conclusion__________________________________________________________________ 57

3 Application to prostate cancer detection _________________________ 59 3.1 In vitro__________________________________________________________________ 60

3.1.1 Objectives __________________________________________________________________ 60 3.1.2 Material and method __________________________________________________________ 61 3.1.3 Results_____________________________________________________________________ 63 3.1.4 Discussion __________________________________________________________________ 67 3.1.5 Conclusion__________________________________________________________________ 69

3.2 In vivo __________________________________________________________________ 69 3.2.1 Objectives __________________________________________________________________ 69 3.2.2 Material and method __________________________________________________________ 69 3.2.3 Results_____________________________________________________________________ 72 3.2.4 Discussion __________________________________________________________________ 77 3.2.5 Conclusion__________________________________________________________________ 79

4 Application to the Visualization of HIFU Lesions _________________ 80 4.1 HIFU therapy follow-up using elastography in vivo_____________________________ 81

4.1.1 Objectives __________________________________________________________________ 81 4.1.2 Material and Method __________________________________________________________ 81 4.1.3 Results_____________________________________________________________________ 82 4.1.4 Discussion __________________________________________________________________ 84 4.1.5 Conclusion__________________________________________________________________ 86

4.2 Monitoring of HIFU lesion formation by passive elastography in vitro _____________ 86 4.2.1 Objectives __________________________________________________________________ 86 4.2.2 Theory _____________________________________________________________________ 87 4.2.3 Materials and Methods ________________________________________________________ 91 4.2.4 Results_____________________________________________________________________ 94 4.2.5 Discussion _________________________________________________________________ 100 4.2.6 Conclusion_________________________________________________________________ 103

Discussion ___________________________________________________ 104 Conclusion (Français)__________________________________________ 107 Conclusion (English)___________________________________________ 111 References ___________________________________________________ 113

3

Annexes _____________________________________________________ 119 Annexe A: List of the publications of the author ___________________________________ 119

Annexe B: Teaching activities __________________________________________________ 121

Introduction

4

Introduction (Français)

1. Détection du Cancer de la Prostate

L’adénocarcinome prostatique est le cancer le plus fréquent et le second en terme de nombre de décès chez l’homme. Au niveau mondial, sa prévalence1 était estimée à plus d’un million de personnes en 1995, dont 896.000 cas dans les pays industrialisés. En 2002 aux USA, 189.000 nouveaux cas et 30.200 décès étaient attendus (American Red Cross, Prostate Cancer Statistics 2002). La probabilité de développer un cancer de la prostate entre la naissance et le décès est de 1/6. Son incidence2 est approximativement 85.000 en Europe (Bray et al. 2002) et 25.000 en France (Menegoz et al. 1997). Sa thérapie est plus efficace lorsque le cancer est diagnostiqué à un stade précoce. Cependant ce carcinome est souvent asymptomatique, et une technique de diagnostic fiable est requise. En effet, une évaluation précise de l’étendue du cancer est d’une importance fondamentale pour choisir une option thérapeutique appropriée. Aujourd’hui, aucune modalité d’imagerie ne permet une détection fiable de l’adénocarcinome prostatique, comme le montrent les récentes revues des techniques existantes (El-Gabry et al. 2001, Bangma et al. 2001). A l’âge adulte, la prostate mesure typiquement 3-5 cm en hauteur (entre la base et l’apex), 3-4 cm de large (entre la gauche et la droite) et 3-4 cm dans la direction antéro-postérieure (de la paroi rectale vers l’os du pubis), pour un poids moyen allant de 30 g à 40 g.

(a) (b) Figure 1: Coupes anatomiques présentant la prostate et les organes voisins (National Library of Medicine’s Visible Human Project) La glande prostatique se trouve entre la vessie et le rectum, et est traversée par l’urètre. Les figures 1a et 1b montrent l’anatomie du système urinaire et reproducteur, permettant de situer la prostate. L’anatomie zonale de la prostate, telle que définie par McNeal (1981), est montrée en figure 2. La prostate est composée de trois différents types de cellules:

1 Prévalence: Dénombrement ou proportion d’une population étant atteinte de la maladie 2 Incidence: Nombre de nouveaux cas dans une population pendant un intervalle de temps donné

Pubic bone

Prostate

Rectum

Muscle Fatty tissues

Introduction

5

- Les cellules glandulaires, qui produisent le liquide séminal servant à nourrir et véhiculer les spermatozoïdes.

- Les cellules musculaire lisses, qui se contractent lors de l’éjaculation et éjectent le liquide séminal vers l’urètre, où celui-ci se mélange aux spermatozoïdes et autres fluides pour former la semence.

- Les cellules stromales, qui forment la structure de la prostate. La glande prostatique peut être divisée en deux entités fonctionnelles séparées, qui se comportent différemment autant dans leurs fonctions normales que dans la maladie. La partie intérieure, généralement dénommée la zone de transition, est de nature glandulaire et stromale. C’est en général ici que se développe l’hyperplasie prostatique bénigne (BPH), qui est communément associée à des symptômes d’obstruction urinaire et d’hématurie3, et qui peut conduire à des anormalités du conduit urinaire supérieur du fait de cette obstruction. La littérature indique que seulement 5 à 15% des cancers de la prostate se développent dans la zone de transition. La portion restante, dénommée zone périphérique, est le plus souvent le siège de développement du cancer (85 à 95% des cas). Cette portion de la prostate n’est que rarement le foyer d’hématurie, de symptômes d’obstruction urinaire, ou d’anormalités du conduit supérieur.

(a) (b) Figure 2: (a) Coupe sagittale : Les plans de section coronal (C) et coronal oblique (OC) sont indiqués par une ligne double. Le verumontanum est indiqué en rouge. Le plan coronal suit les canaux éjaculatoires (e) et l’urètre distale (UD). Le plan coronal oblique suit l’urètre proximale (UP) jusqu’à la vessie (b). (Figure 2bCoupe coronale oblique de la prostate. Les canaux proximaux de la zone périphérique (P) rayonnent latéralement depuis le verumontanum (V). Les canaux de la zone de transition (T, zone ombrée) partent de la partie de l’urètre proche de la base du verumontanum se trouvant entre la zone périphérique et le sphincter (S). Les canaux péri-urètraux partant de l’urètre proximale sont confinés au stroma périurètral (U) et sont dirigés vers la vessie (au-dessus). Adapté de McNeal (1981). Il est depuis longtemps connu que les tumeurs prostatiques sont significativement plus dures que les tissus normaux. C’est pourquoi le toucher rectal (TR) est un outil majeur pour la détection de ce cancer. Cependant cette pratique a ses limites car elle reste subjective, et est limitée à la détection de grosses anormalités tissulaires rigides proches de la surface de la paroi rectale. La meilleure approche pour mettre en évidence la présence d’un cancer de la prostate est l’évaluation de la zone par toucher rectal, et l’utilisation d’un test sérum spécifique de la prostate (PSA). Le dosage de PSA se fait à partir d’une simple prise de sang.

3 Hématurie: sang dans l’urine

Introduction

6

Il fournit un niveau de suspicion de la présence du cancer et une indication globale de son développement, mais ne donne aucune information concernant l’emplacement, la taille et le type de tumeur. En se basant sur un seuil de 4 ng/dl, la sensibilité4 du test de PSA pour détecter un cancer considéré comme « cliniquement significatif » est de 63 à 83% (Harris et al. 2001). L’échographie endorectale (TRUS) fournit une information sur la réflectivité ultrasonore des tissus. Elle est couramment utilisée pour l’examen de la prostate. La fréquence centrale utilisée pour l’examen est généralement dans une gamme allant de 7 à 9 MHz. L’échographie seule a une sensibilité d’environ 71% à 92% pour les carcinomes prostatiques, et d’environ 60% à 85% pour les cas non cliniques. Sa spécificité5 varie entre 49% et 79% selon les études, et sa valeur prédictive positive6 (PPV) est d’environ 30% (Waterhouse et al. 1989), ce qui correspond à une faible fiabilité. De même, ni l’IRM (Quinlan et al. 1995) ni la tomographie par rayons X (El Gabry et al. 2001) ne sont considérés comme adéquats pour la détection du cancer de la prostate. L’IRM avec antenne endorectale a cependant l’avantage de pouvoir détecter l’extension du cancer hors de la prostate avec une sensibilité de 95% (Cornud et al. 2002). Aujourd’hui, l’examen de référence pour évaluer le cancer de la prostate consiste à ponctionner six biopsies, dites en sextant, dans six zones différentes de la prostate (apex, partie médiane et base dans le sens de la hauteur, et droite/gauche dans le sens de la largeur) en se guidant sur l’échographie endorectale pour augmenter les chances de détection. Une biopsie positive offre une certitude de 100% quant à la présence du cancer, mais cet examen reste une procédure d’échantillonnage qui peut passer à côté d’un foyer tumoral et ne permet donc pas d’exclure la présence du cancer en cas de biopsie négative.

2. La thérapie du cancer de la prostate

Le médecin doit déterminer quelle est la thérapie appropriée pour chaque patient. Une surveillance régulière éventuellement combinée avec une hormonothérapie peu éviter un traitement agressif aux patients présentant un cancer à bas risque. Cependant la plupart des patients désireront, et peuvent nécessiter, un traitement actif. Les techniques thérapeutiques standard incluent la prostatectomie radicale (ablation de la prostate par chirurgie), la radiothérapie externe, la Curiethérapie (Bangma et al. 2001) et depuis peu les ultrasons focalisés de haute intensité (HIFU). L’histoire des HIFU, aussi dénommés chirurgie par ultrasons focalisés (FUS), a commencé très tôt (Lynn et al. 1942, Fry et al. 1955), avant même que l’échographie n’ait vu le jour. Les premiers essais ont eu lieu sur des lésions du système nerveux. Cependant les recherches ont été limitées à l’époque par le manque de technique d’imagerie permettant de guider le traitement. Ce n’est qu’avec l’évolution récente de l’imagerie médicale que les HIFU ont connus un regain d’intérêt dans de nombreux domaines (ophtalmologie, prostate, foie, sein, cerveau). Les HIFU sont maintenant reconnus comme une alternative peu invasive à la

4 Sensibilité: Pourcentage de personnes malades pour lesquelles le test est positif (i.e. vrais positifs) 5 Spécificité: Pourcentage de personnes saines pour lesquels le test est négatif (i.e. vrais négatifs) 6 Valeur prédictive positive: Pourcentage de personnes présentant un test positif et qui sont réellement atteintes par la maladie

Introduction

7

chirurgie pour les cancers localisés de la prostate (Chapelon et al. 1992, Foster et al. 1993, Madersbacher et al. 1995, Gelet et al. 2001). Le principe repose sur la création de lésions thermiques localisées produites par un transducteur thérapeutique qui génère de brèves élévations de température. Le transducteur peut être focalisé, créant alors des lésions internes sans endommager la surface, ou non focalisé, créant alors des lésions qui s’étendent en profondeur à partir de la surface. Une intensité ultrasonore élevée est maintenue pendant quelques secondes de manière à ce que les tissus visés atteignent une dose thermique telle qu’une nécrose de coagulation locale et irréversible apparaisse. De multiples lésions élémentaires sont juxtaposées afin de traiter la totalité de la zone ciblée. La session thérapeutique est généralement guidée par échographie endorectale ou par IRM. Comme l’adénocarcinome prostatique ne peut pas être vu de façon fiable, et qu’il est généralement multi-focal (diffus), la totalité de la prostate est généralement traitée. Une limitation importante de cette technique est la difficulté d’évaluer l’étendue des lésions induites. La technique de référence aujourd’hui pour visualiser les lésions HIFU est l’IRM pondéré en T1 avec injection de produit de contraste (Hynynen et al. 1996, Rouvière et al. 2001). L’IRM permet aussi de cartographier les variations de température dans le corps pendant le traitement. Cependant son coût peut être prohibitif, et il requiert l’utilisation d’un équipement HIFU spécifique ne perturbant pas le champ magnétique. Un méthode ultrasonore est désirable car les ultrasons représentent un faible investissement et sont déjà communément utilisés pour l’imagerie de la prostate. Des lésions HIFU ont été décrites comme hypo-échogènes en échographie lors de précédentes études sur le foie de porc (Bush et al. 1993), mais aucune ne concerne la prostate humaine. Selon notre expérience (Gelet et al. 2001) les lésions HIFU dans la prostate humaine restent invisibles à l’échographie, même après dissipation des bulles de gaz hyper-échogènes qui sont créées durant la thérapie. Ces bulles envahissent généralement presque toute la prostate et rendent l’examen échographique difficile ; elles ne sont pas représentatives de la lésion HIFU. Un examen par Doppler de puissance avec injection de produit de contraste permet de voir les lésions HIFU car la zone traitée est dévascularisée. Cependant cette technique sous-estime le volume de la lésion (Sedelaar et al. 1999). Des technique ultrasonores telles que la mesure de vitesse du son (Lu et al. 1996), d’atténuation ultrasonore (Ribault et al. 1998) et de mesure de température par ultrasons (Van Baren et al. 1996) ont aussi été proposées pour évaluer l’étendue des lésions HIFU, mais n’ont rencontré qu’un succès limité.

3. Imagerie de l’élasticité des tissus

Pendant les 20 dernières années, de nombreuses études ont été conduites pour caractériser les propriétés mécaniques des tissus biologiques (Ophir et al. 1999, Greenleaf et al. 2003), qui ont souvent été considérés comme des matériaux élastiques linéaires homogènes et isotropiques. Ces attributs mécaniques incluent le module de cisaillement ou le module élastique (module d’Young), le coefficient de Poisson, ou n’importe laquelle des composantes de déformation ou de cisaillement obtenues en réponse à l’application d’une charge sur les tissus. En général, des lésions peuvent ou non posséder des propriétés ultrasonores qui les rendent détectables par échographie. Comme l’échogénicité et la rigidité des tissus sont en général non corrélés, l’imagerie de la rigidité ou de la déformation des tissus apporte une nouvelle information qui peut être liée à la structure pathologique ou non des tissus (Ophir et al. 1999). L’existence d’un contraste entre tissus sains et malins a été démontrée pour les cancers du sein et de la prostate (Krouskop et al. 1998), comme le montre la table 1.

Introduction

8

Type de tissu Module d’élasticité (kPa)

à 2% de déformation Module d’élasticité (kPa)

à 4% de déformation Normal (antérieur) 62 ± 17 63 ± 18 Normal (postérieur) 69 ± 17 70 ± 14 Hyperplasie prostatique bénigne (BPH)

36 ± 9 36 ± 11

Cancer 100 ± 20 221 ± 32 Table 1: Module d’élasticité des tissus prostatiques, mesuré avec un indenteur par chargement cyclique à 1 Hz (Krouskop et al. 1998), pour deux valeurs de précontrainte (i.e. de déformation initiale). La variation de module du cancer en fonction de la déformation observée est caractéristique d’un comportement mécanique non linéaire. Les méthodes d’imagerie d’élasticité par ultrasons se divisent en trois groupes principaux, se différenciant par le type d’excitation mécanique auquel les tissus sont soumis. Les principes et les limitations de ces trois groupes de méthodes sont explicitées ci-dessous : • Méthodes statiques (ou quasi-statiques) : Une compression (ou élongation) quasi-statique

est exercée sur les tissus, et les composantes du déplacement ou du tenseur des déformations sont estimées. Ces méthodes incluent l’élastographie (Ophir et al. 1991 O’Donnell et al. 1994, Ophir et al. 1999), basée sur une compression globale du milieu, et l’imagerie par force de radiation acoustique (ARFI) qui utilise la force de radiation statique créé par un transducteur focalisé pour exercer une compression locale (Nightingale et al. 2001). En élastographie comme en ARFI, une ou plusieurs composantes de déformation sont estimées, mais la contrainte n’est pas mesurable. Dans le cas de l’ARFI, l’amplitude de la force de radiation induite dans les tissus dépend de la pression acoustique locale et des propriétés acoustiques du milieu. Dans les tissus biologiques, cette amplitude n’est généralement pas connue.

• Méthodes vibratoires : Les tissus sont soumis à une vibration monochromatique de basse fréquence (typiquement de quelques Hz à quelques kHz), et l’amplitude et éventuellement la phase des déplacements ainsi générés sont mesurées. La vibration est généralement induite de façon mécanique par un vibreur, ou de manière acoustique par une force de radiation monochromatique. Parmi ces techniques, la sonoélasticité est basée sur une mesure de déplacement par effet Doppler (Krouskop et al. 1987, Lerner et al. 1988, Lerner et al. 1990, Yamakoshi et al. 1990). La vibro-acoustographie (Fatemi et al. 1998) repose sur une mesure de l’amplitude de l’onde acoustique basse fréquence émise par des tissus soumis à une force de radiation vibratoire. La réponse obtenue par ces techniques dépend des propriétés élastiques du milieu mais est aussi perturbée par les ondes stationnaires inhérentes à une excitation monochromatique. La vibro-acoustographie utilise donc généralement des trains d’ondes courts pour tenter d’effectuer la mesure avant l’établissement du régime stationnaire, alors que la sonoélasticité utilise plusieurs mesures à des fréquences différentes afin de moyenner ces perturbations. Tout comme pour l’ARFI, la force de radiation dépend du milieu et n’est généralement pas connue dans les tissus biologiques.

• Méthodes transitoires : L’élastographie transitoire (Sandrin et al. 1999, Bercoff et al. 2003) repose sur la mesure de la vitesse de propagation et de l’amplitude d’une onde de cisaillement transitoire (pulsée) pour déterminer les propriétés visco-élastiques du milieu. Une barrette linéaire classique connectée à un échographe ultra-rapide (3000 à 5000 images par seconde) est utilisée pour visualiser l’onde de cisaillement, qui se déplace typiquement à une vitesse comprise entre 1 et 10 ms-1. La rapidité de cet échographe est obtenue en transmettant simultanément sur tous les éléments de la barrette afin de former une onde plane, donc au détriment de la focalisation en transmission. Cette méthode a

Introduction

9

l’avantage d’utiliser une excitation localisée dans l’espace et dans le temps, ce qui simplifie la résolution du problème inverse (identification des propriétés mécaniques).

Les techniques d’estimation des déplacement ou déformations quasi-statiques sont basées sur une estimation des délais temporels entre des signaux ultrasonores acquis avant et après déformation. Le déplacement local des tissus est estimé à partir du délai existant entre des segments des signaux pré- et post-compression, et la déformation est estimée à partir du gradient des déplacements. Cette techniques requiert que le déplacement local des tissus se fasse dans la même direction que la propagation de l’onde ultrasonore. Dans le cas de la prostate, la sonoélasticité est capable de détecter le cancer in vitro (Rubens et al. 1995). Une étude sur des prostates de chien in vitro a permis de démontrer la faisabilité de l’élastographie prostatique en utilisant une barrette linéaire couplée à une plaque plane de compression (Kallel et al. 1999). La faisabilité de l’élastographie de la prostate in vivo et en temps réel a été démontrée la même année en utilisant une sonde endorectale standard tenue manuellement pour comprimer la prostate (Lorenz et al. 1999). Les résultats initiaux sur 170 patients ayant subi une prostatectomie radicale après examen élastographique indiquent une sensibilité d’environ 76% et une spécificité d’environ 84% en combinant élastographie et échographie. (Pesavento et al. 2001). L’élastographie transitoire apparaît comme une technique prometteuse mais n’a pas encore été testée sur le prostate. Concernant les HIFU, les études récentes menées in vitro ont montré que la zone de coagulation était durcie par le traitement, ce qui la rend détectable par élastographie (Kallel et al. 1999, Righetti et al. 1999). Ce n’est donc pas une surprise si la sonoélasticité (dénommée « élastométrie dynamique » dans l’article ci-après cité) a aussi permis de visualiser ces lésions (Shi et al. 1999). La même étude a montré qu’il existe un contraste de module allant de 6 à 12 entre du foie sain et une lésion (Table 2). La technique ARFI a également été proposée pour détecter la zone de coagulation pendant la thérapie HIFU, en utilisant la force de radiation produite par le transducteur de thérapie comme excitation mécanique (Lizzi et al. 2003). Type de tissu Normal Lésion HIFU Lobe latéral droit 2.2 19.8 Lobe latéral gauche 1.7 20.7 Lobe médian gauche 4.2 23.6 Table 2 : Modules d’Young, en kPa, mesurés par sonoélasticité dans le foie de porc (Shi et al. 1999) L’IRM aussi permet de visualiser les propriétés élastiques des tissus. L’élastographie par résonance magnétique (MRE) utilise l’onde de cisaillement obtenue par excitation monochromatique pour mesurer le déplacement local des tissus et tenter d’en déduire le module. La MRE a ainsi permis de visualiser les lésions thermiques induites in vitro dans du muscle (Wu et al. 2001) et d’estimer le module de cisaillement des tissus sains et coagulés en fonction de la fréquence d’excitation. (Fig. 3) ainsi que de la température.

Introduction

10

Figure 3 : Module de cisaillement (kPa) des tissus sains (courbe du bas) et des tissus coagulés (en haut) estimé pour diverses fréquences par MRE dans le muscle de porc (Wu et al. 2001).

4. Objectifs

Lors de la thérapie HIFU du cancer de la prostate, le médecin doit choisir la zone à traiter. Il choisi généralement de traiter la totalité de la glande car l’adénocarcinome prostatique ne peut pas être visualisé de manière fiable et est souvent multifocal (diffus). Un objectif important est de traiter efficacement la face postérieure de la prostate, qui est en contact avec la paroi rectale, sans endommager celle-ci. Les statistiques indiquant que la plupart des cancers de la prostate se trouvent dans la zone périphérique, il faut prendre soin d’éviter de laisser des tissus non traités dans cette portion de la glande. Mais si la lésion HIFU abîme la paroi rectale, des complications graves (fistule) peuvent survenir. Les faisceaux de nerfs qui entourent la prostate sont aussi susceptibles d’être endommagés, conduisant éventuellement à l’incontinence et/ou à l’impuissance. Si l’on pouvait s’assurer que les tumeurs sont loin des nerfs, il est probable qu’un traitement efficace pourrait alors être effectué sans pour autant les endommager. Mais lorsqu’une tumeur se trouve proche des nerfs, elle est donc aussi proche de la périphérie de la prostate et le risque de voir le cancer sortir de la capsule prostatique pour envahir les organes voisins est grand. Dans ce cas, il peut être nécessaire d’augmenter la dose thermique autour de ces organes sensibles, et risque de les endommager peut devenir le prix à payer pour éliminer efficacement le cancer. Le question importante, pour les HIFU autant que pour les autres options thérapeutiques, est donc de déterminer si les nerfs peuvent être épargnés ou non tout en assurant le traitement le plus efficace possible. Il apparaît donc qu’une modalité d’imagerie permettant de localiser les foyers tumoraux de façon fiable dans la prostate donnerait au chirurgien l’opportunité d’évaluer les risques et bénéfices de chaque option et de prendre ainsi une décision appropriée. Une fois le traitement HIFU achevé, l’étendue de la lésion et l’éventuelle présence de tissus non traités ne peuvent pas être évalués, à moins de disposer d’un IRM. Cependant même l’IRM ne permettrait pas de voir d’éventuels foyers cancéreux résiduels (Rouvière et al. 2001). Dans la pratique courante, le patient est suivi par dosage de PSA et biopsies, effectués régulièrement, de façon à détecter une éventuelle récurrence du cancer. Une modalité d’imagerie compatible avec l’équipement HIFU et capable de visualiser les lésions thermiques ainsi que de possibles zones mal ou non traitées est donc souhaitable. Elle permettrait de détecter les zones non suffisamment coagulées alors que le patient est encore en salle d’opération, donc de les re-traiter immédiatement, économisant ainsi au patient une

Introduction

11

éventuelle session HIFU supplémentaire. L’efficacité du traitement pourrait alors être assurée pour chaque patient. L’objectif de la présente étude découle directement des deux paragraphes précédents : elle vise à montrer la faisabilité de l’élastographie in vivo en vue de la localisation du cancer de la prostate (chapitre 3), et de la visualisation des lésions HIFU (chapitre 4). Enfin la faisabilité d’une nouvelle technique, dénommée élastographie passive (i.e. sans excitation mécanique), pour visualiser la formation d’une lésion HIFU élémentaire a été étudiée (chapitre 4.2). Cette technique utilise les variations de la vitesse du son avec la température combinée avec l’expansion thermique du milieu pour former une image. Les résultats attendus à plus ou moins long terme sont les suivants : (1) une amélioration du diagnostic du cancer de la prostate, (2) une efficacité accrue des diverses options thérapeutiques et la possibilité d’épargner les tissus périprostatiques grâce à une localisation précise du cancer dans la prostate, (3) la possibilité d’évaluer l’efficacité d’un traitement HIFU pour permettre un complément de traitement immédiat si nécessaire, et (4) de faciliter le développement de nouvelles applications des HIFU grâce à la possibilité de visualiser les lésions. Dans une démarche qualité, les références aux documents internes (expériences, données et rapports) cités sont indiquées en notes de bas de page.

Introduction

12

Introduction (English) 1. Prostate Cancer Detection

Adenocarcinoma of the prostate is the most prevalent malignant cancer and the second cause of cancer-specific death in men. Worldwide, its prevalence7 was 1,014,000 in 1995, with 896,000 cases in industrialized countries. In 2002 in the USA, 189,000 new cases and 30,200 deaths were expected (American Red Cross, Prostate Cancer Statistics 2002). The probability of developing prostate cancer from birth to death is 1 in 6. Its annual incidence8 is approximately 85,000 in Europe (Bray et al. 2002) and 25,000 in France (Menegoz et al. 1997). Its therapy is more effective when cancer is diagnosed at an early stage, but this carcinoma is usually asymptomatic and therefore reliable diagnostic modalities are required. Accurate assessment of the local extent of the disease is fundamentally important in the selection of appropriate local treatment modalities. There is currently a strong need for an imaging modality that could allow reliable detection of prostate adenocarcinoma. A thorough review of the strengths and weaknesses of existing and emerging imaging modalities for prostate cancer can be found in El-Gabry et al. (2001) and in Bangma et al. (2001).

Figure 1: The prostate and surrounding organs (from National Library of Medicine’s Visible Human Project) The prostate gland is located between the bladder and the rectum and wraps around the urethra. Figure 1 shows the prostate and surrounding organs. The zonal anatomy of the prostate is shown in Fig. 2 (adapted from McNeal 1981). The prostate is basically composed of three different cell types: - Glandular cells, which produce a milky fluid that liquefies semen. - Smooth muscle cells, which contract during sex and squeeze the fluid from the glandular

cells into the urethra, where it mixes with sperm and other fluids to make semen. - Stromal cells, which form the structure of the prostate. The prostate gland can be functionally divided into two separate entities that act differently in normal function as well as in disease. The inner part most commonly known as the transitional zone is both glandular and stromal in nature. It is the usual site of development of 7 Prevalence: proportion of people who are found to be with the disease 8 Incidence: rate showing how many new cases occurred in a population during a specific time interval

Pubic bone

Prostate

Rectum

Muscle Fatty tissues

Introduction

13

benign prostatic hyperplasia (BPH), which is usually associated with benign obstructive urinary symptoms (as well as hematuria9) and may lead to upper urinary tract abnormalities. The transition zone is associated with development of prostate cancer only 5-15% of the time. The remaining portion is the peripheral zone that is most commonly associated with the development of prostate cancer (85-95%). This portion of the prostate is rarely associated with hematuria, urinary obstructive symptoms or upper tract abnormalities.

(a) (b) Figure 2: (a) Sagittal Plane : Relationships to other planes of section, coronal (C) and oblique coronal (OC), are shown by double lines. The verumontanum is red. The coronal plane follows the ejaculatory ducts (e) and the distal urethra (ud). The oblique coronal plane follows the proximal urethra (UP) to the bladder (B). (b) Oblique coronal (OC) plane of the prostate. The most proximal ducts of the peripheral zone (P) radiate laterally from the verumontanum (V). Ducts of the transition zone (T, shaded area) exist from urethra just proximal to the base of the verumontanum, lying between the peripheral zone and the sphincter (S). Periurethral ducts arising along the proximal urethra are confined to the periurethral stroma (U) and run proximally toward the bladder (on top). Adapted from McNeal (1981) The best approach to the detection of prostate cancer is by evaluation of this area through digital rectal examination (DRE) and use of serum prostate-specific antigen (PSA). The PSA measurement provides a level of suspicion for the presence of cancer, and an overall indication of its development, but it does not give any information about the location, size and type of the tumor. If a cutoff point of PSA > 4 ng/dl is used, PSA screening has an estimated sensitivity10 of 63% to 83% for "clinically significant" disease using pathological criteria (Harris et al. 2001). It has long been known that tumors of the prostate are significantly stiffer than normal surrounding tissues. This is why digital rectal examination is a major tool in the diagnosis of prostate cancer. However this practice remains subjective, and is limited only to large, stiff tissue abnormalities located close to the rectal wall. Transrectal ultrasonography (TRUS) provides information about the relative ultrasonic reflectivity of tissues, and is routinely used in prostate examination. For TRUS alone, sensitivity ranged from 71% to 92% for prostatic carcinoma and 60% to 85% for subclinical disease. Specificity11 values ranged from 49% to 79%, and positive predictive values12 (PPV) in the 30% range have been reported (Waterhouse et al. 1989). Neither MRI (Quinlan et al. 9 Hematuria: blood in the urine 10 Sensitivity: percentage of people with disease who test positive (i.e. true positive rate) 11 Specificity: percentage of healthy people who test negative (i.e. true negative rate) 12 Positive predictive value: the percentage of people with positive test who turn out to have the disease

Introduction

14

1995) nor CT scans (El Gabry et al. 2001) are considered adequate for prostate cancer detection. However endorectal MRI permits the determination of occult extraprostatic spread in a given individual with a 95% specificity (Cornud et al. 2002). The gold standard remains sextant biopsies guided by transrectal ultrasound (TRUS) examination. This process is a sampling procedure that cannot exclude cancer with 100% certainty.

2. Prostate Cancer Therapy

The physician has to determine what the appropriate therapy for each patient should be. Watchful waiting combined with hormonal therapy can avoid aggressive treatments to low-risk patients. However most patients will expect, and may require, active treatment. Standard therapeutic techniques include radical prostatectomy (surgery), external beam radiotherapy, brachytherapy (Bangma et al. 2001) and more recently high intensity focused ultrasound (HIFU). The destructive ability of high intensity ultrasound had been recognized from the time of Paul Langevin when he noted destruction of school of fishes in the sea and pain induced in the hand when placed in a water tank insonated with high intensity ultrasound. Lynn and Putnam (1942) successfully used ultrasound waves to destroy brain tissue in animals. It was considered as the earliest experiment of therapeutic ultrasound application on biological tissues. But at the time research was limited by the lack of an efficient imaging technique to guide the treatment, and no significant improvement was achieved until the onset of reliable imaging modalities in the 1960’s. HIFU is now established as an effective minimally invasive therapy for localized prostate cancer (Chapelon et al. 1992, Foster et al. 1993, Madersbacher et al. 1995, Gelet et al. 2001). This technique, also called focused ultrasound surgery (FUS), creates brief local elevations of temperature at the focus of an ultrasonic therapy transducer. High ultrasonic intensity is maintained for a few seconds in order for the target tissue to reach a thermal dose that creates irreversible local coagulation necrosis. Multiple elementary coagulation lesions are generated to treat the whole prostate. The therapy session is guided by transrectal diagnostic ultrasound. Because prostate cancer cannot be accurately detected and is usually multifocal, the whole gland is generally treated. Today a major limitation of this technique lies in the difficulty of assessing the extent of the induced lesions. The most effective imaging modality for HIFU lesion imaging is MRI (Hynynen et al. 1996, Rouvière et al. 2001), but its cost may be prohibitive, and MRI-compatible HIFU equipment is required. MRI was also proposed to monitor the temperature changes in the body during HIFU application (Hynynen et al. 1996). Ultrasound is inexpensive and is commonly used in prostate imaging, therefore an ultrasonic imaging method for HIFU lesion imaging seems desirable. Earlier studies on porcine liver reported HIFU-induced lesions as hypo-echoic areas in sonograms (Bush et al. 1993), but none deal with the human prostate. From our experience (Gelet et al. 2001) HIFU lesions in human prostate remain invisible on the sonograms after the dissipation of hyper-echoic gas bubbles created during the therapy. These bubbles are not representative of the lesion. The lesions are devascularised and can be seen using contrast-enhanced power Doppler, but this technique underestimates the extent of the HIFU lesion (Sedelaar et al. 1999). Ultrasonic techniques such as speed of sound (Lu et al. 1996), ultrasonic attenuation (Ribault et al. 1998) and temperature measurements using ultrasound (Van Baren et al. 1996) have also been proposed to assess the extent of the HIFU-induced lesions, with limited success.

Introduction

15

3. Tissue Elasticity Imaging

Over the past 20 years there have been numerous investigations conducted to characterize the mechanical properties of biological tissue systems, which have often been idealized as homogeneous, isotropic and linear elastic materials. These mechanical attributes may include the shear or elastic modulus (Young’s modulus), the Poisson’s ratio, or any of the longitudinal or shear strains that occur in tissues as a response to an applied load. In general, lesions may or may not possess echogenic properties that would make them ultrasonically detectable. Since the echogenicity and the stiffness of tissue are generally uncorrelated, it is expected that imaging tissue stiffness or strain will provide new information that may be related to pathological tissue structure (Ophir et al. 1999). Krouskop et al. (1998) showed that a stiffness contrast exists between malignant and normal tissues in the breast and in the prostate (Table 1). Tissue type Elastic modulus at 2%

precompression (kPa) Elastic modulus at 4% precompression (kPa)

Normal (anterior) 62 ± 17 63 ± 18 Normal (posterior) 69 ± 17 70 ± 14 Benign prostate hyperplasia (BPH) 36 ± 9 36 ± 11 Cancer 100 ± 20 221 ± 32 Table 1: Stiffness of malignant and normal prostate tissues, measured during cyclic loading at 1 Hz (Krouskop et al. 1998) A review of selected elasticity imaging techniques was recently published (Greenleaf et al. 2003). Basically, tissue elasticity imaging methods based on ultrasonics fall into three main groups: • Methods where a quasi-static compression is applied to the tissue and the resulting

components of displacement or of the strain tensor are estimated. These methods include elastography, based on a global compression of the medium (Ophir et al. 1991, O’Donnell et al. 1994, Ophir et al. 1999), and acoustic radiation force imaging (ARFI) that uses localized compression (Nightingale et al. 2001).

• Methods based on a monochromatic low-frequency vibration such as sonoelasticity (Krouskop et al. 1987, Lerner et al. 1988, Lerner et al. 1990, Yamakoshi et al. 1990) which uses Doppler signals to estimate tissue displacement, and vibro-acoustography (Fatemi et al. 1998) which uses ultrasound-stimulated acoustic emission. Stationary waves inherent to CW excitation have to be avoided using short bursts or have to be accounted for in order to be enable to directly relate the measured signal to the elastic properties.

• Transient elastography (Sandrin et al. 1999) relies on the observation of the propagation of a transient (pulsed) shear wave to determine the visco-elastic properties of the tissues. This method offers the advantage of producing a spatially and temporally localized excitation, independent of boundary conditions, that propagates through the medium so that a whole volume can be scanned rapidly. The solution of the inverse problem (identification of the mechanical parameters) is greatly simplified in this case. The implementation of the technique requires the use of an ultrafast ultrasound scanner to track the propagation of the shear wave. The ultrafast scanner is based on an unfocused

Introduction

16

plane wave illumination of the medium instead of adjacent focused beams, resulting in a reduction in resolution.

Among the techniques based on the quasi-static estimation of tissue strain, elastography is based on estimating the tissue strain using time delay estimation. An ultrasound pulse is emitted in the tissues and the backscattered signal is received. Then a static compression is applied to the tissues, and a second transmit/receive sequence is performed. Local tissue displacements are estimated from the time delays between gated pre- and post-compression echo signals, whose axial gradient is then computed to estimate and display the local strain. Rubens et al. (1995) have shown that sonoelasticity was able to detect prostate cancer in vitro. Prostate elastography was also proposed for prostate imaging (Kallel et al. 1999), who obtained elastograms of canine prostates in vitro. Elastography was implemented as a real-time strain imaging modality and initial results showed a sensitivity for prostate cancer detection of approximately 76% and a specificity of 84% using a prospective study on 170 patients (Lorenz et al. 1999, Pesavento et al. 2001). Transient elastography appears as a promising technique for tumor detection but has not been tested on prostate yet (Sandrin et al. 1999, Bercoff et al. 2003). Recent studies have shown HIFU application stiffens the target area, making them detectable by imaging techniques that depict the mechanical properties or behavior of tissues (Kallel et al. 1999, Righetti et al. 1999). Shi et al. (1999) detected HIFU lesions in porcine liver in vitro using sonoelasticity, also named dynamic elastometry, a technique where Doppler signals are used to measure the velocity of the tissues while they are being vibrated at low frequency. Since velocity depends on stiffness, this technique provides a measurement related to the elastic properties of the tissues. They also reported that HIFU lesions in porcine liver are approximately six to twelve times stiffer than the normal tissue liver (Table 2). Comparison between tables 1 and 2 shows that the Young’s modulus of biological tissues can easily contrast by 2 orders of magnitude depending on the tissue type and the amount of pre-compression. Tissue type Young’s modulus in normal

tissue (kPa) Young’s modulus in HIFU lesion

(kPa) Right lateral lobe 2.2 19.8 Left lateral lobe 1.7 20.7 Left medial lobe 4.2 23.6 Table 2 : Stiffness of HIFU lesions in porcine liver (Shi et al. 1999) Magnetic Resonance Elastography (MRE) was also recently proposed to assess the effects of thermal tissue ablation (Wu et al. 2001) by measuring mechanical properties of the lesion in ex vivo porcine muscle using shear waves (Fig 3). Wu et al. also reported a dependence of the elastic modulus on tissue temperature. The local assessment of tissue stiffness during HIFU therapy is also possible with the ARFI technique, using the radiation force of the therapeutic US beam as an elastographic “push” (Lizzi et al. 2003). However the radiation force depends on the ultrasonic properties of the medium and it can only be roughly estimated, therefore precluding the determination of the absolute value of the modulus.

Introduction

17

Figure 3 : Shear (elastic) moduli of the HIFU lesion and normal tissues in the porcine specimen at various shear wave frequencies (Wu et al. 2001)

4. Objectives

When performing HIFU therapy of localized prostate cancer, the clinician has to choose the target area. He usually targets the whole gland because prostate cancer cannot be reliably detected and is usually multifocal (diffuse). An important issue is to properly treat the posterior face of the prostate, which is in contact with the rectal wall, without damaging the rectal wall. Most cancers appear on the posterior face of the prostate, so care has to be taken not to leave untreated tissues in this area. But if the HIFU lesion damages the rectal wall, severe complications (fistula) may arise. Nerve bundles that surround the prostate are also likely to be damaged, resulting in possible incontinence and impotence. If the cancer is located away from these tissues, it is likely that effective therapy could be achieved without damaging them. If the cancer is close to these tissues, increasing the thermal dose around them and running the risk of damaging them may be the price to pay for efficiently eliminating the cancer. An important question is therefore to assess if these tissues could be spared or not, while ensuring that efficient therapy is performed. The use of a reliable imaging modality for prostate cancer detection would give the clinician the opportunity to assess the risks and benefits of both options and to make a appropriate decision. Once the therapy has been performed, the extent of the HIFU lesion, the presence of eventual untreated areas cannot be determined unless MRI is available. Even using MRI, eventual residual cancer foci cannot be seen (Rouvière et al. 2001). PSA measurements and biopsies are used on a regular basis for patient follow-up, in order to detect the eventual recurrence of the cancer. An imaging modality capable of visualizing HIFU lesions and compatible with the HIFU therapy system is desirable. It would allow for the detection and re-treatment of untreated areas while the patient is still in the operating room. Efficient HIFU coverage of the gland would then be ensured. The aim of this study is to investigate the feasibility of using elastography in vivo for the localization of prostate tumors (Chapter 3), and for the visualization of HIFU-induced lesions (Chapter 4). Elastography is expected (1) to improve prostate cancer diagnosis, (2) to improve HIFU targeting on the cancer while sparing important surrounding tissues, and (3) to allow for HIFU treatment efficiency and for immediate re-treatment if necessary. “Passive” elastography (i.e. without an external mechanical stress) was also used to monitor the formation of HIFU lesions in porcine liver in vitro. This technique relies on changes in speed

Introduction

18

of sound and on thermal expansion during tissue heating, and is described in detail in Chapter 4.2. For quality insurance purposes, references to internal documents (experimental data and reports) cited in the text are given in footnotes. At the beginning of each section, publications based on this particular part of the work are also given in footnotes.

Chapter 1. Theory

19

1 Theory (Review) Ce chapitre est dédié aux rappels théoriques nécessaires pour d’appréhender les principes de l’élastographie ultrasonore. Il a pour seule ambition de fournir au lecteur un « glossaire » des termes et de rapides explications des principes de base, et ne se substitue en aucun cas à la lecture des ouvrages et publications référencés pour qui souhaite approfondir ses connaissances. L’élastographie étant une modalité d’imagerie de l’élasticité des tissus, ou plus précisément du comportement élastique des tissus soumis à une contrainte donnée, la première partie du chapitre s’intéresse à la théorie de la mécanique des milieux élastiques. Il y est rappelé les notions de tenseur, de contrainte et de déformation, ainsi que la loi de Hooke généralisée qui relie ces deux dernier par l’intermédiaire des 81 composantes du tenseur d’élasticité. Celles-ci représentent les propriétés élastiques de ce milieu. Les hypothèses permettant de réduire le nombre de constantes de 81 à 2, les constantes de Lamé, sont alors exposées. Sont alors définis le module de compression et de le module cisaillement, ainsi que le module d’Young (module d’élasticité) et de coefficient de Poisson qui en dérivent, et la loi de Hooke. Les paragraphes suivants illustrent certaines propriétés de certains tissus biologiques qui violent les hypothèses précédemment énoncées, obligeant ainsi à être précautionneux dans le choix d’un modèle mécanique pour les tissus observés. La variation du module d’élasticité avec la déformation appliquée, c’est-à-dire la non-linéarité, montre qu’il peut être difficile ou même impossible de pouvoir associer un type de tissu avec un module donné. Le comportement visco-élastique est ensuite défini, et illustre qu’un module peut aussi dépendre de la vitesse (ou la fréquence) de chargement. Il est enfin rappelé que de nombreux tissus biologiques, telles les fibres musculaires, possèdent une orientation spatiale et sont donc anisotropes. La seconde partie de ce chapitre est consacrée aux rappels d’acoustique et propagation d’ondes appliqués au domaine de l’imagerie médicale par ultrasons. Ce chapitre commence avec les équations de propagation pour ensuite définir la vitesse de propagation et l’impédance acoustique, puis passe en revue les phénomènes de réflexion, de diffusion, de réfraction, de diffraction, d’interférences, d’absorption, et de cavitation. Finalement le fonctionnement d’un échographe est décrit. La troisième et dernière partie du chapitre est spécifiquement consacrée à l’élastographie ultrasonore, ou imagerie des déformations des tissus biologiques. Les principes fondamentaux de l’évaluation de la déformation à partir des délais temporels entre les signaux RF acquis à deux états de contrainte différents y sont énoncés. Diverses méthodes employées pour calculer la déformation sont alors passées en revue. Puis les critères de caractérisation du système que sont la variance des délais temporels, le rapport signal-sur-bruit élastographique (SNRe), le filtre des déformations (strain filter), le rapport contraste-sur-bruit (CNRe), et la résolution sont définis. L’altération de l’estimation des délais due à la nature 3D des déplacements des tissus est ensuite expliquée, et le chapitre se termine sur un lexique des artefacts couramment rencontrés en élastographie.

Chapter 1. Theory

20

Elastography is an ultrasonic-based imaging method designed to depict the mechanical behavior of soft tissues. As a prerequisite of this study, the basics of acoustics and mechanics are presented in this chapter. Then the principles and methods of elastography are described.

1.1 Mechanics

1.1.1 Stress

In a continuous deformable body acted upon by external forces, an imaginary internal incision is made over a small plane element ∆S in the body. The result will be that, unless the stress over the element of area is one of direct compression only, there will be some relative movement between the opposing faces of the incised area. In general, these faces will draw apart slightly owing to what are termed internal tensile stresses. They will also slide over each other to some extent as a result of what are termed shear stresses (Wang 1953, Crandall et al. 1959, Hayden et al. 1965, Varga 1966). Before the incision, each face has been exerting a force F

r∆ on the other face. The ratio

∆F/∆S is then called the average stress acting over ∆S. If this is uniform, it will be acting with equal intensity at all points of the elementary surface. If the stress in the material is not uniform, however, the stress vector t

r acting at a point P over the elementary surface ∆S is

defined as the limit

SFt

dS ∆∆

=→

rr

0lim Equation 1.1

The components of the stress vector have the same unit as pressures (Pascals). The orientation of the surface element ∆S may be defined by the unit vector nr normal to the surface and oriented towards the other face. The stresses at point P over the surface ∆S may be resolved into a tensile stress σn given by

ntnrr.=σ Equation 1.2

and a shear stress σt given by ntttt

rrrr.. −=σ Equation 1.3

A negative value for σn will indicate a compressive direct stress along the normal, imparted from the opposite face, whereas a positive value will indicate a tensile stress.

Chapter 1. Theory

21

x3

x1

x2

σ33

σ32σ31

3tr

2tr

1tr

σ23

σ22σ21

σ13

σ12

σ11

Figure 1.1 – The nine stress tensor components

For the full specification of the state of stress at any given point, all components of the stress must be given. In the Cartesian coordinate system Ox1x2x3, there exists at the point P a general component of stress σ12 which denotes the limiting force-to-area intensity ratio, in the sense of equation 1.1, of a component of force along the positive x2 direction, and acting on an element of plane area with normal rising from it in the positive x1 direction. Since such a general component of the stress has two subscripts, there are altogether nine such components (shown in fig. 1.1) that compose the stress tensor Σ

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡=Σ

332313

322212

312111

σσσσσσσσσ

Equation 1.4

It is evident from the definition of stress components that only those with the two like subscripts represent normal stresses, tensile or compressive, according to whether they are positive or negative, while those with two unlike subscripts represent shear stresses.

Figure 1.2 – The elementary tetrahedron Given the general components of stress at P, the stress components, along the direction of the coordinate axes, on any plane element ∆S with a normal vector nr can readily be derived. To this end, the plane element ∆S is assumed to have the shape of a triangle, and the origin of the coordinate system is chosen so that the corners A,B,C of the triangle come to lie on the

x1

x2

x3

A

C

B

Chapter 1. Theory

22

coordinate axes, with OABC forming an elementary tetrahedron (fig. 1.2). Considering the equilibrium of the tetrahedron under the forces acting on it, the values of the stress components on ∆S are obtained by equating the forces along the directions of the coordinate axes (Varga 1966):

nt rr⋅Σ= Equation 1.5

For normal mechanics, the condition of equilibrium requires in addition that the moments of the forces acting upon its faces shall vanish. As a consequence, the stress tensor is symmetric, i.e. for any (i,j), σij = σji. This has six independent general components of stress, three normal stresses, and three shear stresses; these fully define the state of stress at any given point. Note: For some geometries, it is often convenient to use cylindrical coordinates (r,θ,z). The stress tensor is still symmetric and becomes:

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡=Σ

zzzrz

zr

zrrrr

σσσσσσσσσ

θ

θθθθ

θ

1.1.2 Strain

Strain is defined as the relative variation of the distance between particles of a continuous medium undergoing a mechanical stress. Consider 2 points P and Q in an orthonormal set of axes. Note dx1, dx2 and dx3 the 3 components of vector PQ . The distance between these two points is:

∑=

==3

1

2

iidxPQdl Equation 1.6

When the solid is deformed, these two points respectively move to P’ and Q’. Note

')( PPPu = the displacement of point P, ')( QQQu = the displacement of point Q, and

)()(),( PuQuQPdu −= the variation in displacement between points P and Q. The distance between points P’ and Q’ is:

∑=

+==3

1

2)('''i

ii dudxQPdl Equation 1.7

Which can be written as:

( ) ∑∑==

++=3

1

23

1

22 2'i

ii

ii dudxdudldl Equation 1.8

The components of the displacement vector ),( QPdu can be written as:

∑= ∂

∂=

3

1jj

j

ii dx

xudu for i=1, 2, 3 Equation 1.9

Chapter 1. Theory

23

Then equation 1.8 becomes

∑∑= =

+=3

1

3

1

22'i j

jiij dxdxdldl ε Equation 1.10

where the strain component εij is defined as:

⎟⎟⎠

⎞⎜⎜⎝

⎛

∂∂∂

+∂

∂+

∂∂

= ∑=

3

1

2

21

k ji

k

i

j

j

iij xx

uxu

xuε Equation 1.11

Strain is a tensor composed of 9 components:

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡=

332313

322212

312111

εεεεεεεεε

ε Equation 1.12

Note that equation 1.20 is symmetric, i.e. ∀{i,j}∈{1,2,3}2, εij = εji . As a consequence, only 6 independent components are required to completely define the strain tensor: Strain is a ratio between distances and is therefore dimensionless. The strain component εij corresponds to the deformation in the direction xi of a surface normal to xj. The components ε11, ε22, ε33 are the normal components and represent either the contraction (εii < 0) or the elongation (εii > 0) of the medium. The other components are the tangential components and represent the distortions, or shear. If displacements are small, the term of second order can be neglected. Then each strain component εij can be written as:

⎟⎟⎠

⎞⎜⎜⎝

⎛

∂

∂+

∂∂

≈i

j

j

iij x

uxu

21ε Equation 1.13

1.1.3 Hooke’s law

No simple mathematical model can describe the mechanical behavior of the wide variety of existing materials. Many simplifications are required if the model is to be useful. Under the assumption that a small quasi-static compresion is applied, biological tissue can be modelled as a linear elastic solid. This model assumes that the material is:

• A single-phase solid: As an example, fluids and poro-elastic materials (composed of a solid matrix filled with a fluid) do not comply with this assumption, and exhibit fluid flow.

• Homogeneous: Although most biological materials have a hierarchical structure on a microscopic scale, properties are supposed to be position invariant at a macroscopic scale.

• Elastic: The stress and strain state of the material depends only on current conditions/ The material returns to its initial shape and dimensions as soon as external stresses are

Chapter 1. Theory

24

removed, and the relation between stress and strain does not depend on time. Materials that do not comply with these assumptions are usually plastic materials, which are permanently deformed once an external stress has been applied, or visco-elastic materials, whose behavior depends on the previous stress/strain state, i.e. on the “history” of the material.

• Linear: Each stress component is a linear function of the strain components. Superposition holds.

Under these assumptions, there exist a set of 81 (34) coefficients Cijkl that relate stress and strain:

∀{i,j}∈{1,2,3}2, ∑∑= =

=3

1

3

1k lklijklij C εσ Generalized Hooke’s law - Equation 1.14

The quantities Cijkl are the components of the elasticity tensor and characterize the mechanical properties of the material. They are equivalent to pressures and are given in Pascals. As σij is symmetric, the elasticity tensor is also symmetric in i and j, i.e.:

∀{i,j,k,l}∈{1,2,3}4, Cijkl = Cjikl Similarly, the elasticity tensor is symmetric in k and l because the strain tensor is symmetric: ∀{i,j,k,l}∈{1,2,3}4, Cijkl = Cijlk

As a consequence, only 6x6 = 36 independent elasticity constants are required to define the linear elastic properties of the material:

⎥⎥⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢⎢⎢

⎣

⎡

=

121212131223123312221211

131213131323133313221311

231223132323233323222311

331233133323333333223311

221222132223223322222211

111211131123113311221111

CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC

ijklC Equation 1.15

To further simplify the model, the following assumption is often added:

• Isotropic: The material properties are independent of orientation Under this assumption, only two independent constants (λ, µ) are needed to describe the mechanical properties of the material. Using Kronecker’s symbol δij (i.e. δij=1 if i=j, or δij=0 if i≠j), the components Cijkl of the elasticity tensor can be written as:

Cijkl = λ δijδkl + µ (δikδjl + δilδjk) The relation between stress and strain becomes:

Chapter 1. Theory

25

⎥⎥⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢⎢⎢

⎣

⎡

⎥⎥⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢⎢⎢

⎣

⎡

++

+

=

⎥⎥⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢⎢⎢

⎣

⎡

12

13

23

33

22

11

12

13

23

33

22

11

222

000000000000000000200020002

εεε

εεε

µµ

µµλλλ

λµλλλλµλ

σσσσσσ

The coefficients λ and µ are known as the Lamé constants. λ is the bulk modulus (fr: coefficient d’élasticité de compression) and µ is the shear modulus (fr: coefficient d’élasticité de cisaillement). Two related constants are commonly used in engineering: the Young’s modulus E and the Poisson’s ratio υ. The Young’s modulus represents the ratio between the applied stress and the strain along the axis of the applied stress under tensile load of a bar. The Poisson’s ratio is the ratio between the lateral strain (perpendicular to the axis of the applied stress) and the axial strain (along the axis of the applied stress). If the stress is applied along x1, we can write:

11

11

εσ

=E Equation 1.17

11

33

11

22

εε

εευ −=−= Equation 1.18

The Young’s modulus is expressed in Pascals. The Poisson’s ratio is a dimensionless ratio. Stiff materials exhibit a large Young’s modulus, whereas the Young’s modulus of soft materials is small. Although there are some exceptions for composite materials, the Poisson’s ratio is always between 0 and 0.5. Materials with a Poisson’s ratio υ=0.5 are incompressible materials; any axial strain on such a material induces corresponding lateral deformations to keep the volume constant. A material with a Poisson’s ratio υ=0 is totally compressible; when under axial compression or expansion, it does not deform laterally but its volume changes. The Young’s modulus and the Poisson’s ratio are related to the Lamé constants:

( )µλ

µλµ++

=23E Equation 1.19

( )µλλυ+

=2 Equation 1.20

Or reciprocally:

( )( )υυυλ

211 −+=

E Equation 1.21

( )υµ

+=

12E

Equation 1.22

Of course, the behavior of biological tissues is much more complex than the behavior of linear isotropic elastic solids. It is necessary to have a clear understanding of the limitations of this model. In the following paragraphs, we will discuss the various properties of soft biological tissues that may not comply with the assumptions mentioned above.

Equation 1.16: Hooke’s law for a linear elastic solid

Chapter 1. Theory

26

1.1.4 Non-linearity

For linear materials the stress-strain curve is a line with constant slope, the slope is the Young’s modulus. Non-linear materials exhibit a non-linear stress-strain relationship (Fig. 1.2). For example, many materials become stiffer as they are being compressed. A non-linear material cannot be characterized by a unique modulus, but its modulus E(ε) is dependent on the applied strain ε (also referred to as the applied pre-compression) and is locally equal to the slope of the stress-strain curve:

εσε

ddE ≈)( Equation 1.23

Krouskop et al. (1998) have shown that prostate cancer, normal glandular breast tissues and normal fibrous breast tissues behave non-linearly. Breast carcinoma is highly non-linear.

Fig. 1.2: Typical non-linear stress/strain relationship, shown for muscle, artery, skin and prostate (Erkamp et al. 2000) In elastography, the non-linear behavior of soft tissues may induce enhancements or deterioration of the strain contrast between malignant and normal tissues depending on the chosen amount of pre-compression (Erkamp et al. 2000). Therefore care has to be taken in the choice of an appropriate pre-compression to enhance the strain contrast of the malignant tissues to be detected. 1.1.5 Visco-elasticity

Viscoelasticity indicates time dependent mechanical behavior of a particular class of materials. Thus, the relationship between stress and strain is not constant but depends on the loading history. The major characteristic of a viscoelastic material is hysteresis or energy dissipation. This means that if a viscoelastic material is loaded and unloaded, the unloading curve will not follow the loading curve. The difference between the two curves represents the amount of energy that is dissipated or lost during loading. An example of hysteresis is shown below in Fig. 1.3:

Chapter 1. Theory

27

Fig. 1.3: Hysteresis. The path to the first loading point (1) is different during loading and unloading. In many materials, after a few loading and unloading cycles (2, 3, 4, ..) the amount of hysteresis under cyclic loading is reduced and eventually the stress-strain curve becomes reproducible. A linear isotropic viscoelastic material cannot be characterized using a unique real Young’s modulus. Its modulus is complex and is a function of frequency f:

E = E ’ + i.E ” Equation 1.24 The real part E ’ is the storage modulus (for shear) and determines the elastically stored energy. The imaginary part E” is the loss modulus and determines the energy dissipated per cycle and volume. There are two major types of behavior characteristic of viscoelasticity. Creep is increasing deformation under constant load. This contrasts with an elastic material in which the deformation remains constant regardless of how long the load is applied. Creep is illustrated schematically below:

Fig 1.4: Illustration of the creep phenomenon. Once a step load was applied and maintained, deformation continues to increase until it reaches an equilibrium (from University of Michigan, Ann Arbor, MI, USA, College of Engineering, Class of Biomechanics) The second significant behavior is stress relaxation. This means that the stress will be reduced or will relax under a constant deformation. This behavior is illustrated in Fig. 1.4.

Chapter 1. Theory

28

Fig. 1.5: Illustration of the stress relaxation phenomenon. When a step deformation is suddenly applied, stress first undergoes an overshoot, then decays to its new equilibrium. The relaxation time τ measured during stress relaxation characterizes the viscoelastic behavior of the material. Available data show that breast tissues are viscoelastic, and that the relaxation time could be used to characterize different tissue types in the breast (table 1.1). Tissue type Relaxation time at 25°C

(ms) Relaxation time at 38°C

(ms) Normal fat 420 ± 100 100 ± 40 Normal glandular tissue 540 ± 120 460 ± 80 Fibroadenoma 760 ± 60 740 ± 80 Invasive ductal carcinoma 520 ± 80 400 ± 80 Table 1.1: Relaxation times of various breast tissues. Data courtesy of Dr. T. Krouskop. 1.1.6 Anisotropy

In order to decrease the number of constants required to characterize a biological material, we assumed the material is isotropic, i.e. it has the same properties in all directions. However some tissues, like muscle fibers, are oriented and behave differently depending on the direction of the applied stress. As a consequence, the Lamé constants are not sufficient to characterize such materials. As a conclusion, elastic constants are intrinsic properties that characterize a material, whereas strain represents the behavior of the material under a given stress. The interpretation of a strain image (elastogram) requires the investigator to be aware of the limitations and of possible artifacts: for example, low strain can be observed where the material is stiff, but it could also be observed in a soft material if low stress was applied.

1.2 Basic Ultrasound Physics & Acoustic Imaging

An understanding of the physics of ultrasound is essential to the successful clinical application of diagnostic ultrasonographic techniques. This section provides a brief review of the principles of medical ultrasound. Most of this section was directly taken from the book

Chapter 1. Theory

29

“Ultrasound Physics and Instrumentation” written by Hedrick et al. (1995). More detailed information on ultrasound physics and applications such as wave propagation, transducer design, medical imaging, and biological effects can be found in the above-mentioned book and others (Wells 1977, Christensen 1988, Allard 1993, Kremkau 2002), but are beyond the scope of this short review. 1.2.1 Sound waves and the wave equation

Sound is mechanical energy that is transmitted through a medium. Changes in the pressure of the medium are created by forces acting on the molecules, causing them to oscillate about their average positions. Ultrasound is defined as high-frequency (> 20 kHz) pressure or mechanical waves that humans cannot hear. The changes in pressure when vibrating molecules interact with neighboring molecules are conveyed from one location to another. The term propagation describes this transmittal to distant regions remote from the sound source. The movement of particles can be illustrated by looking at the movement of an audio speaker. As the speaker moves forward, the air molecules immediately in front are pushed together, producing a region of increased air density, characterized by a small zone of increased pressure (compression). When the speaker is pulled back, a zone of decreased molecular density results (rarefaction). The originally affected molecules collide with adjacent molecules to propagate the action of the speaker. Molecules do not travel from one end of the medium to the other. Rather, the effect is transmitted over long distances because of neighbor-to-neighbor interactions. The fundamental principles governing wave propagation are the conservation of mass (or equation of continuity) and the conservation of momentum (or Euler’s equation). The wave equation is derived from these two equations13:

( ) ( ) Xuuu+∇+∇∇+=

∂∂

•2

2

2

µµλρt

Equation 1.25 - The wave equation

where ρ is the density of the medium, t the time, u the particle displacement vector, λ and µ are the Lamé constants, and X the body forces. For elastic solids, equation 1.25 can be divided into two equations:

13 The following notation is used:

⎟⎟⎠

⎞⎜⎜⎝

⎛∂∂

∂∂

∂∂

==∇321

,,xxx

grad ϕϕϕϕϕ is the gradient of the scalar field φ(x1,x2,x3)

3

3

2

2

1

1

xv

xv

xvdiv

∂∂

+∂∂

+∂∂

==∇ • vv is the divergence of the vector field v = (v1,v2,v3)

⎟⎟⎠

⎞⎜⎜⎝

⎛

∂∂

−∂∂

∂

∂−

∂∂

∂∂

−∂

∂==∧∇ 2

2

12

21

22

21

32

23

12

23

22

22

32

,,xv

xv

xv

xv

xv

xv

curl vv is the curl vector

⎟⎟⎠

⎞⎜⎜⎝

⎛

∂

∂+

∂

∂+

∂

∂

∂∂

+∂∂

+∂∂

∂∂

+∂∂

+∂∂

=∇=∇∇ • 23

32

22

32

21

32

23

22

22

22

21

22

23

12

22

12

21

12

2 ,,xv

xv

xv

xv

xv

xv

xv

xv

xv

vv is the

Laplacian of the vector field v.

Chapter 1. Theory

30

2

22

2 t∂∂

+=∇

ϕµλ

ρϕ Equation 1.26 : Longitudinal wave

2

22

t∂∂

=∇ψψ

µρ Equation 1.27: Transverse wave

where a scalar potential φ and a vector potential ψ = (ψ1, ψ2, ψ3) are used to represent the displacements ( ψu ∧∇+∇= ϕ ).

The solutions to equation 1.26 and 1.27 are two waves that propagate through the medium. Waves are divided into two basic types: longitudinal and transverse. Longitudinal waves are those in which particle motion is along the direction of the wave energy propagation, i.e. particles vibrate back and forth in the same direction as the wave is travelling. Equation 1.26 defines a longitudinal wave. Transverse waves, sometimes referred to as shear waves or stress waves, are those in which the motion of the particles is perpendicular to the direction of propagation of the wave energy. The wave motion resulting from a stone thrown into a pool of water is an example of a transverse wave. The solution to equation 1.27 is a transverse wave. In soft tissues, transverse waves are usually negligible compared to the magnitude of longitudinal waves. The speed at which a wave propagates through the medium is called acoustic velocity (c). It is not the same as the particle velocity (v), which refers to the speed at which the particles vibrate back and forth across their mean positions. The acoustic velocity depends on the mechanical properties of the propagation medium. The acoustic velocity cL of a longitudinal wave in a fluid is determined by the density (ρ) of the medium and by its bulk modulus (Κ) through equation 1-2. The bulk modulus gives the change in volume of a substance as the pressure on it is changed.

ρρ

µλ Κ=

+=

2Lc Equation 1.28: Longitudinal velocity

The acoustic velocity cT of the transverse wave is determined from Eq.1-3 by the density and the shear modulus (µ) of the medium:

ρµ

=Tc Equation 1.29: Transverse velocity

In most soft tissues, the acoustic velocity is within a few percentage points of 1540 ms-1 for longitudinal waves, and is on the order of 1-10 ms-1 for transverse waves. Acoustic velocity exhibits a slight dependence on sound frequency (this phenomenon is called dispersion) and on temperature, but these small differences have little importance in clinical imaging. The wavelength (λ) is the extent of a complete wave cycle, i.e. the distance between two consecutive density zones. The frequency (f) is the number of cycles that occur in one second, and is the reciprocal of the period (τ) of the wave. Diagnostic ultrasound operates over a range of 1-20 MHz. Wavelength, frequency and acoustic velocity are related by equation 1-1:

Chapter 1. Theory

31

τλ cfc

== Equation 1.30

1.2.2 Intensity

The intensity of an ultrasonic beam is the amount of energy flowing through a cross-sectional area per second. Simply, it is the rate at which the energy is transmitted by the wave over a small area. The study of potential biological effects is linked to intensity. Traditionally, acoustic intensity is expressed in units of watts per centimeter squared. At any point P through which an ultrasound beam passes, the instantaneous intensity i is given by:

c

tPptPi i

ρ),(

),(2

= Equation 1.31

where pi(P,t) is the instantaneous acoustic pressure at point P, t the time, c the velocity of sound, and ρ the density. Often the time-averaged intensity (ITA) is of interest. For a continuous wave, the time-averaged over one cycle is:

cpITA ρ2

2

= Equation 1.32

where p is the peak pressure amplitude. The spatial-averaged intensity (ISA) over a given area is also commonly used. The half-value layer (HVL) of material is the thickness that will reduce the intensity to half its original value. The power W is a measure of the total energy transmitted per unit time summed over the entire cross-sectional area of the beam:

AreaIntensity xW = Equation 1.33 Ultrasonic power, averaged over a time period, is referred to as the temporal average power. 1.2.3 Reflection

If a sound beam is directed to a smooth interface (e.g. the boundary between different tissue types) larger than the width of the beam, it will be partly reflected. These interfaces are called specular reflectors (from the Latin word speculum, mirror). The resulting angle of reflection (Φr) equals the angle of incidence (Φi). These angles are defined relative to a line drawn perpendicular to the surface of the interface. In diagnostic ultrasound, the same device transmits and receives the sound waves, so maximum detection of the reflected echo occurs when the sound beam strikes the interface with normal incidence. A large rough-surface interface deflects the ultrasound beams in multiple directions; this is called diffuse reflection. What conditions result in a reflection of energy ? A useful analogy would be throwing a baseball against a brick wall; not much energy will be transferred to the wall. Conservation of energy would permit the ball to transfer all its energy to the wall and simply stop at the surface of the wall, but conservation of momentum prevents this from occurring because of the differences in mass. In ultrasound the quantity analogous to momentum is acoustic impedance (Z). This quantity is a measure of the resistance to sound passing through the medium. Acoustic impedance is measured in kg/m2/s, which is given a special name, the rayl (from the name of John William Strutt, Lord Rayleigh, 1842-1919). cZ ρ= Eq.1-34

Chapter 1. Theory

32

If the acoustic impedance is the same in one medium as in another, sound will be readily transmitted from one to the other. A difference in acoustic impedances causes some portion of the sound to be reflected at the interface. The reflection coefficient (R) for intensity is the ratio between the reflected acoustic intensity Ir to the incident intensity Ii. For normal incidence it is expressed as follows:

2

12

12⎟⎟⎠

⎞⎜⎜⎝

⎛+−

==ZZZZ

IIR

i

r Equation 1.35

Z1 and Z2 are respectively the acoustic impedances of media no. 1 and no. 2. The transmission coefficient (T) is the ratio of the acoustic intensity It transmitted into medium 2 to the incident intensity. It is calculated directly by the formula:

( )

RZZZZ

II

Ti

t −=+

== 142

12

21 Equation 1.36

It does not matter which impedance is the larger or smaller for two materials composing the interface. The same percentage of reflection occurs at the interface, whether sound is going from a high acoustic impedance to a low acoustic impedance, or vice-versa, and regardless of the thickness of either medium. When a transducer scans a patient, multiple interfaces are encountered along the path of travel of the ultrasound beams. A percentage of the incident beam intensity is reflected and transmitted at each interface. A series of echoes is subsequently detected. 1.2.4 Scattering

Another important interaction between ultrasound and tissue is scattering, or nonspecular reflection, which is responsible for the internal texture of organs in the image. The scattering occurs because the interfaces are small, less than several wavelengths across, and is called Rayleigh scattering. Each interface acts as a new separate sound source, and sound is reflected in all directions. These nonspecular reflections have a strong frequency dependence (f 2 to f 6). Many factors influence the fraction of incident intensity that is reflected at an interface toward the transducer – the acoustic impedance mismatch, the angle of incidence, the size of the structure compared with the wavelength, the shape of the structure, and the texture of the surface of the interface. The combination of these factors is described by the term reflectivity. 1.2.5 Refraction

Refraction is the change in direction of the portion of wave that is transmitted at an interface. If the beam strikes the interface at an angle other than 90 degrees, the transmitted part will be refracted or bent away from the straight-line path. Refraction of sound obeys Snell-Descarte’s law, which relates the angle of transmission (Φt) to the relative velocities of sound (ci, ct) in the two media:

t

i

t

i

cc

=ΦΦ

sinsin

Equation 1.37: Snell-Descarte’s law

Chapter 1. Theory

33

This bending occurs because the portion of the wavefront in the second medium travels at a different velocity from that in the first medium. This does not generally present any difficulty in diagnostic ultrasound because the velocity of ultrasound in soft tissue is relatively constant. However under certain conditions refraction of the wave can cause misregistration in diagnostic ultrasound because the formation of the image is predicted on the assumption that the ultrasound beam always travels in a straight line through the tissues. An object can appear at the wrong location because the assignment of position is based on the projected straight line path of the beam whereas the true position of the object is actually offset from the assumed path. A similar effect caused by the refraction of light is seen when an object under water is viewed from above. 1.2.6 Diffraction

Diffraction causes the ultrasound beam to diverge or spread out as the waves move father from the sound source. The rate of divergence increases as the size (diameter) of the sound source decreases. The lateral resolution of the beam and the sensitivity of the ultrasonic system are both affected by divergence. 1.2.7 Interference and Speckle

Sound waves demonstrate interference phenomena or the superposition of waves (algebraic summation). If waves with the same frequency are in phase, they undergo constructive interference and result in an increased amplitude. If waves with the same frequency are out of phase, they undergo destructive interference and result in a decrease in amplitude. Every combination from completely constructive to completely destructive interference can occur, resulting in a complex wave summation. This interference is important in the design of an ultrasonic transducer because it affects the uniformity of the beam intensity through the ultrasonic field. Focusing of the beam in real time imaging and in therapeutic applications is based on the principle of wave interference. Scattering by each small interface within soft tissues causes multiple reflected wavefronts to interfere. The resulting interference pattern is not constant but changes with time, even in a homogeneous medium. The fluctuating signal causes variations in the amplitude of the backscattered signal received by the transducer. This phenomenon is called ultrasonic speckle and is responsible for brightness nonuniformities in sonograms of a homogeneous object. 1.2.8 Absorption and Thermal Index (TI)

Absorption is the process whereby ultrasonic energy is transformed into other energy forms, primarily heat. It is responsible for the medical applications of therapeutic ultrasound. Absorption is the only process whereby sound energy is dissipated in a medium. All other modes of interactions (reflection, refraction, scattering, and divergence) decrease the ultrasonic beam intensity by redirecting the energy of the beam. The absorption of an ultrasonic beam is related to the beam’s frequency and to viscosity and relaxation time of the medium. The relaxation time describes the rate at which molecules return to their original positions after being displaced by a force. The ability of molecules to move past one another determines the viscosity of a medium. If the frequency is increased the molecules must move more often, thereby generating more heat from the drag caused by

Chapter 1. Theory

34

friction (viscosity). Also, as the frequency increases, less time is available for the molecules to recover during the relaxation process. Molecules remain in motion, and more energy is required to stop and redirect them, again producing more absorption. The rate of absorption is directly proportional to the frequency. The peak amplitude of acoustic pressure follows an exponential function: ( )zpp α−= exp0 Equation 1.38 where p is the peak amplitude of pressure of the beam at a distance z, p0 the original peak pressure of the beam, α the absorption coefficient, and z the distance traversed by the beam. The thermal index is an indicator of the potential biological effects of temperature elevation. The thermal index is defined by the AIUM (American Institute of Ultrasound in Medicine) standards as the ratio of the in situ acoustic power (W’) to the acoustic power required to raise tissue temperature by 1 °C (Wdeg):

deg

'WWTI = Equation 1.39

Three thermal indices corresponding to soft tissue (TIS), bone (TIB), and cranial bone (TIC) have been developed for application to different examinations. 1.2.9 Attenuation

Attenuation includes the effects of both scattering and absorption in the characterization of amplitude reduction as the ultrasound wave propagates through a medium. It is also described by an exponential function, but the absorption coefficient is replaced by the attenuation coefficient a: )exp(0 zapp −= Equation 1.40 The attenuation coefficient is given by the sum of the scattering coefficient as and the absorption coefficient α:

α+= saa Equation 1.41 The unit for these coefficients is the neper (Np) per centimeter. Decibels (dB) per centimeter are commonly used. The conversion to decibels for attenuation loss expressed in nepers is given as:

)()()( 686.810ln

20NpNpdB aaa ≈= Equation 1.42

The effect of frequency must be included in any specification of attenuation coefficient. In a first approximation the attenuation coefficient increases linearly with frequency in biological tissues. In this case it is often expressed in units of nepers per megahertz (Np/MHz). High frequency sound waves are attenuated faster than low-frequency sound waves. Thus the ability to penetrate tissue is reduced at higher frequencies. In addition, a reflector positioned at progressively greater depths generates progressively lower-intensity returning echoes. In ultrasound scanners, a method called time time-gain compensation (TGC) is used to increase the strength of the received signal by amplification of a function of time, i.e. of depth.

Chapter 1. Theory

35

1.2.10 Cavitation and Mechanical Index (MI)

As the ultrasound wave propagates through the medium, regions of compression and rarefaction are created. Thus localized regions are subject to increases and decreases in pressure in an alternating fashion and these cause gas bubbles to form and grow and to exhibit dynamic behavior. This phenomenon is known as cavitation. In stable (noninertial) cavitation microbubbles already present in the medium expand and contract during each cycle in response to the applied pressure oscillations. The bubbles may also grow as dissolved gas leaves the solution during the negative-pressure phase, a process called rectified diffusion. Each bubble oscillates about the expanding radius for many cycles without collapsing completely. At a characteristic frequency which is a function of the size of the bubble, the vibration amplitude of neighboring liquid particles in maximized. This condition is called volume resonance. Oscillations of a gas bubble can may produce high shearing forces in the nearby surrounding areas, and may give rise to microstreaming. Biomolecules or membranes subjected to such phenomena can fragment or rupture. Transient (inertial) cavitation is a more violent form of microbubbles dynamics in which short-lived bubbles undergo large cycle changes over a few acoustic cycles before completely collapsing. This phenomenon produces highly localized (within 1 µm3) shock waves, very high temperatures (up to 10,000°K) and pressures (108 pascals or higher). Theses effects may result in the decomposition of water to free radicals, and may also drive chemical reactions. The general consensus is that transient cavitation is a threshold effect that depends on the peak negative pressure p- of the acoustic wave and on frequency f, exhibiting a p2/f dependence. The cavitation threshold is predicted by the mechanical index (MI):

fpMI

−

= Equation 1.43

where p- is expressed in MPa, f in MHz. 1.2.11 Echo ranging

In diagnostic ultrasound, reflections of the sound beam from interfaces along the ultrasonic path are of primary interest. An ultrasound wave is transmitted into the body, strikes an interface (acoustic mismatch between two media), and is partially reflected to the transducer. The reflected waves arising from the various impedance mismatches result in ultrasonic detection of interfaces within the body. If the velocity of sound is known and constant along the path of travel, a system that can generate an ultrasonic pulsed wave and detect the reflected echo after a measured time permits the distance to an interface to be determined. In diagnostic ultrasound the distance z between the transducer and an interface is measured from the return-trip time-of-flight τ of the ultrasonic wave between the transducer and the target:

2τcz = Equation 1.44

where c is the velocity of sound and τ the time of travel.

Chapter 1. Theory

36

1.2.12 Basic ultrasound instrumentation

There are three requirements for diagnostic medical ultrasound: generation of a beam, reception of the returning echoes, and processing of the signal for display. All scanning devices are derived from the basic A-mode scanner. A-mode scanning is an echo-ranging technique in which an ultrasound beam is directed along a single path into the body. Detected echoes from interfaces along this path are displayed as a series of spikes. The position of the spikes along the display axis denotes the depth of the interface; the height of the spikes denotes the strength of the echo. Two-dimensional ultrasound imaging is performed with B-mode scanning. The image, or sonogram, is created by moving the transducer to a different position or orientation. Mechanical scanners displace the transducer, whereas linear and phased arrays use electronic selection of small adjacent piezoelectric crystals to scan the body. Attenuation of the A-mode signals is compensated by TGC, and the envelope of the processed A-mode data is displayed in gray-scale format. The direction of the beam is determined from the position and orientation of the transducer. Gray-scale allocation has an important influence on the ability of human vision to detect an object in a sonogram. The ultrasonic wave is generated by a transducer made of a piezoelectric material. An electric voltage applied to such a material causes expansion and contraction of the crystal. When placed over the body, sound waves (mechanical waves) are passed into the body. In echo ranging systems the transducer must pause after transmitting the sound wave to “listen” for the returning echoes. It must be pulsed for transmission after an appropriate listening time has elapsed. The maximum pulse repetition frequency (PRFm) is limited by the maximum depth R to be sampled and by the velocity of ultrasound c in the medium, as shown in the following equation:

RcPRFm 2

= Equation 1.45

In B-mode scanning, the maximum frame rate is determined by the number of scan lines and the maximum pulse repetition frequency. Ideally, for each pulse, a short packet of ultrasound energy of the appropriate frequency is directed into the body. In practice, a pulsed transducer generates a range of different frequencies described by the bandwidth B. The frequency spectrum of the transducer can be analyzed by measuring in the Fourier domain the amplitude versus the frequency of an echo returning from a flat steel target located at a fixed distance. The absolute bandwidth is defined as the width of the portion of the frequency spectrum that is located above a –6 dB level from the maximum amplitude of the spectrum, called the full width at half maximum (FWHM) amplitude. The quality factor (Q-value) of a transducer is defined by the center frequency f0 and the bandwidth of the transducer, or alternatively by the ratio of energies stored and lost in the transducer per cycle:

cycleper lost Energy cycleper storedEnergy 0 ==

Bf

Q Equation 1.46

Chapter 1. Theory

37

A high-Q transducer stores energy in the crystal and therefore loses very little each cycle. After being stimulated by the voltage pulse, it vibrates (rings) for an extended duration, producing a long pulse. A low-Q transducer, on the other hand, is easily damped and generates a short pulse. Most of the energy lost per cycle in converted to heat inside the transducer. To avoid heating of the transducer above its Curie temperature (the temperature at which the material loses its piezoelectric behavior), high-Q transducers are preferred for therapeutic applications. The quality factor is often expressed by its reciprocal, the fractional bandwidth b:

0

1fB

Qb == Equation 1.47

Diagnostic ultrasound uses high bandwidth transducers (typically b ranges from 0.4 to 0.8) to generate short pulses. The pulse duration (PD), or temporal pulse length, is the time interval for one complete pulse. A more formal definition is the elapsed time from initiation of the pulse to a point 20 dB below the maximum peak-to-peak pressure amplitude of the wave. The pulse duration is inversely proportional to the bandwidth. The spatial pulse length (SPL) can be calculated from the pulse duration and the velocity of sound:

2.PDcSPL = Equation 1.48

In most diagnostic applications, the amplitude of the emitted ultrasonic pulse h(t) can be approximated by a Gaussian function of time t:

2

2

2exp

21)(

tt

tthσσπ

−= Equation 1.49

where σt is the standard deviation of the pulse in the time domain. The duration of such a pulse, defined from the –20 dB point below the maximum amplitude, is:

tPD σ10ln8= Equation 1.50 Gaussian pulses have a Gaussian frequency spectrum H(f):

( )2

20

2exp

21)(

ff

fffH

σσπ−

−=

where σf is the standard deviation of the pulse in the frequency domain. The –6 dB FWHM bandwidth of such a pulse is given by:

fB σ2ln8= Equation 1.51 The standard deviation of the pulse in the time domain and in the frequency domain are related by the following equation:

ft πσ

σ2

1= Equation 1.52

The axial resolution specifies how close together two objects can be along the axis of the beam and yet still be detected as two distinct entities. The axial resolution also specifies the smallest object detectable along the axis of the beam. The best possible spatial resolution is the spatial pulse length divided by 2. The factor 2 is due to the return-trip time-of-flight. Lateral resolution describes the ability to resolve two objects adjacent to each other that are perpendicular to the beam axis. It also refers to the ability of the ultrasound beam to detect single small objects across the width of the beam. The lateral resolution is determined by the

Chapter 1. Theory

38

beam width. A single object smaller than the sound beam produces a signal the entire time it is within the beam; thus the object appears to be the same size as the width of the beam. Lateral resolution can be improved by focusing the transducer crystal. Focusing limits the useful nearfield depth because the beam diverges rapidly beyond the focal zone. The focal zone is defined as the region where intensity has a value within 3 dB (50%) of the maximum along the transducer axis. The focal zone may not be symmetrical around the focal point, which is the point of maximum intensity. The beam is most narrow at the focal point.

1.3 Ultrasonic Elastography

Elastography is an ultrasonic imaging modality, capable of displaying the internal strains induced in a soft material undergoing an axial stress (Ophir et al. 1991, Ophir et al. 1999). Its principle, methods, and characteristics are explained in this chapter. It should be noted that elastography, as a strain imaging modality, is not necessarily ultrasonic but could be performed using other existing imaging modalities, such as MRI (Wu et al. 2001). 1.3.1 Principle

Ultrasonic elastography relies on radio-frequency (RF) ultrasonic signals that can be acquired using standard ultrasound scanners. When a stress is applied to the tissues, these are being compressed (or elongated, depending on whether the stress is compressive or tensile) and displaced. The local displacement of each particle along the direction of propagation of the ultrasonic beam induces a shift in the time domain of the backscattered signal, due to changes in the time of flight. Assuming a constant speed of sound, the induced time shift τ is directly proportional to the local displacement d. This process is described in figure 1.6:

Figure 1.6: Pre- and post-compression RF signals. The displacement of the tissues induced a time shift. The compression (or elongation) also induces a compression (or respectively stretching) in the time domain of the ultrasonic signal (Céspedes et al. 1993), similar to a frequency modulation, as shown below:

Figure 1.7: Compression in the time domain of the RF signal The process of elastographic acquisition is as follows:

Time

Time Pre-compression RF signal

Post-compression RF signal

Time delay

Pre-compression Post-compression

Time

Chapter 1. Theory

39

- Acquire pre-compression RF data - Apply a small compression - Acquire post-compression RF data - Estimate strain from the acquired signals. The following paragraphs explain how the axial strain can be estimated from the pre- and post-compression RF signals. 1.3.2 Methods

1.3.2.1 Gradient of the time delays

The pre- and post-compression RF lines are segmented into short windows, and the time delay τ between corresponding windows is estimated. If the displacement of the particles inside the window is d, and assuming a constant speed of sound c, the time shift τ is given by the variation in the round-trip time of flight of the ultrasonic signal:

d = cτ / 2 Equation 1.53 The time delay is usually obtained from the position of the maximum of the cross-correlation function (which is a quantitative measurement of the similarity between two delayed signals) between the corresponding pre- and the post-compression windows. The maximum value of the normalized cross-correlation function is an indicator of the quality of the time delay estimate. In order to obtain sub-sample precision, a parabolic or cosine interpolation is performed around the maximum of the function (Céspedes et al. 1995). A cyclic bias may appear, depending on the chosen interpolation method, and may induce the zebra artifact (Céspedes et al. 1995, Alam et al. 2000). However the error would be much larger if no interpolation were used. Other time-efficient estimators can be used to obtain the time delays, using the position of the minimum of the sum of absolute differences between the corresponding segments (Zhu et al. 2002), or the root of the phase of the cross-correlation of the analytic signals (Pesavento et al. 1999). The process is repeated for each window, and a profile of the axial displacement along the RF signal is obtained. Axial strain ε is given by the gradient of the axial displacements in the axial direction z (where z = ct / 2):

tzd

∂∂

=∂∂

=τε Equation 1.54

Practically, the gradient is obtained using two consecutive windows separated by a window shift ∆T. The axial strain measured between the two windows is given by the variation of the time delays divided by the window shift (figure 1.8):

T∆−

= 12 ττε Equation 1.55

Chapter 1. Theory

40

Figure 1.8: The strain is estimated using the time delays in two consecutive windows (from J. Ophir’s slides,

available online at www.elastography.com ) As discussed earlier, the ultrasonic signal also undergoes a compression (or elongation) in the time domain, i.e. its shape changes. Therefore the cross-correlation decreases with increasing strain, resulting in an increased noise in time delay estimation (Céspedes et al. 1999). Temporal stretching of the post-compression RF signal prior to time delay estimation improves the accuracy of the time delay measurements (Céspedes et al. 1993, Varghese et al. 1996). The process is repeated along multiple RF lines in order to form an image of the strain (elastogram). The major disadvantage of this method is the noise induced by the gradient operation. Various methods have been proposed to decrease the noise using filtering, either explicitly (O’Donnell et al. 1994) or implicitly by means of a least-squares estimator (Kallel et al. 1997) or of a staggered strain estimator (Srinivasan et al. 2002). 1.3.2.2 Stretching factor

Using an adequate temporal stretching of the post-compression signal improves the cross-correlation between the pre- and post-compression segments. Instead of looking for time delays, the stretching factor itself can be used as a direct measurement of strain. The pre- and post-compression signal can be modeled as (Alam et al. 1998):

r1(t) = s(t) * p(t) + n1(t) Equation 1.56 r2(t) = s(t /α - t0) * p(t) + n2(t) Equation 1.57

where s(t) is the one-dimensional scattering distribution of the elastic target, p(t) is the impulse response of the ultrasonic system, n1(t) and n2(t) are uncorrelated renditions of random noise, α is a temporal stretching factor close to unity because the applied strain is generally small, t0 is a time delay, and * denotes convolution. An iterative approach is used to find the stretching factor α that maximizes the normalized cross-correlation function between the two signals (Alam et al. 1998, Brusseau et al. 2001).

τ2 ∆T

Pre-Compression RF signal

Post-Compression RF signal

τ1

Chapter 1. Theory

41

This can be implement using an exhaustive search, a binary search or a hierarchical search. If the strain ε inside the window is small (i.e. ε <<1), it can be approximated by:

ε ≈ 1 – α Equation 1.58 Because adaptive stretching does not contain any inter-window operation (gradient), it does not suffer from this type of degradation and it results in lower noise in strain estimates than gradient-based methods. A limitation of this method however is its computational complexity. 1.3.2.3 Methods in the frequency domain

The compression of the RF signal in the time domain is analogous to a frequency modulation and corresponds to a shift of the spectrum in the frequency domain. Spectral estimators have been proposed for elastography. The strain (or the stretching factor) can be estimated from an estimation of the frequency shift, using either the centroid (Konofagou et al. 1999) or spectral cross-correlation (Varghese et al. 2000). These estimators were shown to be less precise than correlation-based estimator for low strains, but more robust to decorrelation. As such, they represent an alternative for strain estimation in noisy environment (for example for low sonographic signal-to-noise ratio, or for signal decorrelation induced by excessive strain). 1.3.3 System characterization

Performances of elastographic imaging systems can be quantified using signal to noise ratio, contrast to noise ratio, sensitivity, dynamic range, and resolution. These characteristics are described in the following paragraphs. 1.3.3.1 Standard deviation of time delay estimates

The fundamental limitation in elastography is the smallest achievable standard deviation in time delay estimates. A theoretical expression of such limit is given by the modified Ziv-Zakai lower bound (ZZLB), whose expression was derived by Weinstein and Weiss (1984). The smallest variance is achieved for high sonographic signal to noise ratio (SNRs). In this case, the lower bound of the standard deviation of the time delay estimate is equal to the Cramér-Rao lower bound ( 2

CRLBσ ) and was given by Walker and Trahey (1995):

⎥⎥⎦

⎤

⎢⎢⎣

⎡−⎟

⎠⎞

⎜⎝⎛ +

+≅ 1111

)12(23 2

222322

SNRsBfBT oCRLB ρπ

σ Equation 1.59

T is the window length, B the absolute bandwidth (fractional bandwidth b = B/f0), f0 is the central frequency, ρ is the normalized cross-correlation coefficient between the two signals, and the signal to noise ratio SNRs is the ratio of root mean square (rms) amplitudes. There exist two type of errors: jitter and false peaks. False peaks occur when a secondary correlation peak exceeds the true peak, they are relatively large in magnitude (typically the error is a multiple of the wavelength, i.e. ~kλ) and from our experience are usually related to poor correlation between signals. They appear as discontinuities and can be removed through

Chapter 1. Theory

42

non-linear processing, for example using a median filter on the estimated time delay. Jitter errors occur in the CRLB region when signal decorrelation, noise, and finite window length cause a slight displacement of the true correlation peak. They are small, but they cannot be removed and they pose a fundamental limit to the performance of a time delay estimator. 1.3.3.2 Signal to noise ratio (SNRe)

The elastographic signal to noise ratio (SNRe) is a quantitative indication of the noise level in the strain images. It is defined in an area where strain is supposed to be uniform as the mean to standard deviation ratio (MSDR, Céspedes et al. 1993):

ε

ε

σµ

=SNRe Equation 1.60

1.3.3.3 The Strain Filter

Meunier and Bertrand (1995) derived the cross-correlation coefficient ρ as a function of the axial expansion α . This result was used to derive the strain filter, i.e. the upper bound of the SNRe as a function of strain (Varghese et al. 1997). The strain filter for a single compression experiment has a typical “band-pass” shape with maximum SNRe around 1% strain (figure 1.9). For low strains, the difference in time delay between consecutive windows is small and hidden by the noise in time delay estimates, resulting in high correlation but noisy strain estimates due to jitter in time delay estimates (Walker et al. 1995). For high strains, there exist an abrupt strain threshold above which the shape of the post-compression signal changes and decorrelation noise occurs, resulting in false peaks and the elastogram exhibits a typical “salt and pepper” decorrelation noise. Using a multi-compression sequence, the strain filter can be extended toward high strains and has a “high-pass” characteristic (Varghese et al. 1996).

Figure 1.9: Typical shape of a strain filter The SNRe in the CRLB region is given by:

⎥⎥⎦

⎤

⎢⎢⎣

⎡−⎟

⎠⎞

⎜⎝⎛ +

+∆=

11113

)12(2

22

23

SNRs

BfBTTSNRe oub

ρ

πε Equation 1.61

Two important characteristics are obtained from the strain filter: the sensitivity εmin is the smallest measurable strain at a given SNRe level (usually at half the maximum SNRe), and

Strain

SNRe

Dynamic range

Sensitivity

Chapter 1. Theory

43

the dynamic range DR is the range of strains that can be reliably estimated using the elastogram. The dynamic range is defined using the minimum and maximum measurable strains at a given SNRe level as (Varghese et al. 1997):

⎟⎟⎠

⎞⎜⎜⎝

⎛=

min

maxlog20εε

DR Equation 1.62

1.3.3.4 Contrast to noise ratio (CNRe)

The contrast-to-noise ratio is a quantitative measurement of the detectability of a target of a given size inside a homogeneous background and is defined as (Bilgen et al. 1999):

( )22

2

2bt

btCNReσσ

εε+

−= Equation 1.63

where εt, εb are the average strain in the target and in the background, and σt

2, σb2 are the

variances of the strain estimates in the target and in the background. 1.3.3.5 Axial and lateral resolution

Elastographic axial resolution Ra was shown to be proportional to the ultrasonic wavelength λ,and inversely proportional to the bandwidth of the US system used to acquire the data (Righetti et al. 2002, Srinivasan Righetti et al. 2003). Axial resolution is approximately equal to k.(Z+∆Z) where Z=cT/2 is the spatial window length, ∆Z=c∆T/2 is the spatial window shift, k is a constant that depends on how resolution is defined (typically 0.5 ≤ k ≤ 1) the criteria that is used to define the resolution. However the best achievable resolution is limited by the spatial duration of the ultrasonic pulse, which is given by the wavelength λ:

Ra ≈ max( k.(Z+∆Z), ka.λ ) with 0.5≤ k ≤1 and ka ≈1.5 Equation 1.64

Lateral resolution Rl is proportional to the beam width d as long as the pitch between consecutive RF lines is smaller than the beam width, with a proportionality factor kl that depends on the criteria used to define resolution, but is on the order of 0.7 (Righetti et al. 2003):

Rl = kl.d Equation 1.65 1.3.3.6 Decorrelation due to 3D motion

Mechanics is essentially a 3D problem, and in practice displacement are not purely axial, but also have components in the lateral and elevational directions, referring respectively to in-plane and out-of-plane motion in the image. Scatterers move out of the ultrasonic beam, other scatterers move in, resulting in a change in the shape of the RF signal. The correlation between pre- and post-compression signal decreases because we no longer image the same tissues (Kallel and Varghese 1997). As a consequence elastograms become noisy, and eventually decorrelation noise dominates resulting in a majority of false peaks seen as “salt

Chapter 1. Theory

44

and pepper” noise in the elastograms. Because the beam width of the US system used to acquire the data is usually small, elastography is very sensitive to small lateral and/or elevational motion. 2D or 3D companding methods have been proposed to minimize decorrelation induced by such motion, or even to use it to measure lateral displacements, lateral strains, and Poisson’s ratio (Insana et al. 1997, Chaturdevi et al. 1998, Konofagou et al. 1998, Bai et al. 2002). These methods do not restrict the search area to corresponding segments of the same RF line when looking for time delay estimates, but also search adjacent ultrasonic signals. Lateral search can be performed using a standard ultrasound scanner, but 3D scanning is required to track out-of-plane motion. These methods were shown to improve the axial elastogram in the presence of lateral or elevational motion. However the precision of the lateral or elevational components of the displacement is poor, and results in noisy lateral strain elastograms. 1.3.3.7 Zebra, Worm and Underline artifacts

Elastography, like other imaging modalities, is prone to showing artifacts. The zebra artifact (Ophir et al. 1999) appears as cyclic horizontal bands of alternate black and white in the elastograms. They often go through the whole width of the image. The spatial frequency of the bands increases with increasing strain. This artifact is due to the cyclic bias error introduced by sub-sample interpolation (Céspedes et al. 1995, Alam et al. 2000). The worm artifact occurs when large window overlaps are used (Ophir et al. 1999), which produce correlated noise patterns as well as an amplification of the noise in the gradient of the displacement. This artifact also appears as horizontal structures like the zebras, but are generally much thinner (vertically) and shorter (horizontally). They do not go through the entire width of the image and are strain independent. Normalized cross-correlation is sensitive to changes in signal amplitude (Céspedes 1993). In the presence of a spike, the estimated time delay is essentially the delay of the spike. The underline artifact (Alam et al. 2000) occurs in the presence of a sudden variation in the amplitude of the RF signal. It appears as a strain overshoot (high strain) immediately followed by a strain undershoot when the measurement window enters a high amplitude region, and an undershoot followed by an overshoot when the window enters a low amplitude region. The undershoot is expected to underestimate the true strain, but not to appear as an area of expansion (i.e. the undershoot does not cross the zero-strain limit). For a local spike whose duration is of the order of one window length, an overshoot-undershoot-overshoot pattern can be observed. The underline artifact is approximately one window long, and has sharp transitions. The undershoot may be large enough to exhibit positive strains (i.e. expansion). Non linear signal transformations such as logarithmic compression or soft limiting have been proposed to minimize this artifact. In prostate elastography, this artifact is likely to occur at the interface between a water-filled balloon and the rectal wall, or around hyper-echoic calcifications. Finally, elastography shows high strain concentrations around stiff inclusions surrounded by a soft medium (Ophir et al. 1999). This is a well known mechanical phenomenon that has been described previously (Goodier 1933, Mushkelishvili 1966). The strain concentrations are not an artifact, as they represent an existing mechanical phenomenon.

Chapter 2. Development & Characterization

45

2 System Development & Characterization Ce chapitre décrit le système d’élastographie développé au laboratoire pendant cette étude, puis montre comment ce système d’imagerie a été validé puis évalué sur des fantômes. Dans la première partie, le dispositif d’acquisition des données RF à partir d’un échographe équipé d’une sortie RF analogique est décrit. Sa particularité est d’être basé sur une sonde échographique endorectale sectorielle entourée d’un ballon servant à exercer une compression radiale sur la prostate tout en assurant le couplage acoustique. Dans la seconde partie, le traitement effectué sur les données RF en vue d’obtenir l’image des déformations (élastogramme) est détaillé. Le principe de l’évaluation des délais temporels par l’inter-corrélation à délai zéro, qui correspond simplement à remplacer le calcul classique de corrélation ρ(τ) au délai τ par un décalage temporel de la fenêtre d’un délai τ avant de calculer la corrélation ρ(0) à retard nul, est expliqué. On montre alors l’avantage de cette méthode sur la technique classique pour éviter les erreurs de faux pics dans le cas de délais (i.e. de déplacements) non négligeables devant la taille de la fenêtre. Sont ensuite expliquées les améliorations qui ont été utilisées soit systématiquement, telles que le gradient espacé (ou entrelacé), soit optionnellement comme la compensation des déplacements latéraux, l’étirement adaptatif et le cumul des déformations par multi-compression. La troisième partie montre la validation expérimentale du système sur des fantômes homogènes de géométrie cylindrique, ainsi que sur fantôme contenant une inclusion cylindrique, et compare les valeurs expérimentales avec les prédictions théoriques ainsi que des simulations par éléments finis. La quatrième partie est consacrée à une estimation expérimentale de la qualité d’image fournie par le système. Les fondements théoriques basés sur l’estimation du rapport signal-sur-bruit élastographique (SNRe) et de la résolution axiale sont rappelés. Il y est montré que pour la géométrie cylindrique le SNRe dépend de la profondeur à cause de l’atténuation ultrasonore et de la non-uniformité de la déformation. Sont alors étudiées la déformation moyenne et le SNRe obtenus pour plusieurs valeurs de compression, ceci afin de choisir la valeur permettant de maximiser la qualité d’image. Enfin il est montré que la meilleure qualité d’image, évaluée en terme de SNRe et de valeur maximale de déformation engendrée, a été obtenue par l’acquisition successive et le cumul de petits pas de déformation.

Chapter 2. Development & Characterization

46

This chapter first describes the imaging system that was designed and implemented for this work. Then it continues with an experimental validation of the system and an evaluation of the attainable elastographic signal to noise ratio on phantoms.

2.1 Data acquisition

The elastography system combines radial compression using a balloon with radial sector scan imaging. The imaging system is based on a commercially available ultrasound scanner (Combison 311, Kretz, Austria), equipped with a transrectal probe (IRW 77AK, Kretz, Austria). The scanner was slightly modified to output the analog ultrasonic radio-frequency (RF) echo signal, the frame trigger and the line trigger. The imaging probe is a rotating single element transducer with a fixed focus. The RF lines are radial, uniformly distributed around the axis of symmetry of the probe. Unless otherwise specified, acquisitions were performed with the scanner operating at 5.5 MHz. The scanner provides 8 frames per second (fps) and a 0.4° pitch. The transducer has a 52 mm focal length, a 35 mm focal depth at –6 dB (manufacturer’s data, not verified) and the received RF signal has a 40% full width at half maximum (FWHM) amplitude bandwidth (Ribault et al. 1999). For the purpose of HIFU therapy, the system also included a therapy probe (fig 2.1).

Figure 2.1 : Imaging and compression device for HIFU lesion imaging RF signals from the ultrasound scanner were initially amplified using a wideband amplifier followed by an anti-aliasing (low-pass) filter with a 10.7 MHz cutoff frequency (BLP10.7, Mini-Circuits). RF data were initially acquired using a 12-bit 60-MHz CompuScope CS6012/PCI analogue-to-digital converter (ADC) (Gage, Canada), and later using a CompuScope 14100 ADC (14-bits, 100-MHz, 128-Msamples memory depth) to increase the frame rate. The digitizer was synchronized with the ultrasound scanner using the line and the frame triggers. The transrectal imaging probe is covered by a latex transducer cover filled with a coupling liquid. It is mounted on a computer-controlled motorized holder that allows stable positioning of the probes. The balloon is inflated using a 60 ml syringe to apply a compression, either manually or using a computer-triggered pneumatic device. Using this setup, the main component of the compression was aligned with the direction of the ultrasonic beam, as long as the probe and the balloon were centered. In case the imaging probe and the balloon are not properly aligned, a strain projection artifact appears. This artifact can be compensated for when the eccentricity of the imaging probe is less than 63% (de Korte et al. 1999). Compressing using the balloon applies a uniform pressure and uniform stress on the rectal

Liquid-filled Balloon

HIFU probe

Imaging probe

Prostate

Imaging plane

Chapter 2. Development & Characterization

47

wall, and thus minimizes artifacts due to non-uniform stress fields (Konofagou et al. 1996). The amplitude of the stress applied using the balloon is independent of the direction, whereas the use of the imaging probe as a compressor requires the use of a directional compensation for non-uniform applied stress and generates an angle-dependent decrease in the elastographic signal to noise ratio (SNRe) (Lorenz et al. 1999). Using the balloon, the imaging probe can be attached to a motorized table to avoid undesired manual motion. The counterpart is the loss in ease of use, as compared to the hand-held compression using the probe. Although elastography is meant to image quasi-static strain assuming elastic materials, the time interval between the pre- and post-compression frames was kept short to minimize decorrelation due to undesired patient motion. At low frame rates, occasional decorrelation between frames prevented a systematic use of multi-compression, therefore only single compression was used. In the future, occasional decorrelations may be removed by detecting and eliminating the respective frames from the calculation of the image. In vivo, all consecutive frames within a 1.5 s-compression were digitized to minimize undesired inter-frame motion. In vitro, the acquisition was triggered by the displacement of the balloon on the central A-line in the region of interest, disregarding the time interval between frames, to ensure that a controllable and uniform displacement was applied between digitized frames. The position of the balloon was detected and tracked by the software using simple thresholding of the RF signal.

2.2 Data processing

The ADC presented up to 67 ns jitter (corresponding to a 50 µm error in displacement estimates) between the trigger event and the time when the first sample was acquired. These jitters result in erroneous displacement estimates when adjacent RF lines are compared, or when applying a 2D-kernel filter to the results. To avoid these errors, the RF lines were re-aligned by adding a delayed version of the line trigger to the RF signals, and using software detection of the delayed pulse. Sub-sample precision measurement of the jitters was achieved by setting a constant slope (1 mV/ns) to the delayed pulse. RF signals were re-aligned numerically using the shifting property of the Fourier Transform. A ±2 ns standard deviation (residual jitter) was found when detecting the synchronization pulse on the shifted signals, corresponding to ±1.5 µm standard deviation in displacement estimates. Jitter errors induce no error in strain estimates because they result in a constant bias along each single RF line, which is eliminated by the gradient operation. However jitter errors corrupt strain estimates when a 2D-kernel filter is applied to displacement estimates before the gradient is taken. This is because each RF line suffers a different bias. Jitter compensation is illustrated in figure 2.2.

Chapter 2. Development & Characterization

48

0 50 100 150 200 250 300-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

time (ns

voltage (V)

Original signal

0 50 100 150 200 250 300-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

time (ns

voltage (V)

Translation only

0 50 100 150 200 250 300-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

time (ns

voltage (V)

Both slopes assumed equal

0 50 100 150 200 250 300-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

time (ns

voltage (V)

Accounting for different slopes

Figure 2.2: Jitter compensation by re-alignment of a synchronization pulse added to the RF signal. (a) Original signals, (b) alignment after shifting by an integer number of samples, (c) sub-sample precision alignment assuming all signals have exactly the same slope, (d) sub-sample precision alignment accounting for different slopes. The delay between the arrows shows the jitter. In (d), the rising edge is realigned with a 2 ns standard deviation. Interference spikes were observed in the signal spectrum (fig. 2.3). RF data were filtered by numerically removing frequency components outside the transducer bandwidth to minimize noise and interference. Frequency components between 3.0 and 9.3 MHz were preserved (-35 dB cutoff).

Figure 2.3: Amplitude spectrum of the RF signal, measured in a gelatin-based phantom14

14 Experiment internal ref. 2002-1208-PH1-Fantome_c8-1.doc

Jitter

(a) (b)

(c) (d)

Frequency (MHz)

Amplitude spectrum (dB)

Chapter 2. Development & Characterization

49

2.2.1 Displacement estimation using correlation at zero lag

The pre- and post-compression RF lines s1(t) and s2(t) were segmented in overlapping windows x(t) and y(t) (window length T, window shift ∆T). Radial displacements d were estimated from time delays measured between corresponding segments of the pre- and post-compression RF signals. The time delay τ was obtained from the position of the maximum of the normalized cross-correlation function ρ(τ) (Ophir et al. 1991). The cross-correlation coefficient is calculated either in the time domain or in the frequency domain with zero padding. In either case the window length will influence the correlation value. When using a window length of N samples, cross-correlation will be exact at zero lag, but only N–M samples will be used in the calculation of the cross-correlation value at lag M or –M (due to M samples being multiplied by zeros on the edges). Using this method the effective window length is N–M; it is lag-dependant and the value of the normalized cross-correlation function decreases with increasing lags. This phenomenon induces jitter errors, and false peak errors can occur when the true peak is associated with a large lag. To minimize these errors a large window length (much larger than the expected time delay) has to be used. The unbiased cross-correlation value, i.e. normalized by N/(N–M), can be used but false peaks will also appear due to small correlation peaks being amplified by large lags. Finite window length has a second consequence. When the signals x and y are shifted versions of the same signal, some distinctive portions of the signals may be lost when selecting windows xn and yn. An example is shown in Figure 2.4(a). Figure 2.4(b) shows the false peak error (correlation 0.883; lag = –3 samples) on the corresponding cross-correlation function. In this specific case the true peak was the secondary peak located at lag 11 samples.

(a) (b) Figure 2.4: (a) Distinctive cycles at the beginning of the post-compression signal are lost on the pre-compression

signal 15. (b) Corresponding cross-correlation function. In order to overcome these problems, the true cross-correlation function was calculated for each lag M using the zero-lag correlation: The post-compression signal s2(t) was first shifted by M samples, then the post-compression window was selected from the shifted RF signal, and the cross-correlation value was calculated between the two segments at lag zero.

15 Experiment internal ref. 2000-0503-T27-C2223, line 50, window 2341:2496

Chapter 2. Development & Characterization

50

In the case of our previous example, the resulting cross-correlation function is shown in Figure 2.5(c) as a function of the lag M. It can be seen that the true peak (lag = 11 samples) is found with a high correlation (0.990). To gain a better insight of the effect of shifting windows, Figure 2.5(a) and 2.5(b) show the pre- and post-compression signals shifted respectively by the erroneous lag and by the true lag. The good correspondence with the true peak is obvious in fig. 2.5(b), and the mismatch with the false peak is visible in fig 2.5(a).

(a) (b)

(c) Figure 2.5: (a) Pre- and Post-Compression signals shifted by the false peak at –3 samples, (b) Pre- and Post-Compression signals shifted by the true peak at +11 samples, (c) Cross-correlation function samples obtained from shifted signals at zero lag. A major advantage of the zero-lag correlation algorithm is that a short window length can be used without the elastogram being degraded by finite window length effects. It also allows for the estimation of delays larger than the window size. It should be noted that an accurate and time-efficient time delay estimator could be implemented by combining an initial guess of the time delay to re-align the signals, for example using a fast zero-crossings detector (Srinivasan and Ophir 2003) or a sum of absolute differences (Zhu et al. 2002), and the standard cross-correlation calculated at zero lag after re-alignment. The zero lag estimation was implemented through a time efficient code based on cumulated sums. Computationally-efficient displacement estimation was performed by restricting the search range to the expected range of displacements. For example, when the balloon was inflated by 0.2 mm (detected by thresholding on the RF signal), the search range was restricted to –0.1 to +0.3 mm. This scheme also minimized the occurrence of false peaks, as

Chapter 2. Development & Characterization

51

the search range only contained 2-3 peaks. Care had to be taken to choose the search range large enough to ensure all displacement estimates fall within the chosen range, otherwise low-correlation erroneous displacement estimates (false peaks) were found. Once the radial displacement map was obtained, individual false peaks were identified and replaced by a smooth, filtered estimate. False peaks were identified by comparing a filtered version of the displacement map (using a 5x5 median filter) with the initial displacement map. False peaks are large errors that approximately correspond to an integer number of wavelengths. Any sample that differed from the filtered estimate by more than 75 µm (half a wavelength) was considered to be a false peak, and replaced by the corresponding filtered displacement. The resulting displacement map is identical to the initial displacement map, except for individual false peaks. Radial strain was calculated from the gradient of the radial displacements in the radial direction. After false peak removal, a 5x5 (2 mm x 2° aperture) median filter was applied to the displacements before the gradient was calculated. The filter increased the elastographic signal to noise ratio at the cost of resolution, whereas the overall size of the image was unchanged. The size of the filter kernel was kept small to minimize the additional loss in resolution. Unless otherwise specified, all elastograms were calculated using the zero-lag correlation with multi-compression, 1 mm (1.3 µs) window length T and 0.5 mm window shift ∆T , non-overlapping windows, false peak removal, without lateral motion compensation and without adaptive stretching (these techniques are explained in the following paragraphs). 2.2.2 Staggered strain estimates (interleaved gradient)

Strain does not necessarily need to be estimated from consecutive displacement estimates, but can be obtained using the staggered strain estimator (Srinivasan et al. 2002) (interleaved gradient) from any two displacement estimates along a line, i.e. skipping over N samples:

TNdd iNi

i ∆−

≈ +

.ε Equation 2.1

The effective window shift using this estimator is N∆T. The elastographic SNRe was shown to increase linearly with the square root of the window shift (Equation 1.42). Using the staggered strain estimator results in a N improvement in SNRe, with no significant degradation in resolution (Srinivasan et al. 2002). Strain estimation using non-overlapping windows is a particular case of staggered strain estimation, where the effective window shift (i.e. the number of skipped samples N) is chosen so that non-overlapping windows are used in the gradient operation. This is true for:

TTN

∆≥ , N integer Equation 2.2

Displacements estimates obtained from overlapping windows are not independent but partly correlated. For this reason, the time delay estimation error in consecutive overlapping windows are not random, but partly correlated. This phenomenon is likely to induce non-

Chapter 2. Development & Characterization

52

random noise and/or artifacts in the estimated strain. Using non-overlapping windows prevents this phenomenon. 2.2.3 Correction for lateral displacements

Lateral motion is likely to occur in vivo and was investigated. Each post-compression segment was compared with pre-compression segments obtained from 2K+1 adjacent RF lines, where 2K+1 was the lateral search range. Cross-correlation was therefore obtained for each time delay τ and each lateral shift δl. Radial displacement was obtained from the time delay associated with the maximum of the cross-correlation coefficients, and lateral displacement was obtained from the corresponding lateral shift (Konofagou et al. 1998). Unlike the method proposed by Konofagou and Ophir, A-lines were not interpolated. Instead, the lateral displacement δl was obtained by a parabolic interpolation of the cross-correlation function in the lateral direction (i.e. at constant time delay, over three consecutive lines), resulting in a significant improvement in calculation time. 2.2.4 Adaptive stretching

Adaptive stretching was shown to improve the elastographic SNRe (Alam et al. 1998, Brusseau et al. 2001, Srinivasan and Kallel 2002) and was also investigated. Strain was calculated from the gradient of the displacements using a two-iterations adaptive stretching estimator, described by Srinivasan and Kallel et al. (2002). The first iteration calculated an elastogram using a uniform global stretching (Varghese et al. 1996). This initial elastogram was smoothed and used as an initial guess to perform local stretching of the post-compression RF. In a second iteration, the final elastogram was calculated using the locally-stretched RF signals. 2.2.5 Multi-compression

High SNRe can be achieved using a multi-compression sequence, i.e. by cumulating small compression steps (Varghese et al. 1996). Consecutive RF frames were acquired during the compression, and individual displacement maps were calculated as described in the previous paragraphs. The multi-compression elastogram was obtained from the gradient of the sum of the displacements. While this is analytically identical to cumulating individual elastograms (sum of the gradients), experiments showed better elastograms using the gradient of the sum. In practice, the numerical gradient operation amplifies noise in the elastograms. Only one gradient operation is used when cumulating displacements, whereas multiple gradients are calculated to cumulate elastograms. This may explain why cumulated elastograms were noisier than the gradient of cumulated displacements.

Chapter 2. Development & Characterization

53

2.3 System validation on phantoms16

2.3.1 Homogeneous phantom

The elastography system (without the HIFU probe) was validated using a homogeneous phantom made of 6% by weight gelatin (225 Bloom from porcine skin, SKW Biosystems, France), 3% agar and 91% de-ionized and degassed water. The phantom was an 80 mm-radius x 70 mm high cylinder. A 40 mm diameter cylindrical hole at the center of the phantom represents the rectal cavity, inflated by a balloon. The phantom was left unconfined, immersed at the bottom of a tank filled with degassed water. The imaging probe was held vertical. A radial compression was applied from the inside using the balloon. The position of the balloon was used to trigger the acquisition, using a multi-compression sequence with a 0.15 mm displacement step. Elastograms were calculated using 2 mm windows, 0.5 mm window shift, without lateral motion compensation and without adaptive stretching. The experimental elastograms are shown in figure 2.6. Increasing decorrelation (in blue) is observed with increasing compression steps. The average strain (red bar) is erroneous in the last figure because of decorrelation noise. (a) (b) (c)

Figure 2.6: Experimental single-compression elastograms for (a) 0.15 mm displacement of the balloon, 0.23% average strain, (b) 0.60 mm displacement, 0.86% average strain, (c) 1.05 mm displacement, 1.57% average

strain. Dimensions are given in mm, strain in %. A simulation of the experiment was performed. The phantom was represented by an unstructured, triangular finite element mesh (FEM). The FEM model (Kallel et al. 1996) was designed to calculate the strain in an inhomogeneous, isotropic, elastic body which is subjected to natural boundary conditions set by the internal pressure change, in a plane strain state. The uniform pressure (natural or Neumann boundary condition) was applied at the inner rectal wall. In order to simulate the conditions met in transrectal prostate elastography in-vivo, the outer boundaries of the model are first left free and then confined laterally in regions corresponding to the pubic and Coccyx regions. The Poisson’s ratio was set to 0.495, accounting for a quasi-incompressibility condition. Severe strain decay with depth was observed because of the cylindrical geometry (Shapo et al. 1995, Souchon and Soualmi 2002). Strain decays linearly with 1/r2, where r is the radial distance in the cylindrical coordinates attached to the cylindrical phantom. The experimental and simulated average strain profiles are shown in figure 2.7. A 1/r2 fit was matched to these profiles and is shown in the figure.

16 Publications based on, or including part of, this section:Srinivasan Kallel et al. 2002, Souchon Soualmi et al. 2002.

-50 0 50

20

40

60

0

0.2

0.4

0.6

0.8

-50 0 50

0.5

1

1.52

2.5

-50 0 50 0

1

2

3

4

Chapter 2. Development & Characterization

54

(a) (b)

Figure 2.7: (a) Experimental and simulated strain profiles, (b) experimental strain profile and 1/r2 fit. Minimal lateral motion was expected because of the radial geometry. Using a plate compressor (Ophir et al. 1991), the material is allowed to expand laterally with increasing lateral motion on the edges of the plate. For radial compression, both theory and FEM simulations of a homogeneous phantom show that displacement is purely radial, with no lateral (tangential) motion. In phantom experiments, the “curtain noise” (decorrelation due to large lateral motion on the edges of the phantom) previously reported for a plate compression (Konofagou et al. 1998) was not observed using a balloon compressor. However lateral motion was observed when no confinement prevented the phantom from drifting. 2.3.2 Phantom with a stiff inclusion

In order to validate the capability of the elastography system to visualize an inclusion, a stiff cylindrical inclusion (12% by weight gelatin, 3% agar, 85% water) was embedded in a cylindrical phantom made of 62% gelatin, 2% agar and 92% water. The inclusion was approximately 1.5 times stiffer than the background (Kallel et al. 2001) and 15 mm in diameter. Corresponding FEM simulations were performed. Fig. 2.8 shows the experimental setup, with the imaging probe vertically inserted into the phantom.

Experimental elastograms (figs. 2.9a and 2.9b) and simulated strain map (fig. 2.9c) are in good agreement. The high strain rim around the inclusion was due a very soft layer of gel that appeared at the interface of the inclusion a few days after the phantom was prepared. Strain concentrations were observed around the inclusion, with a typical “butterfly” pattern. Strain concentration is a known mechanical behavior that has been largely investigated in the literature (Goodier 1933, Mushkelishvili 1963). Strain decay was also observed. In order to compensate the strain decay, the strain was multiplied by r2 before being displayed to perform strain decay compensation. This method is equivalent to a time-gain compensation (TGC) in sonography. It amplifies strain (as well as noise) in distal low strain areas, and provides a

uniform image of the homogeneous phantom. r2 strain decay compensation (figure 2.9b) was shown to improve the visual

Figure 2.8: Experimental setup

Chapter 2. Development & Characterization

55

perception of the elastograms, providing a uniform background and enhancing strain concentrations.

(a) (b) (c) Figure 2.9: (a) raw experimental elastogram, (b) experimental elastogram after strain decay compensation, (c) simulated strain map (no ultrasonic noise) after strain decay compensation.

2.4 Experimental characterization

2.4.1 Materials and methods

A homogeneous phantom (6% gelatin, 2% agar, 92% water) was used for experimental system characterization. Four sets of elastograms were acquired from different areas in the same phantom, for displacements ranging from 0.15 mm (0.23% average strain) up to 1.05 mm (1.57% average strain) at the balloon interface, by 0.15 mm steps. Elastograms were calculated using 2 mm windows, 0.4 mm window shift, conventional gradient (no staggering), lateral motion correction (search kernel +/- 1 line), and two-iteration adaptive stretching. The strain profiles were averaged over a 120° aperture (300 RF lines). The average strain inside the phantom was measured. For each individual elastogram, depth-dependent SNRe was measured at constant depth, using the mean to standard deviation ratio in strain. Experimental SNRe was also measured on a multi-compression sequence with eight 0.15 mm-displacement steps, using the same RF data set as for single-step compressions. SNRe was predicted from equation 1.61. 1/r2 strain decay and depth-dependent sonographic SNR were included into the theoretical model. Ultrasonic parameters were f0=5.5 MHz central frequency, B/f0=40% fractional bandwidth, α=54 Np/m attenuation at 5.5 MHz. Attenuation in the phantom was estimated from experimental measurements of sonographic SNR with depth17. It was similar to the attenuation measured in the prostate in vivo (0.83 dB/cm/MHz, i.e. 53 Np/m at 5.5 MHz) (Ribault et al. 1999). Processing parameters were set to T=2.7 µs window length, ∆T=0.53 µs window shift 2.4.2 Simulation and experimental results

Figure 2.10 shows (a) the theoretical SNRe plotted vs. depth for different values of the attenuation coefficient, from 0 (no attenuation, strain decay only) up to 100 Np/m. SNRe was expected to decrease at large depth due to low sonographic SNR. A threshold is observed near 60 mm for 100 Np/m attenuation. It corresponds to the threshold between the Cramér-Rao

17 Experiment internal ref. 2002-0923 and 2002-0701

Chapter 2. Development & Characterization

56

region and the Barankin region. In the strain filter this threshold is usually associated with high strains, but here it is due to an extremely low sonographic SNR. In (b) SNRe is shown for different applied strain, assuming a constant attenuation. As strain increases, a low SNRe area appears near the balloon. It is associated with high strains that induce signal decorrelation. The edge of the decorrelation area shifts deeper as the applied strain is increased. A second threshold is observed near 90 mm, it corresponds to the edge of the Cramér-Rao region and is associated with low sonographic SNR. The second edge does not shift significantly when the applied strain changes. SNRe increases in deep areas as the applied strain increases, as long as high strains do not induce decorrelation. Overall, for a given applied strain, high SNRe is achieved only within a limited depth range. This result suggests that this high SNRe area can be positioned at the area of interest by changing the applied strain. A possible application would be to increase SNRe in deep areas by increasing the applied strain.

20 30 40 50 60 70 80 90 1000

2

4

6

8

10

12

Depth (mm)

SNRe

SNRe at 1% average strain

Strain decay only Attenuation 50 Np.m-1 Attenuation 100 Np.m-1

20 30 40 50 60 70 80 90 1000

2

4

6

8

10

12

Depth (mm)

SNRe

SNRe vs. Depth

0.25% 0.5% 1% 2% 4% 8%

(a) (b) Figure 2.10: Theoretical SNRe plotted vs. depth (a) for different attenuations at 1% applied strain and (b) for different applied strains with a 54 Np/m attenuation coefficient. Figure 2.11 shows (a) the average strain profiles obtained for single-step compressions ranging from 0.23% up to 1.57%. For average strains up to 0.64% (displacement of the balloon up to 0.45 mm), smooth strain profiles are observed. Between 0.86% and 1.09% average strain (0.60 to 0.75 mm displacement), noisy strain estimates corrupt the strain profile in the proximal region, whereas the strain profile is still smooth between 50 and 65 mm. For higher strains, the whole profile is noisy. (b) Experimental SNRe was maximal for small applied strain (up to 0.64%, i.e. displacements up to 0.45 mm) but did not exceed 3.0. A similar shape was obtained using non-overlapping windows staggering with an increase in maximum SNRe (approximately 4.0). The experimental SNRe values are lower than the theoretical values. As predicted by theory, for large applied strains, decorrelation occurs at first in the high-strains area near the balloon (for 0.86% applied strain, SNRe is low between 20 and 40 mm), and for increasing strains the low SNRe area extends deeper into the gel. Experimental results differ from theory as the maximum strain does not increase in deep areas when the applied compression is increased. Strain decay and attenuation effects do not explain this behavior. From experimental observations, it is very likely that increasing strains were associated with increasing lateral and/or elevational displacements, which in turn induced signal decorrelation and decreasing SNRe. This observation suggests that in practice small compression steps (0.15 to 0.45 mm displacements) are desirable to achieve high SNRe. Each individual small compression step results in accurate displacement estimates, but in poor elastographic signal-to-noise ratio. High SNRe can then be improved using a multi-

Chapter 2. Development & Characterization

57

compression scheme. Variations in ultrasonic beam width, in beam overlap, and out-of-focus effects are also likely to degrade the experimental SNRe.

(a) (b) Figure 2.11 : (a) Experimental strain profiles and (b) experimental SNRe, averaged over four realizations, using single-step compression, for average strains ranging from 0.23% up to 1.57%. The strain is plotted vs. depth, the balloon/phantom interface is located at 20 mm. Figure 2.12 shows the experimental SNRe and the average strain profile using an eight-step multi-compression sequence without staggering (total displacement 8*0.15 mm, total applied strain 2.0%). Strains as high as 5% were induced near the balloon while SNRe was increased up to 6.0 (to be compared with SNRe 3.0 for 0.23%-0.64% average strains). No decorrelation was observed near the balloon, and SNRe decreases almost linearly with depth. This result shows that high strains and high SNRe can be achieved using a multi-compression sequence. Using this technique, inter-frame strain and inter-frame decorrelation due to undesired displacements are minimized.

20 30 40 50 60 700

1

2

3

4

5

6

7

Depth (mm)

Stra

in (%

)

Strain Profile & SNRe

Strain (%)SNRe

Figure 2.12 : Strain profile and SNRe plotted vs. depth for a multi-compression sequence (average strain 2.0%) 2.4.3 Conclusion

Experimental SNRe was measured in a homogeneous phantom and compared with theoretical values for typical data processing parameters (2 mm window length, 0.4 mm window shift). The theoretical model accounted for strain decay with depth, and for depth-dependent sonographic SNR due to ultrasonic attenuation. Experimental SNRe was lower (maximum 3.0) than theoretical SNRe (maximum 12), and increased to 4.0 using non-overlapping

Chapter 2. Development & Characterization

58

windows (staggering). The shape of the curve (SNRe vs. depth) was similar between theory and experiments, showing low SNRe near the balloon for high applied strains, a region where SNRe was maximum, and a decrease with depth. The theoretical SNRe values obtained using a 1D model (i.e. assuming that scatterer motion occurs only in the direction of propagation of the ultrasonic pulse) predicted that high SNRe could be achieved at large depth by increasing the applied strain, experiments showed that large applied strains did not increase SNRe in the distal region. This phenomenon was most likely due to increasing lateral and/or elevational displacements associated with increasing strains. These undesired displacements were not accounted for in the theoretical model. The experimental results suggest that unless undesired displacements can be avoided or at least minimized, small compression steps (0.15 to 0.45 mm at the balloon interface) are preferred for achieving optimum SNRe. The highest SNRe was obtained using a multi-compression sequence that demonstrated the feasibility of imaging high strains at high SNRe.

Chapter 3. Prostate Cancer Detection

59

3 Application to prostate cancer detection Le présent chapitre traite de la faisabilité de la détection du cancer de la prostate par le système d’élastographie décrit précédemment. Il est organisé en deux étapes, allant d’une observation in vitro à) l’application clinique in vivo. La première partie est consacrée à l’étude de l’élastographie de la prostate humaine in vitro. Les objectifs principaux sont de décrire l’aspect de la prostate telle que vue en élastographie, d’identifier les structures visibles, et de déterminer si le cancer peut être détecté. La méthode choisie pour ceci consiste à mettre en correspondance et à comparer l’élastographie avec les coupes anatomo-pathologiques sur des prostates humaines récemment excisées. Ceci a été réalisé sur 15 pièces opératoires. Les résultats de cette première partie confirment que non seulement le cancer, mais aussi d’autres types de nodules et même l’anatomie prostatique sont observables par élastographie. En effet l’élastographie a permis de voir l’anatomie zonale de la prostate (verumontanum, zones périphérique et de transition) et un certain nombre de nodules bénins et malins. Les tumeurs de petites taille (<0.5 cm3) n’étaient cependant pas visibles. Afin d’augmenter la spécificité du diagnostic, des critères de différenciation entre les caractères bénins et malins sont proposés, en particulier basés sur le contraste et sur la netteté des contours. La seconde partie rapporte notre expérience de détection du cancer de la prostate par élastographie in vivo. L’objectif de cette étude était de déterminer si l’élastographie utilisant le ballon comme compresseur pouvait permettre de détecter le cancer en situation clinique. A cette fin 69 patients devant être traités par une thérapie HIFU ont été examinés. Ce choix était motivé par la simplicité de mise en œuvre, la sonde d’imagerie et le ballon étant déjà présents sur le dispositif HIFU. A cause des contraintes de temps liées à la thérapie, chez chaque patient 3 images seulement ont pu être acquises pour couvrir la totalité de la prostate. Il s’est avéré que le passage du laboratoire (in vitro) à la clinique n’était pas immédiat. La grande majorité des premiers élastogrammes étant noyés dans un important bruit de décorrélation, il a fallu identifier les critères permettant de recouvrer, ou tout du moins d’essayer d’approcher, la qualité d’image auparavant observée in vitro. C’est ainsi que le rôle majeur de la cadence d’acquisition des signaux RF a été mise en évidence, et le système d’acquisition a été modifié en conséquence. La faisabilité de la détection du cancer par notre système d’élastographie a alors été montrée par 5 cas pour lesquels la lésion visible dans les élastogrammes était confirmée par les biopsies et l’échographie. Le nombre de cas confirmés est resté faible d’une part à cause du peu d’images par patient mais aussi à cause du grand nombre de cas dans lesquels les biopsies et l’échographie étaient contradictoires.

Chapter 3. Prostate Cancer Detection

60

3.1 In vitro18

3.1.1 Objectives

Elastography is expected to be able to detect prostate cancers that are stiffer than normal tissues. Although there is supporting evidence that many prostate cancers possess such characteristics, it is likely that some do not. The sensitivity of the technique will therefore depend on the percentage of prostate cancers that have sufficient stiffness contrast to make them detectable in elastograms. It is also expected that other benign prostate structures may exhibit a stiffness contrast that will make them visible in elastography. Among these structures, some may be mistaken for cancer. In order to estimate how reliable the cancer detection is, the positive predictive value (PPV: percentage of areas identified as possible malignant tumors that turn out to be cancer) of the technique needs to be estimated. It is expected that both sensitivity and PPV can be increased as the reader gains improved knowledge of the elastographic appearance of the various prostate structures that can be encountered. Using sonograms in conjunction with elastograms with perfect registration is also easily feasible (as both sonograms and elastograms are generated from the same set of RF data) and is likely to improve the performances. For example, it has been reported that breast carcinoma appears larger in elastograms than in sonograms in vivo, whereas fibroadenoma has the same size in both imaging modalities (Ophir et al. 1996a, Ophir et al. 1996b, Garra et al. 1997, Hall et al. 2003).

Figure 3.2 : Typical pathology slice. The peripheral zone (PZ) and the two lobes of the transition zone (TZ) are visible. The urethra (U) is the central opening located anterior to the verumontanum (V). A malignant tumor was marked in black in the PZ by the pathologist. The size of the tumor is usually much smaller than the outline suggests because care is taken not to overwrite the tumor boundaries when marking the slice. The tumor cannot be seen from macroscopic examination. In order to investigate the performances of elastography for prostate cancer detection, elastograms were acquired from 15 prostates in vitro and were compared with pathology. A typical pathology slice is shown in figure 3.2 to illustrate prostate anatomy. The objectives of the study were (1) to identify benign prostate structures visible in the elastograms, and (2) to assess the capability of our elastographic system to detect malignant tumors. For this purpose, sensitivity and PPV were estimated by a radiologist. This study also aims at providing

18 Publications based on, or including part of, this section: Souchon Hervieu et al. 2003

TZ TZ U

PZ

PZ PZ

V

Anterior

Posterior

Chapter 3. Prostate Cancer Detection

61

information about the prostate elastogram quality that can be obtained in a controlled experiment and that can be expected in vivo if undesired effects can be minimized. 3.1.2 Material and method

15 prostates were obtained from patients who underwent radical prostatectomy for prostate cancer at the urology department at Edouard Herriot Hospital in Lyon. These patients were diagnosed with moderately aggressive tumors (Gleason score 6-8) at clinical stage T1 (non palpable) or T2 (palpable) (TNM classification, Sobin et al. 2002). The prostates were embedded in a cylindrical (100 mm height, 90 mm radius) gelatin-based gel (10% by weight gelatin, 90% water) within one hour after surgery. The gel had a 40-mm diameter cylindrical hole in its center to insert the imaging probe vertically. The composition of the gel was chosen from previous experiments showing that clean elastograms were obtained on porcine liver using these proportions19. The prostate was usually positioned upside down, i.e. the apex near the top and the base at the bottom of the gel, and the imaging probe was held vertically from above. Using this convention, the ultrasound probe had the same orientation as in the real clinical situation, and the left side of the images corresponds to the right side of the prostate (standard convention in radiology). Occasionally the shape of the prostate did not allow for stable positioning using this orientation, and the prostate was positioned with the apex at the bottom of the gel, resulting in non-standard orientation of the images (flipped left/right). In vivo, the posterior face of the prostate (peripheral zone) is adjacent to the rectal wall. It was positioned toward the imaging transducer, in order to compress and image from the same direction as in vivo. The gel was left for 1-2 hours in a refrigerator to solidify. Then the gel was immersed in a tank filled with water and confined between three evenly spaced vertical rods and a rigid 50-mm wide confinement plate designed to simulate the pubic bone. The water temperature was 20-22°C. The experimental setup is shown in figure 3.1.

Figure 3.1: (a) Experimental setup showing the acquisition system (left), the ultrasound scanner (back), the water tank and the transrectal imaging probe (right). (b) The imaging probe (white) is held vertical, three vertical rods (gray) stabilize the gel, the curved plate was not used. (c) Zoom on the prostate and on the confinement plate. The elastography system described in the previous chapter was used to acquire a multi-compression sequence. Multi-compression was chosen to achieve high SNRe (2.4 – Experimental characterization). The displacement of the balloon served as a reference to apply uniform steps. Twelve 0.15-mm displacement steps were used. The extremity of the apex was found in transverse B-mode scans, and the location of the probe was recorded as the origin. Then transverse elastograms were acquired every two mm from the apex to the base.

19 Reports Ref. 2003-0429 and 2002-0701, Experiment Ref. 2002-0627

(a) (b) (c)

Confinement plate

Chapter 3. Prostate Cancer Detection

62

The prostates were returned to the histology department of the hospital by the end of the day, or were preserved in formaldehyde to be returned the day after. Histologic examination is always performed after radical prostatectomy to assess the extent of the malignant tumors and this information determines the therapeutic orientation of the patient after surgery. Therefore the prostates had to be returned undamaged. A preliminary experiment conducted on rabbit and rat liver embedded in an identical gel showed that this procedure created minimal tissue damage (only one cell layer was affected on microscopic examination)20, provided that the temperature of the mixture was 28-29 °C at the time tissues were immersed. At lower temperatures, the gel started to solidify immediately when in contact with the tissues, resulting in undesirable trapped air bubbles around the tissues. At higher temperatures (between 30-40 °C), a 500 µm layer of coagulation necrosis was observed on the liver capsule and was deemed unacceptable for prostates. Liver tissue was chosen because it was expected to be more sensitive than prostate tissue. Once the data acquisition was finished, used materials were cleaned using a disinfectant active against HIV-1 and HBV (Surfanios, Laboratoires Anios, France), or Javelle water (bleach solution). The multi-compression elastograms were processed using 1-mm window length, 0.5-mm window shift, non-overlapping windows, no lateral motion compensation, no adaptive stretching. The elastograms were normalized by the mean strain value in the elastogram (samples with correlation < 0.90 were rejected for average strain calculation). For normalized strains below 1.0 (i.e. for strain < average strain) the normalized strain was shown using a linear grayscale, whereas the square root of the normalized strain was displayed (similar to a gamma correction) for normalized strains higher than 1.0. On a grayscale that ranges from 0 to 1, 70% of the gray levels (0-0.7) were allocated to strain below the average, and 30% (0.7-1) to strains above the average (Fig. 3.3). This scheme was used in order to visually enhance low strain areas where cancer is expected, while minimizing saturation in high strain areas. The drawback of this process it to enhance noise in areas where strain is close to zero, eventually enhancing positive (tensile) strains corresponding to noisy estimates. A red correlation mask was applied to hide noisy areas where normalized the cross-correlation was less than 0.75. Figure 3.3: Gray level (g) allocation vs. normalized strain. ε0 denotes the average strain in the image. Increased dynamic range (70% of the gray levels) was allocated to low strains, and white saturation was minimized using

20 Experiments Ref. 2002-0502 and 2002-0624

0 0.5 1 1.5 20

0.2

0.4

0.6

0.8

1

Gra

y le

vel

Black

White

20εε

=g

Normalized strain ε/ε0

02εε

=g}

}

Chapter 3. Prostate Cancer Detection

63

Microscopic evaluation of the extent of the cancer was performed according to standard medical practice, and malignant tumors were marked on pathology (transverse) slices. The slices were registered with elastograms using information on slice thickness and slice orientation provided in the anatomy-pathology reports. An initial reading was performed by Dr. Valérie Hervieu and Dr. Florence Mège-Lechevallier (pathologists) to identify benign prostate structures visible in elastograms. Then a blind reading of was performed by Dr. Olivier Rouvière (radiologist) using the corresponding elastograms and sonograms. True positives (TP) (suspicious areas whose position correspond to cancer in the corresponding pathology slice) and false positives (FP) (suspicious areas that did not correspond to any cancer on pathology) were recorded. No true negative (TN) was found because all patients had been previously diagnosed with prostate cancer using biopsies. False negatives (FN) (tumor not visible in the elastogram) were defined as the number of tumors seen in pathology that were not visible in the corresponding elastogram (i.e. FN = total number of tumors minus TP). Sensitivity (SE) and positive predictive value (PPV) were determined. Sensitivity is the relative number of tumors that were correctly identified (i.e. SE = TP divided by total number of tumors), the positive predictive value is the relative number of suspicious areas found in elastograms that were real tumors (PPV = TP/(TP+FP)). No true negatives were included in the study, therefore specificity could not be determined. Results were split into two categories between cancer in the peripheral zone (PZ) and cancer in the transition zone (TZ), as cancer detection is expected to be more difficult in the transition zone because of its heterogeneity. Results were also split between “clinically significant” (volume ≥ 0.5 cc) and “clinically insignificant” (volume < 0.5 cc) cancers, expected to be hardly detectable. 3.1.3 Results

Of the 15 prostates, only 13 were included in the study: one was excluded because of a error in prostate orientation (prostate #2), another because the gel was broken (prostate #13).

Chapter 3. Prostate Cancer Detection

64

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.41%)

-30 -20 -10 0 10 20 3020

30

40

50

60

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.41%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 3020

30

40

50

60

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.09%)

-30 -20 -10 0 10 20 3020

30

40

50

60

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.09%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 3020

30

40

50

60

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.17%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.17%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 3020

30

40

50

60

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.29%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.29%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 3020

30

40

50

60

(a) (b) (c) (d) Figure 3.4: Series of (a) pseudo-color elastograms, (b) grayscale elastograms, (c) sonograms and (d) corresponding pathology slices, acquired every 2 mm on a human prostate in vitro21. The elastograms were normalized by the average strain and shown using a nonlinear color map to enhance low strain areas. The color bar shows the normalized values. The same scale was used in all four columns. A typical series of elastograms is presented in figure 3.4, with the corresponding sonograms generated using the same set of RF data. The embedding gel was anechoic, therefore decorrelation was observed around the prostate. In these images, prostate anatomy is clearly visible in the elastograms. The peripheral zone appears as a high strain (bright) area with a characteristic “ω” shape near the apex (top image), it becomes flat in the mid-gland and near the base. At the center of the peripheral zone, the verumontanum appears as a stiff (dark) inverted “v”-shaped area. The two lobes of the transition zone are clearly visible as two low strain (dark) lobes. Strain contrast between the stiff transition zone and the high strain peripheral zone was consistently observed in all 13 cases. The contrast may be related to a strain decay with depth, but it is more likely to be representative of a true stiffness contrast because high strains are also observed in the lateral and anterior faces of the peripheral zone.

21 Experiment Ref. 2002-1129-JM

Chapter 3. Prostate Cancer Detection

65

0

0.5

1

2.25

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.41%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

0.5

1

2.25

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.41%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 3020

25

30

35

40

45

50

55

60

65

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.94%)

-30 -20 -10 0 10 20 3020

30

40

50

60

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.94%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 3020

25

30

35

40

45

50

55

60

65

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.35%)

-30 -20 -10 0 10 20 3020

30

40

50

60

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

x (mm)

y (m

m)

Normalized Elastogram (mean strain 2.35%)

-30 -20 -10 0 10 20 3020

30

40

50

60

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 3020

25

30

35

40

45

50

55

60

65

(a) (b) (c) (d) Figure 3.5: Prostatic glandular hyperplasia (blue outline) visible in (a) the pseudo-color elastogram, (b) grayscale elastogram, (c) the sonogram and (d) the corresponding pathology slice. Cancer (red outline) is also visible on the left side of the prostate (right side of the images)22. The series represents 3 consecutive planes spaced by 2 mm. Figure 3.5 shows glandular hyperplasia in the transition zone visible in (a-b) the elastogram, (c) the sonogram and (d) the corresponding pathology slice. It is visible as a light pink nodule in the pathology slice. Hyperplasia is a general medical term referring to excess cell replication. Glandular hyperplasia is a common and noncancerous growth of glandular tissues of the prostate. It usually begins with microscopic nodules in young men and may overgrow with age in the transition zone and form BPH (benign prostatic hyperplasia). Glandular hyperplasia was identified in the elastograms in 4 out of 13 prostates, and was depicted as very stiff circular or ellipsoidal areas with sharp edges enhanced by high strains. In the figure cancer is also visible in the peripheral zone. It exhibits a lower strain contrast than glandular hyperplasia.

22 Experiment Ref : 2002-1127-PG

Chapter 3. Prostate Cancer Detection

66

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.34%)

-30 -20 -10 0 10 20 30

30

40

50

60

70

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.34%)

-30 -20 -10 0 10 20 30

30

40

50

60

70

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 30

30

40

50

60

70

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.38%)

-30 -20 -10 0 10 20 30

30

40

50

60

70

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.38%)

-30 -20 -10 0 10 20 30

30

40

50

60

70

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 30

30

40

50

60

70

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.33%)

-30 -20 -10 0 10 20 30

30

40

50

60

70

-0.2

0

0.2

0.4

0.6

0.8

1

1.44

1.96

2.56

x (mm)

y (m

m)

Normalized Elastogram (mean strain 1.33%)

-30 -20 -10 0 10 20 30

30

40

50

60

70

0

50

100

150

200

250

x (mm)

y (m

m)

Echography

-30 -20 -10 0 10 20 30

30

40

50

60

70

(a) (b) (c) (d) Figure 3.6: Myomatous nodule (orange outline) in the prostate, visible in (a) the pseudo-color and (b) grayscale elastograms, (c) the sonogram and (d) on the corresponding pathology slice23. Cancer (black outline) is not visible in the elastogram. The series represents 3 consecutive planes spaced by 2 mm. In figure 3.6 a myomatous nodule is visible in (a-b) the elastograms, (c) the sonogram and appears as a dark pink area on (d) the corresponding pathology slice. Myomatous nodules are composed of fibrous (muscular) hyperplasia, they are benign and common in the prostate. They are usually hypoechoic in the sonograms. Four were identified in the elastograms and seen as ellipsoidal areas. Their outline was visible as a rim of high strains, whereas their interior had no significant strain contrast compared to surrounding tissues. On some occasions, they were seen as decorrelation areas (correlation < 0.75) due to low sonographic signal-to-noise ratio inside the nodule. In figure 3.6d, a small malignant tumor is marked with a black outline on the pathology slice. Its volume was considered negligible (a few malignant cells) and was therefore not reported on the anatomy-pathology report. It was not visible in the elastograms. Multiple glandular nodules are also visible in the pathology slice, and nodule contours can be seen in the elastograms.

Large tumors (≥ 0.5 cc) Insignificant Tumors (< 0.5 cc) Peripheral zone Transition zone Peripheral zone Transition zone

No. of cancers 10 6 19 6 True positive 4 1 0 0 False positive 5 0 0 0 Sensitivity (%) 40 % 17 % 0 % 0 % PPV (%) 44 % 100 % N/A N/A Table 3.1: Total number of cancers, true positives and false positives observed from elastograms in 13 prostates in vitro24.

23 Experiment Ref : 2002-0912-BB 24 Document Ref : 2003-0612.xls

Chapter 3. Prostate Cancer Detection

67

Table 3.1 shows the total number of malignant tumors identified in each zone of the prostate, the number of true positives identified from the elastograms by the radiologist, and the number of false positives. From the litterature, cancer is mostly expected in the peripheral zone. In the peripheral zone, ten “large” tumors (≥ 0.5 cc) were identified by microscopic examination. Four were visible in the elastograms. They had Gleason scores 6 (3 cases) or 7 (1 case), stade T1 (2 cases) or T2 (2 cases), and volume 1.0 to 20.4 cc. Five false positives were found in the peripheral zone. In the transition zone, six “large” tumors were identified by histology, only one (Gleason score 7, stade T2, volume 1.4 cc) was correctly identified as a true positive in the elastograms, the five remaining tumors were not visible. No false positive was found in the transition zone. Numerous “insignificant” tumors were identified by histology in the peripheral (19) and transition (6) zones. None was visible in the elastograms, and no “insignificant” (< 0.5 cc) false positive was found (i.e. all false positives were bigger than 0.5 cc). As a consequence, sensitivity was 0% for “insignificant” tumors and the PPV value could not be determined. 3.1.4 Discussion

The sensitivity, specificity, and PPV values given in table 3.1 are too low to be of any clinical interest. But these values are very preliminary and should not be considered as the true performances of elastography for prostate cancer detection. First the number of prostate samples is small. Second, the resolution, SNRe, and CNRe of the elastograms suffers from the short bandwidth of our ultrasound system. Image quality would improve with high-bandwidth transrectal probes currently used in radiology. Third, cancer detectability is likely to depend on the amount of pre-compression but this parameter was not exploited in the current study. Finally, the stiffness contrast between cancer and normal might have been affected by the low temperature of the gel. These limitations are discussed below. In our experiments, prostate elastography was therefore performed with a sub-optimal system. It is highly likely that the true performances were underestimated. High sensitivity, high specificity and high PPV are required for “significant” tumors, especially in the peripheral zone where most cancers are expected. 40% sensitivity and 44% PPV were found for “large” tumors in this region. The transition zone is usually heterogeneous and cancer detection in this area using conventional sonography is considered to be difficult. 17% sensitivity and 100% PPV were found in the transition zone. Of course, this value is not statistically significant because of the small number of cases. Still, it is remarkable that no false positive was found in the transition zone in spite of the heterogeneous aspect of this zone. It is also interesting to note that none of the benign structures identified by the pathologists was mistakenly considered as malignant by the radiologist, even during a first blind reading performed without the a-priori knowledge provided by the pathologists. Although the topic is still subject to controversy, in the standard medical practice “insignificant” tumors are considered not to have an impact on the patient prognosis25 ; their detection is not critical. None of these small tumors were detected in the elastograms. This was expected because the detectability of a tumor is related to its stiffness contrast and to its 25 Prognosis : The probable outcome or course of a disease; the patient's chance of recovery.

Chapter 3. Prostate Cancer Detection

68

size (Bilgen 1999). Some prostate tumors may not exhibit a significant stiffness contrast when compared to normal tissues and may therefore not be detectable in elastography. This lack of contrast can be expected for well-differentiated (i.e. low Gleason grade) malignant cells. The main criteria used by the radiologist to identify the tumors was the presence of a low-strain area consistently visible in several consecutive planes. Low strain regions that appeared in one elastogram only could not be reliably identified as tumor, benign structure, noise or artifact. This observation suggests that a single elastogram is not sufficient to detect prostate cancer, and that prostate cancer detection in a full clinical implementation would benefit from a large number of slices with a thin inter-slice spacing. Stiffness contrast is expected to depend on the amount of pre-compression because of the nonlinear behavior of prostatic adenocarcinoma (Krouskop et al. 1998): tumors are expected to become stiffer with increasing pre-compressions between 2% and 4%, while the stiffness of normal tissues does not change significantly in this range. Increasing the pre-compression is therefore likely to increase the stiffness and the strain contrast (Varghese and Ophir 2000) of the prostate tumors in the elastograms. However an optimal pre-compression may need to be determined because the nonlinear behavior of tissues may in some cases induce a loss or an inversion in strain contrast (Varghese and Ophir 2000), as was observed by Hall et al. (2003) on breast fibroadenoma in vivo. Hall et al. observed that individual strain images could be misleading, whereas the strain image sequence during a continuously increasing compression for various breast pathologies was unique. The largest tumor found in our study was 20.4 cc in volume26, it occupied the whole transition zone and invaded a part of the peripheral zone. The tumor was not seen in the transition zone in the elastograms, but the extension into the peripheral zone was visible (true positive). Interestingly, the average strain in this prostate (0.78%) was lower than in the other prostates included in the study (min 0.89%, max 2.63%, mean 1.9%). This observation suggests that the average strain value may be an indicator of the presence of a large tumor. Alternatively, non-normalized elastograms could be used to identify the tumor. Using the average strain values or non-normalized elastograms requires that reproducible stress is applied on all slices, which may be difficult in vivo. Some tumors may not be identified in the elastograms because of registration issues: errors are likely to occur, and the imaging plane may not be exactly parallel to the pathology slice. In four cases the left/right orientation of the prostate was not mentioned in the pathology report. In one case, the shape and location of two stiff structures visible in elastograms had good correspondence with two small tumors identified in pathology, but with a 10 mm mismatch in distance (i.e. five slices apart). These structures might really be the cancer, but the mismatch was not deemed acceptable. The above results are expected to be biased because of the temperature dependence of tissue stiffness (Wu et al. 2001). Biological tissues were shown to become stiffer at low temperatures. The temperature was approximately 5-10 °C inside the embedding gel during the experiments. At 15-18°C the gel began melting and compression could not be applied. It is therefore likely that the stiffness contrast between cancer and normal tissues in this experiment was not representative of the stiffness contrast at body temperature in vivo.

26 Ref.: 2002-1008-JLL

Chapter 3. Prostate Cancer Detection

69

3.1.5 Conclusion

Transverse elastograms acquired in vitro on 13 prostates were compared with pathology slices. Prostate elastograms with sharp edges and significant contrast were acquired in vitro, and might be obtained in vivo if undesired motion is minimized. The peripheral zone, the two lobes of the transition zone and the verumontanum were clearly visible in all prostates. Glandular hyperplastic nodules were seen as stiff ellipsoidal areas in four cases, and myomatous nodules were seen either as low-contrast areas surrounded by sharp high strain contours, or as decorrelation areas due to low sonographic signal-to-noise ratio. Four out of ten malignant tumors were detected from the elastograms in the peripheral zone, and one out of six in the transition zone. Tumors with volume less than 0.5 cc were not visible in the elastograms. It is expected that cancer detection performances could be improved by exploiting the nonlinear behavior of tissues.

3.2 In vivo

3.2.1 Objectives

The application of elastography to prostate cancer detection was investigated in vivo on 69 patients undergoing HIFU therapy. The objectives were (1) to investigate the feasibility of prostate elastography using a balloon compression and (2) to quantify the percentage of prostate tumors that could be detected using elastography, using sonography and biopsies as a reference. 3.2.2 Material and method

The imaging probe and the balloon required for prostate elastography in vivo were already part of a HIFU machine used for prostate cancer therapy in Lyon (Ablatherm®, Edap-Technomed, France). The acquisition system described in the previous chapter was connected to the HIFU machine, so that the research protocol on prostate elastography was included in the ongoing study on HIFU therapy. Ultimately, a stand-alone elastography system would be required for prostate cancer detection, but this would require a new approval from the Ethics Committee. At the current stage of the study the combined setup allowed for substantial savings in time. Elastograms were acquired from patients undergoing HIFU therapy of prostate cancer. The active part of the system combined a transrectal HIFU therapy probe and an imaging probe (Fig. 3.7), and was covered by a latex transducer cover filled with a coupling liquid. A computer-controlled motorized holder allowed a stable positioning of the probe. The HIFU probe was present but was not used.

Chapter 3. Prostate Cancer Detection

70

Figure 3.7: A transrectal imaging probe (IP) was connected to a HIFU probe (HP), it provided transverse scans of the prostate (perpendicular to the figure). Elastograms were acquired in three parallel planes at the apex (A), mid-gland (M) and the base (B). The figure shows the bladder (BL) and the urethra (UR), the seminal vesicles (SV) and the ejaculatory ducts (ED). For the first 27 patients (group A), compression of the prostate was manually applied by continuously inflating the balloon using a 60 ml syringe. Computer-controlled pneumatic injection was implemented for the next 42 patients (group B). Compression using the balloon applied a uniform pressure and uniform stress on the rectal wall, and thus minimized artifacts due to non-uniform stress fields (Konofagou et al. 1996). The amplitude of the stress applied using the balloon was independent of the direction, whereas the use of the imaging probe as a compressor would require the use of a directional compensation for non-uniform applied stress and generates an angle-dependent decrease in the elastographic signal to noise ratio (SNRe) (Lorenz et al. 1999). The counterpart of the stable positioning and uniform compression is the loss in ease of use, as compared to the hand-held compression using the probe. The time interval between the pre- and post-compression frames was kept short to minimize decorrelation due to undesired motion of the patient. For this purpose, consecutive frames were acquired during a continuous compression. For patient group A, the acquisition was triggered by tracking the displacement of the balloon on the central A-line using a simple thresholding procedure. This scheme ensured that a known and uniform displacement was applied between digitized frames. A 0.25 mm displacement step was shown to be effective on phantoms and was chosen. The total tissue thickness between the balloon and the pubic bone that confines the anterior part of the prostate is usually in the range 35-45 mm, so the 0.25 mm displacement step corresponded to a 0.5-0.7% average strain. This level of applied strain has been shown to be in the optimal range for achieving maximum elastographic contrast-to-noise ratio (Varghese and Ophir 1997). The inter-frame time interval was approximately 1.5 s (i.e. 0.7 frame per second). For patient group B, all consecutive RF frames were digitized at the highest achievable frame rate (8 fps) to minimize inter-frame decorrelation. RF data were filtered by numerically removing frequency components outside the transducer bandwidth to minimize noise and interference. Strain was calculated from the gradient of the displacements, which were estimated using the position of the maximum of the cross-correlation function (Ophir et al. 1991) at zero lag with non-overlapping staggering and without stretching or lateral motion estimation. The normalized cross-correlation coefficient

A M B BLUR

SVED

IP

HP

Chapter 3. Prostate Cancer Detection

71

and the time delays were calculated using T=1 mm window length (1.3 µs) and ∆T=0.5 mm window shift (50% window overlap). A 5x5 (2.5 mm x 2°) median filter was applied to the displacements before the gradient was calculated. The filter increased the elastographic signal to noise ratio at the cost of resolution, whereas the overall size of the image was unchanged. The length of the filter kernel was matched to the window length to minimize the additional loss in resolution. At low frame rate (patient group A), decorrelation between frames prevented a systematic use of multi-compression, therefore the elastograms were calculated using a single compression. At higher frame rate (patient group B), multi-compression elastograms were generated using 12 consecutive RF frames acquired in 1.5s. In the future, occasional decorrelations may be removed by detecting and eliminating the respective frames from the calculation of the image. The patients were positioned on the table in lateral decubitus position, under general or spinal anesthesia. The patients were under the effects of anesthetics for the purpose of the therapy, and therefore were not able to hold their breath. A pre-compression was first applied to ensure that good contact was made between the balloon and the rectal wall. On two patients, the pre-compression was roughly estimated to be 15-20% of the initial antero-posterior (AP) dimension of the prostate, using the thickness of the prostate measured on the post-therapy MRI as a reference. Limited time was available to acquire elastograms. Therefore only three elastograms were acquired in parallel and approximately equally spaced imaging planes at the apex, mid-gland and base (Fig. 3.6). The imaging planes were chosen independently of any a-priori knowledge regarding the position of the tumor. These limitations imply that this study only reports incidental finding of the tumor, as opposed to a systematic scan that should be performed in a diagnostic application for effective tumor detection. In a future application, the acquisition could be performed over many planes to create a volumetric image. The influence of acquisition frame rate on correlation in vivo was investigated by comparing elastograms acquired at low frame rate (group A) and elastograms acquired at high frame rate (group B). A specific acquisition was also performed to calculate the loss in correlation as a function of the inter-frame delay. No compression was applied during this acquisition, so that changes in correlation with inter-frame delay can only be related to uncontrollable patient and/or organ motion, and to non-stationary RF signals (for example because of the presence of pulsatile blood flow). 100 consecutive RF frames were acquired at 8 fps (inter-frame delay ∆t =125 ms) during 12.5 s. Correlation was used as an indicator of the quality of the estimated time delays. The normalized cross-correlation coefficient was calculated using the zero-lag method without lateral motion compensation and without adaptive stretching, and averaged over the whole field of view. The average cross-correlation in the elastogram calculated using consecutive frames was compared with the average correlation obtained when skipping frames, thus simulating lower acquisition frame rates. In order to determine if adaptive stretching and lateral motion compensation were effective in vivo, the same elastograms were also calculated using a two-iterations adaptive stretching algorithm (Srinivasan and Kallel 2002). The first iteration calculated an elastogram using a 0.5% global stretching (Céspedes et al. 1993). This initial elastogram was smoothed and used as an initial guess to perform local stretching of the post-compression RF. In a second iteration, the final elastogram was calculated using the locally-stretched RF signals. In both iterations lateral motion estimation (Konofagou et al. 1998) was performed by searching for

Chapter 3. Prostate Cancer Detection

72

corresponding RF segments on five adjacent A-lines (corresponding to 2° aperture) to compensate for unavoidable lateral motion in vivo. Instead of interpolating intermediate RF lines, the position of the peak of the cross-correlation function was directly interpolated from the 2D cross-correlation function in the lateral direction. The normalized cross-correlation coefficient and the time delays were calculated using T=2 mm window length (2.6 µs) and ∆T=0.4 mm window shift (80% window overlap). A 5x5 (2 mm x 2°) median filter was applied to the displacements before the gradient was calculated. The repeatability of the acquisition was verified by an evaluation of the standard deviation of the average strain inside the prostate, measured from ten elastograms acquired at the same location. Five independent measurements of the standard deviation were obtained from different patients. Biopsies and sonograms were used as a reference to identify the tumors. The prostate and tumor contours were manually delineated from the sonogram. The average strain was measured inside the whole prostate, inside the tumor and outside the tumor. Areas of excessive decorrelation (correlation ρ<0.75) were excluded when measuring the average strain. 3.2.3 Results

3.2.3.1 Influence of frame rate on correlation and elastogram quality

(a) (b) Figure 3.8: Illustration of (a) a poor quality multi-compression elastogram and (b) an elastogram considered as acceptable in vivo. Areas marked in red correspond to correlation less than 0.50. Elastogram (b) shows the prostate (medium gray) with a central stiff (dark) urethral probe, whereas the prostate cannot be seen in elastogram (a). Eighty one multi-compression acquisitions were performed on 27 patients included in group A (frame rate 0.7 fps), each corresponding to 8 RF frames. Out of 567 (7x81) single-compression elastograms calculated using these data, only 5 elastograms (i.e. 0.9%) were deemed acceptable. The other elastograms were dominated by false peaks associated with low correlation (most samples had correlation less than 0.75), and the prostate could not be seen in the images. The elastograms were also calculated using lateral motion compensation and adaptive stretching in an attempt to determine the eventual benefits of these techniques for this specific data set. No improvement was observed in the elastograms. For the 42 patients included in group B (frame rate 8 fps), 126 multi-compression acquisitions were performed (12 RF frames each), and 1386 single-compression elastograms were calculated. The prostate was visible in almost all single-compression and multi-compression

-60 -40 -20 0 20 40 60

10 20 30 40 50 60 70 80

-60 -40 -20 0 20 40 60 10

20

30

40

50

60

70

80

Chapter 3. Prostate Cancer Detection

73

elastograms, and high correlation (> 0.95) was observed between consecutive RF frames in high sonographic SNR areas. 29% of the multi-compression elastograms were deemed of poor quality (similar to the elastogram presented in figure 3.8a) and were mostly associated with non-uniform and/or very small (< 0.1 mm) displacement of the balloon. The other elastograms were considered as acceptable (similar to the elastogram shown in figure 3.8b). Figure 3.9(a) shows the average correlation measured in vivo between frame number N and frame N-1, i.e. for consecutive frames (blue line), and between frame number N and frame 1, i.e. for increasing inter-frame delays (red line). Correlation was averaged over the whole field of view (from 9 to 56 mm from the transducer), including deep areas that have low sonographic signal to noise ratio. Between consecutive frames (inter-frame delay 125 ms), average correlation between 0.80 and 0.85 was consistently achieved. Correlation was below 0.80 for a couple of frames only. When time delays are cumulated in a multi-compression sequence, the frames with lower correlation could be identified and removed in order not to alter the sequence. The red line shows the average correlation of each frame with frame 1, i.e. the inter-frame delay associated with frame number N is (N-1)x125 ms. As the inter-frame delay increased, correlation rapidly decreased from 0.82 down to 0.65 after 1 s. Then correlation remained low, varying between 0.57 and 0.68. This result shows that the inter-frame delay has to be kept minimal in vivo in order to achieve high correlation between RF frames. (a) (b)

Figure 3.9: (a) Average correlation achieved in vivo between frame number N and frame N-1 (blue), and between frame number N and frame 1. Correlation was averaged over the whole field of view. Acquisition frame rate was 8 fps, so that each vertical line corresponds to a 1-s interval. Correlation was averaged over the whole field of view (9-56 mm from the transducer). (b) Average correlation vs. inter-frame delay in the high sonographic SNR area (9-24 mm from the transducer). Arrows show the average correlation expected for patient groups A and B. Figure 3.9(b) shows the correlation averaged in the area where sonographic SNRs is the highest (9-24 mm from the transducer). The error bars correspond to the standard deviation. The highest correlation (0.98±0.01) was achieved using 125-ms inter-frame delay, with low standard deviation. Then for increasing inter-frame delays the average correlation rapidly decreased down to 0.92 after approximately 1 second. The inter-frame delay for the elastograms acquired on patient group A was approximately 1.5 s. Figure 3.8 shows that after 1.5 s correlation has already reached its minimum (0.64 in the whole field of view, and 0.91±0.04 in high SNRs areas). At this level, less than 1% of the acquisitions provided acceptable elastograms. Much better elastograms were obtained for patient group B. In this group the inter-frame delay was 125 ms, corresponding to 0.83 average correlation in the whole image and to 0.98±0.01 in high SNRs areas.

Patient group B

Patient group A

Patient group B Patient group A

Chapter 3. Prostate Cancer Detection

74

3.2.3.2 Reproducibility

Figure 3.10: Series of 9 elastograms acquired at the same location in order to assess the reproducibility of the compression27. A urethral probe is visible as a very stiff central area (zero strain) associated with decorrelation noise in its sonographic shadow. The standard deviation in the average strain (measured over five series of 10 identical acquisitions) ranged from 4% to 15% (average 10%) of the average strain. Figure 3.10 shows one of these series. Although good visual correspondence in the overall strain pattern was observed, some discrepancies were visible when looking at the details. Some differences can probably be explained by respiratory motion, which resulted in imaging slightly different tissue planes. Differences are also expected because of the elastographic noise (limited elastographic SNRe). This result suggests that a single elastogram may be misleading (even if it is a multi-compression elastogram), but that multiple acquisitions would allow to differentiate between real structures (that would remain between consecutive frames, or that would appear and disappear on a cyclic basis) and random noise. Real-time acquisition and feedback would be able to provide such information. The reproducibility of the elastograms was deemed satisfactory for the purpose of detecting relatively large lesions, but care has to be taken in their interpretation when looking for small lesions.

27 Experiment Ref. 2002-1127-PD

Urethral probe

Chapter 3. Prostate Cancer Detection

75

3.2.3.3 Lateral motion correction and adaptive stretching

As shown in Fig.3.11, no significant difference was observed between multi-compression elastograms processed without (top row) and with (bottom row) lateral motion correction and adaptive stretching. The first column shows a cancer (dark blue low strain area), the second column shows a HIFU lesion that was generated on the right side of the prostate (low strain, medium blue on the left of the images), and the third column shows a HIFU lesion that covers the whole prostate (low strain, medium blue area). A black correlation mask hides areas where correlation ≤ 0.75. Pseudo-colors were used instead of grayscale in order to increase the dynamic range of the images. Pseudo-colors were useful to enhance differences in the images for the comparison between the processing schemes. However they could be misleading if used for diagnostic purpose in vivo because of the artificial segmentation they induce. Because of the high frame rate, the inter-frame strain and the lateral motion were small. Adaptive stretching was shown to improve elastograms for high strains only (Alam et al. 1998) and inter-frame lateral motion was too small to significantly degrade the elastograms. It is also likely that the small improvement expected with lateral motion correction and adaptive stretching is hidden by the 5x5 samples median filter applied to the axial displacements. Without the median filter, both series of elastograms were too noisy for interpretation. Figure 3.11: Comparison between elastograms generated without (top row) and with (bottom row) lateral motion correction and adaptive stretching for three different images (columns). A black correlation mask hide areas of decorrelation noise, mostly associated with low sonographic signal-to-noise ratio. Pseudo-color display was chosen to enhance the visual dynamic range. Note the similarity between the two rows. 3.2.3.4 Prostate cancer detection

In spite of the limited number of imaging planes, prostate cancer confirmed by biopsies and sonograms was found in the elastograms in 5 cases28. Two are presented below to show the feasibility of using elastography to detect prostate cancers. The images follow the conventional orientation used in radiology (i.e. the left of the image is the right of the patient, and vice-versa). The applied pre-compression was approximately 15-20%.

28 Experiments Ref : 2002-0202-MR, 2002-0301-JBF, 2002-0723-AF, 2002-0912-HK, 2002-1106-FP

-30 -20 -10 0 10 20 3020

30

40

50

60

-30 -20 -10 0 10 20 3020

30

40

50

60

-30 -20 -10 0 10 20 3020

30

40

50

60-30 -20 -10 0 10 20 30

20

30

40

50

60

-30 -20 -10 0 10 20 3020

30

40

50

60-30 -20 -10 0 10 20 30

20

30

40

50

60

Chapter 3. Prostate Cancer Detection

76

(a) (b) Figure 3.12: (a) Sonogram and (b) elastogram of the prostate (black outline) and the cancer (white outline) in patient X. The images are 60x45 mm in size. In the elastogram, the greyscale corresponds to 0 (black) to 1.5% (white) strain and areas where correlation is less than 0.75 are hidden by a medium grey mask. Figure 3.12 shows the sonogram and the corresponding elastogram acquired for a patient diagnosed with stage T2 carcinoma. This patient was included in group A (i.e. low frame rate acquisition). The prostate (black outline) and the tumor (white outline) were manually delineated from the sonogram. The tumor was visible on the sonogram as a hypo-echoic area, and on the elastogram as a stiff area in the left posterior part (bottom right of image) of the prostate. Biopsies confirmed the presence of a carcinoma in this region. The black area at the bottom of the sonogram is the liquid inside the balloon. The prostate was seen as a hypo-echoic area that contains two hyper-echoic calcifications that give rise to characteristic distal shadows. The two calcifications were clearly depicted as low strain areas surrounded by strain concentrations, as expected from experiments with phantoms containing a stiff inclusion. The sonographic shadows behind the two calcifications generated an area of corresponding decorrelation noise in the elastogram. Between the two calcifications, the urethra was visible as a central area of high strains. A third stiff nodule, neither visible on the sonogram nor found by biopsies, was seen on the elastogram between the calcifications. On the elastogram, a white rim of high strains outlined the edge of the prostate capsule (which is a thin soft layer around the prostate, ~1 mm thick). It corresponds to a soft layer of fatty peri-prostatic tissues. The average strain was 0.19% inside the tumor and 0.72% outside (strain contrast ratio of 3.8). A stiff area was also seen on the right of the patient (left of the image) and might possibly show a tumor not visible in B-mode. Decorrelation areas were observed inside the balloon, due to the low sonographic signal-to-noise ratio (SNRs) in the liquid, and around the prostate. Decorrelation on the left side was associated with low SNRs, whereas possible slip conditions may be responsible for the decorrelation along the right side of the prostate.

(a) (b) Figure 3.13: (a) Sonogram and (b) elastogram of a prostate cancer (white outline) (Patient Y, 60x45 mm images,

greyscale 0-1.5% strain)

Chapter 3. Prostate Cancer Detection

77

Figure 3.13 shows the corresponding sonograms and elastograms for another patient (group A), who was diagnosed with stage T2 carcinoma. The tumor (white outline) had similar location and shape in both images. Biopsies confirmed the location of the carcinoma. The strain contrast ratio between the tumor (0.08% strain) and the other tissues (0.76% strain) was 9.5. The elastogram also showed two unidentified stiff areas, and noisy strain estimates that were mostly associated with low SNRs in sonographic shadows. 3.2.4 Discussion

The comparison between the elastograms in patient groups A and B showed that high frame rate was necessary but not sufficient to produce high-quality elastograms in vivo. The loss in correlation with increasing inter-frame delay was probably due to uncontrollable patient and/or organ motion. The highest average correlation was achieved for 125 ms inter-frame delay (8 fps). It is likely that higher frame rates would provide better correlation. As a consequence, the quality of the elastograms would be improved, provided that the compression rate is matched to the acquisition frame rate. Indeed an excessive compression rate is likely to result in signal decorrelation (because of the inter-frame strain and of lateral and elevational motion), whereas an insufficient compression rate is likely to result in low SNRe (total applied strain too small). In such a case, one could be expected to increase the total applied strain (and SNRe) using longer compression/acquisition sequence with a small strain rate, but it is likely that resolution would then degrade because tracking the same tissue sample would be made impossible by the increased lateral and/or elevational motion. The loss in correlation with increasing inter-frame delay can be explained by uncontrollable patient and/or organ motion. It is likely that an optimum compression rate exists between these two extremes, and will be subject to future investigations. Undesired displacements of the prostate due to the respiration was observed as lateral motion on the longitudinal B-scans. This observation suggests that respiratory motion could be measured, and therefore compensated for, by acquiring elastograms in longitudinal scans and estimating lateral displacements (Konofagou et al. 1998). Alternatively, this motion itself might be used for the generation of elastograms, instead of an externally applied compression. However high frame rate was not sufficient to acquire high-quality elastograms. At 8 fps, 29% of the elastograms were not acceptable because the prostate contours and/or structure could not be seen. Most of these elastograms were associated either with very small displacement of the balloon (<0.1 mm) and the low strain was hidden by jitter noise, or with significant non-uniform displacement of the balloon. As opposed to the case where compression is applied using a rigid plate, the balloon exerts a uniform stress on the rectal wall but not a uniform displacement. Displacements are expected to be larger near soft tissues and smaller near stiff tissues. However only slightly non-uniform displacements are expected and were observed in vitro, whereas significant displacement asymmetry was occasionally observed in vivo. It might be explained by the contact that exists between the balloon and the therapy probe, and/or by non-uniform boundary conditions (for example rectum stenosis due to previous radiotherapy or surgery). In order to achieve better control on the direction of the compression, a specific balloon could be designed to inflate mainly in the direction of the prostate, while preventing inflation and elongation in other directions. Higher compression rates would be achieved (for an identical fluid flow, prostate compression would be increased)

Chapter 3. Prostate Cancer Detection

78

and risks for the patient would be reduced because the total amount of liquid required to induce strain would be small. A major limitation encountered in the analysis of the elastograms was the lack of correlation between biopsies and sonography. Biopsies are the “gold standard” to detect the presence of malignant tissues, but they do not provide information about the size and the shape of the tumor, so this information was sought from sonograms. However most tumors were not visible in the sonograms (false negatives). The number of reliable cases where cancer could be confirmed by both biopsies and sonograms was therefore considerably restricted. Confirmation by sonography only was not used because many hypo-echoic areas found in sonograms were associated with negative biopsies (false positives). Confirmation by biopsies only would be possible but was not used because of inaccurate registration (the prostate is divided into 6 large volumes corresponding to the apex/mid-gland/base and to left/right and biopsies are sampled in these regions) and because it would not provide information about the size and shape of the tumor. Histo-pathological confirmation would have been able to provide the needed information, but due to HIFU being a minimally invasive therapy the prostate was not removed for pathology examination after treatment. The limited number of imaging planes (3 per prostate) was the second major limitation. Because of this poor sampling, chances of imaging the plane where the cancer was were few. Many cancers may have been missed because of this reason. Moreover, the analysis of elastograms acquired in vitro (chapter 3.1.4) showed that assessing the presence of a low-strain area in multiple adjacent elastograms was an important criteria to determine if the lesion was to be considered as cancer or not. Imaging planes were too few in vivo to use such a criteria. As a consequence, rapid data acquisition is required to acquire more imaging planes. In a full clinical implementation, fast data processing is also necessary to allow the radiologist to identify noisy elastograms immediately and to acquire them again, thus avoiding “gaps” in the imaging sequence. The strain contrast ratio measured between cancer and normal tissues was high. This is probably not representative of the strain contrast ratio of the majority of cancers, for many reasons. First, strain contrast was measured in two cases only, which is not statistically significant. Second, the detectability of a lesion depends on its size and on its contrast, and with no prior training in prostate elastography in vivo only large and high-contrast tumors were expected to be found. Third, strain contrast may be improved by optimizing the amount of pre-compression, i.e. using the non-linear properties of tissues, as prostate tumors were reported to behave non-linearly (Krouskop et al. 1998). Setting strain contrast thresholds in vivo is also made difficult by the unclear dependence of strain on depth, which was theoretically expected using a cylindrical compression model (Souchon and Soualmi 2002) but not observed in vivo. Ponnekanti et al. (1992) showed a reduction in strain decay inside a material compressed between two parallel plates. Similarly, the confinement of the anterior part of the prostate by the pubic bone is likely to reduce the effect of strain decay. A major cause of decorrelation in the elastograms was the low sonographic signal-to-noise ratio (SNRs) in parts of the prostate, due either to hypo-echoic tissues, ultrasonic attenuation, or shadow areas. Figure 3.14 shows the average normalised cross-correlation coefficient plotted as a function of SNRs for every time delay estimation performed during the calculation of the elastogram presented in Fig. 3.12. The average correlation decreased from 0.98 at 50 dB SNRs down to 0.9 at 15 dB SNRs, and a steeper decrease was observed for SNRs below 15 dB. (Céspedes et al. 1997). The standard deviation increased with decreasing

Chapter 3. Prostate Cancer Detection

79

SNRs. These results explain a large part of the decorrelation noise present in the elastograms in vivo. It can be seen that the correlation coefficient, and therefore the quality of the elastogram, was strongly dependent on the sonographic SNRs. The use of a high transmit gain on the ultrasound scanner could improve the quality of the elastograms.

Figure 3.14: Normalised cross-correlation coefficient vs. sonographic SNRs for the elastogram shown in

Fig.3.12 3.2.5 Conclusion

Prostate elastograms were acquired in vivo on 69 patients. In spite of the limited number of imaging planes (3 per patient), prostate cancer confirmed by biopsies and sonograms was found in the elastograms in 5 cases. It was seen as a stiff area in the peripheral zone of the prostate. Using a pre-compression of 15-20%, cancer presented a high strain contrast ratio (3.8 to 9.5) with the surrounding tissues. High acquisition frame rate was shown to be necessary but not sufficient to acquire clinically useful elastograms. At 8 fps, 71% of the elastograms clearly showed the prostate. The other images were associated with non-uniform compression resulting in lateral and/or elevational motion. It is expected that a better control on the direction of the compression will improve these results. A major limitation of our system was its limited bandwidth (~40% at 5 MHz center frequency). Indeed it was shown that all performance characteristics of elastographic systems increase with the absolute bandwidth B (Srinivasan et al. 2003): axial resolution (Ra α 1/B), signal-to-noise ratio (SNRe α B3/2) and contrast-to-noise ratio (CNRe α B3). In clinical practice today, prostate scanning is performed at 7-9 MHz with 60%-80% bandwidth transducers. As a consequence, the elastographic acquisition system needs to be improved in terms of bandwidth, frame rate, compression directivity, and real-time feedback, and more imaging planes need to be acquired before a conclusion can be drawn regarding the sensitivity, the specificity and the positive predictive value of elastography. Yet the results of this study confirmed the feasibility of elastographic detection of prostate cancer in vivo.

Chapter 4. HIFU Lesions

80

4 Application to the Visualization of HIFU Lesions

Ce dernier chapitre est consacré à la visualisation par élastographie des lésions thermiques générées par ultrasons focalisés de haute intensité (HIFU). Il se présente sous la forme de deux études traitant tout d’abord de la faisabilité in vivo pour de grosses lésions en cours de et immédiatement après traitement HIFU, puis de la faisabilité de visualiser la formation d’une petite lésion HIFU élémentaire in vitro pendant les quelques secondes que dure le tir. La première partie de ce chapitre décrit l’étude clinique mise en place sur 69 patients recevant un traitement HIFU pour un cancer localisé de la prostate. Chez ces patients, des élastogrammes ont été acquis avec le système décrit précédemment alors que la moitié de la prostate était traitée, puis immédiatement après que la totalité de la glande soit traitée. Les lésions visibles dans les élastogrammes ont été comparées aux dimensions des zones cibles choisies et, pour quelques patients, aux IRM acquis 48 h après HIFU. Les lésions HIFU étaient visibles in vivo du fait du contraste de déformation existant entre tissus nécrosés rigides et tissus sains plus mous. Le contraste de déformations était de l’ordre de 1.6 à 3.2. La taille des lésions correspondait avec la dimension de la zone cible et a été confirmée par les images IRM. La prostate était globalement durcie par le traitement, comme l’a montré la déformation moyenne de la prostate après HIFU qui a baissé de 54±26% par rapport à sa valeur avant traitement. Il a été possible de voir les lésions HIFU dans les élastogrammes immédiatement après leur formation, ce qui rend envisageable le guidage du traitement sur des éventuelles zones insuffisamment traitées. La seconde partie de ce chapitre est dédié à une nouvelle application de l’élastographie pour la visualisation des lésions thermiques, que nous avons baptisée élastographie passive. En effet, la contrainte mécanique exercée par le ballon est supprimée, et les seuls effets thermiques des HIFU permettent de générer un élastogramme. Ce procédé dérive des travaux sur la mesure ultrasonore de température, et a pour originalité d’étendre son champ d’application aux températures élevées pour lesquelles l’expansion thermique et la coagulation des tissus peuvent apparaître. Il permet l’acquisition d’images pendant la formation de la lésion. La déformation mesurée est composée d’une part d’une déformation (dilatation thermique) réelle des tissus, et d’autre part d’une déformation du signal RF qui est la résultante de la variation de la vitesse du son avec la température. La théorie permettant de lier la déformation mesurée à partir des signaux RF à la déformation réelle des tissus et à la variation de vitesse du son est d’abord explicitée. Des élastogrammes sont alors simulés pour une lésion HIFU élémentaire. La technique a été testée in vitro sur du foie. Expérimentalement, la zone de coagulation est apparue comme une zone d’expansion apparente, dont les contours correspondaient bien avec la taille des lésions prédites par simulation et avec l’examen macroscopique des échantillons, bordée par une zone de compression apparente. Il a été possible de suivre la progression du front de coagulation sur l’image. Le fort contraste existant dans les images expérimentales et leur comparaison avec les images simulées suggèrent que la coagulation induit une expansion significative des tissus, liée à un changement de structure plutôt qu’à une dilatation thermoélastique.

Chapter 4. HIFU Lesions

81

4.1 HIFU therapy follow-up using elastography in vivo29

4.1.1 Objectives

After HIFU therapy of prostate cancer, it is of primary importance to ensure that no residual area has been left untreated in the target zone. For this purpose, the physician needs an image of the HIFU lesion. Gadolinium-enhanced MRI is the gold standard to show HIFU lesions (Hynynen et al. 1996, Rouvière et al. 2001), but it does not show eventual residual cancer foci. Moreover its availability is restricted and its cost may be prohibitive, so an ultrasound-based imaging method would be desirable. Sedelaar et al. (1999) showed that the lesions are visible as devascularized areas using Power Doppler, but their volume was underestimated by 25-35% when compared with histology. Elastography was shown to be able to show HIFU lesions in vitro (Kallel and Stafford 1999, Righetti et al. 1999). Righetti et al. found that a lesion/background strain contrast threshold of –2.5 to –3.5 dB (0.67-0.75) was able to completely define the entire zone of tissue damage. In this study, we investigated the feasibility of using elastography in vivo to image HIFU lesions during and immediately after therapy. Comparison of the lesion position and area between elastograms and MRI was used to assess the accuracy of the method. 4.1.2 Material and Method

The acquisition system used for this study was identical to that used for cancer detection. It was based on a transrectal imaging probe covered with a balloon and attached to a therapy probe (Fig. 3.6). Multi-compression RF sequences were acquired at 0.7 fps on 27 patients (group A) and at 8 fps on 42 patients (group B). The central frequency of the imaging probe was 5.5 MHz. Continuous compression of the prostate was applied by inflation of a balloon that covered the HIFU and imaging probes. Inter-frame displacements were estimated using cross-correlation at zero lag with 1 mm windows and 0.5 mm window shift. Strain was estimated from the gradient of the cumulated displacements using non-overlapping windows after 5x5 median filtering of the displacements. The acquisition system, data processing schemes, patient groups (A and B) and imaging planes (3 at the base, mid-gland and apex) were identical to those used for prostate cancer detection in vivo. More imaging planes would be required to fully depict the lesion volume, but for this feasibility study only three were acquired because limited time was available in the operating room for data acquisition. Elastograms were acquired at identical locations (1) immediately before HIFU, (2) during therapy once a lateral lobe was treated, (3) and within a few minutes after HIFU. In two patients (group A), the size and position of the HIFU lesion was estimated using MRI. Fat-saturated gadolinium-enhanced T1-weighted turbo spin echo (TSE) images were acquired to visualize the HIFU lesions, as described in Rouvière et al. (2001): all MR images were obtained on a 1.5-T MR unit (Siemens Symphony, Erlangen, Germany) using a pelvic phased-array coil, after intravenous injection of 0.1 mmol/kg (0.2 ml/kg) of gadoterate

29 Publications based on, or including part of, this section: Souchon Chapelon et al. 2001, Ophir et al. 2002, Souchon Gelet et al. 2002, Souchon Curiel et al. 2003, Souchon Detti et al. 2003a and 2003b, Souchon Rouvière et al. 2003

Chapter 4. HIFU Lesions

82

meglumine (Guerbet Dotarem, Roissy, France). The following parameters were used : TR 1.093 ms, effective TE 12 ms, slice thickness 4 mm, interslice gap 0.4 mm, 3 excitations, echo train of 3, 30% phase over-sampling and frequency-selective fat saturation. This technique was shown to provide accurate visualization of HIFU induced lesions, but not to predict the existence of residual viable cancer. MR scans were acquired 48 h after therapy, whereas elastograms were acquired no later than 5 minutes after the end of the treatment. The distance from the transverse imaging plane to the apex served as a reference for registration between MR images and elastograms. The prostate and the lesion contours were manually delineated from MR images and used to calculate the relative lesion area, defined as the ratio of the surface of the HIFU lesion surface to the surface of the prostate. For these two patients, HIFU lesions visible in the elastograms were characterized by their strain contrast ratio and their relative lesion size. The prostate contour was manually delineated from the sonograms, whereas the HIFU lesion was manually delineated from the elastograms. The lesion/background strain contrast ratio (Cε) was measured when normal tissue was still visible in the elastogram. It was defined as the ratio of the average strain inside the lesion to the average strain inside remaining normal tissues. The relative area of the HIFU lesion in the prostate was measured from the elastogram and was compared to the relative lesion area measured from MRI. After the whole prostate is treated, no normal tissue should remain and no lesion/background strain contrast is expected (i.e. Cε cannot be defined). In this case, the strain ratio (sr) was defined as the ratio between the average strain measured in the whole prostate after HIFU to the average strain in the whole prostate before HIFU. Such a factor quantifies the variation in average strain before and after HIFU. Although this strain ratio is not strictly a contrast ratio because it is calculated from two different elastograms, its value is consistent and comparable with the strain contrast ratio Cε defined above. Because of the 10% inter-frame variability in average strain (chapter 3.2.3), the uncertainty in measurement of the strain ratio is 20%. 4.1.3 Results

As already discussed in chapter 3.2, approximately 99% of the acquisitions failed for patients in group A because of an insufficient acquisition frame rate (0.7 fps). In these elastograms the prostate could not be seen. Only two patients (X,Y) had acceptable elastograms and were selected to have post-HIFU MRI. It is interesting to note that these two patients were part of the few patients in group A who had acceptable prostate cancer elastograms before HIFU. Prostate cancer elastograms for these two patients were shown in the previous chapter in figures 3.12 and 3.13. Their HIFU lesion elastograms are presented in figures 4.1-4.3. In Fig. 4.1 the corresponding sonogram and elastogram are shown for patient Y during HIFU therapy. The target area (arrows) was 18 mm high starting from the rectal wall, and 25 mm wide (the entire right side of the gland). On the sonogram, a large part of the prostate has a hyper-echoic appearance because of the presence of cavitation gas bubbles created during the application of HIFU. These bubbles disappear after a few minutes and are not representative of the position and size of the HIFU lesion. On the elastogram, the HIFU lesion (white outline) is shown as a stiff (dark) area that extends laterally on 35 mm from the right edge of the prostate to the central part, and penetrates up to 20 mm from the rectal wall towards the anterior face of the prostate. The HIFU lesion exceeds the target width, but outside the target area its penetration depth is only 10 mm. The left end side and the anterior side of the prostate

Chapter 4. HIFU Lesions

83

(top of the image) are not treated and are depicted as soft tissues (bright). The large size and the a-priori knowledge of the location of the HIFU lesion make it easier to detect than tumours. Measurements inside the prostate show a 0.32% average strain in the HIFU lesion and 1.01% outside the lesion (strain contrast ratio of 3.2).

(a) (b)

Figure 4.1: Corresponding (a) sonogram and (b) elastogram of a HIFU lesion (white outline) in the right lobe of the prostate (black outline) for patient Y. The lesion was not visible in the sonogram. Arrows show the target area. Image dimension is 60x45 mm, the grayscale corresponds to 0-1.5% strain, a medium gray correlation mask was applied where correlation < 0.75.

(a) (b)

Figure 4.2: (a) MRI and (b) elastogram of a HIFU lesion (white outline) for patient X. The shape and size of the lesion are similar in both images. Two calcifications can be seen inside the HIFU lesion (arrows). Significant pre-compression can be observed in the elastogram when comparing the shape of the prostate (black outline) in both images. Image dimensions are 60x45 mm, grayscale 0-1.5% strain, medium gray mask for correlation < 0.75. Figure 4.2 shows the corresponding MRI and elastogram for patient X after HIFU treatment. The HIFU lesion is expected to spread laterally over the whole width of the prostate, and to penetrate 18-20 mm deep into the prostate starting from the boundary between the rectum and the prostate. The rectal wall and deep prostatic tissues should be undamaged. The HIFU lesion (white outline) is depicted on the elastogram by a stiff area that extends 18-22 mm deep starting from the rectal wall. Deeper tissues show higher strains as well as decorrelation noise due to low SNRs in this area. The average strain is 0.36% inside the HIFU lesion and 0.56% in untreated tissues (strain contrast ratio of 1.6). The calcifications observed before HIFU therapy (Fig. 3.12) were also seen inside the HIFU lesion (two arrows). The MR image shows the HIFU lesion as a hypo-intense zone (white outline) surrounded by a hyper-intense peripheral rim of edema that appeared over 48 h following HIFU therapy. MRI resolution and SNR are similar to those reported by Rouvière et al. (2001). Sharp contours are not seen in the MRI probably because of the inflammatory aspect of the edema. The MR image confirms the slightly asymmetric shape of the HIFU lesion. Both MRI and elastogram show that the rectum and the anterior part (top of image) of the prostate were not damaged. The surface of the HIFU lesion represents 57% of the surface of the prostate on the elastogram, and 44% on the

Chapter 4. HIFU Lesions

84

MRI. The discrepancy may arise from the changes in volume that occurred in the prostate during the 48 h elapsed between elastogram and MRI acquisition (Rouvière et al. 2001). The direction of the imaging planes in elastograms and in MRI may also be slightly different.

(a) (b)

Figure 4.3: (a) MRI and (b) elastogram of a HIFU lesion (white outline) for patient Y. This elastogram was acquired once both lobes were treated, and can be compared to Fig. 4.1 showing only one lobe treated for the same patient. The extension of the HIFU lesion from the rectal wall to the urethra (bright central area) is clearly visible in the elastogram and is confirmed by MRI (60x45 mm image, 0 to 1.5% strain) Figure 4.3 presents the elastogram and corresponding MRI for patient Y after HIFU treatment. The HIFU lesion is expected to have similar characteristics as for patient X. The lesion (white outline) is depicted on the elastogram by a low strain area that extends 15-18 mm deep starting from the rectal wall, leaving an untreated area in deeper tissues. The average strain is 0.42% inside the HIFU lesion and 0.77% in untreated tissues (strain contrast ratio of 1.8). The urethra is visible as a central spot of high strains. The urethra is visible as a circular central black area on the MR image because of the presence of a trans-urethral catheter (the catheter was not present when elastograms were acquired). The MR image confirms that the HIFU lesion extends over the full width of the prostate, and is bounded by the rectal wall and the plane of the urethra. The surface of the HIFU lesion represents 51% of the surface of the prostate on the elastogram, and 47% on the MRI. The rectum and the anterior part of the prostate appear undamaged in both the MRI and the elastogram. 4.1.3.1 Variation of the average strain after HIFU therapy

In patient group A the elastograms were too noisy to allow the average strain to be measured. In patient group B, the average strain ranged from 0.23% to 1.54% before therapy, and from 0.06% to 0.98% after therapy. Because of the large variations between patients, the average strain was not considered to be a relevant parameter to assess the efficacy of the treatment. The strain ratio sr (average strain after HIFU/average strain before HIFU) measured on 29 patients was 0.54 ± 0.26. The inverse of the strain ratio is equivalent to a 2.5 ± 1.7 strain contrast ratio between normal and coagulated tissues. It is consistent with the strain contrast ratio (1.6 to 3.2) measured during therapy between the HIFU lesion and undamaged tissues for patients X and Y. 4.1.4 Discussion

Large HIFU lesions were clearly seen in the elastograms as stiff areas. Their position and size were confirmed by MRI in two cases. This demonstrates the feasibility of using elastography to assess the extent of the treated region immediately after HIFU therapy was performed. It was also possible to briefly interrupt the therapy to acquire elastograms that showed the HIFU lesion, thus showing that elastographic guidance would be possible to detect and target

Chapter 4. HIFU Lesions

85

10 12 14 16 18 2010

15

20

25

30

35

Vol. MRI (cm3)

Vol. Elasto (c

m3)

Equality

untreated areas. Real-time monitoring using elastography while the tissues are being heated could be possible and would allow accurate targeting, but may be challenging because both stiffness and speed of sound depend on temperature. The strain contrast ratio measured in vivo between normal tissues and HIFU lesions in the prostate in vivo was 1.6 to 3.2, which is larger than and consistent with the strain contrast threshold (1.3 to 1.5) used by Righetti et al. (1999) to differentiate HIFU lesions from normal tissues in canine liver in vitro. A difference may exist because of the different nature of liver and prostate tissues and of the use of different boundary conditions, so comparison with in vitro results should be made carefully. The strain contrast ratio is also likely to vary during prostate cooling after HIFU application, because of the temperature dependence of the stiffness of tissues (Wu et al. 2001). It may also depend on the pre-compression because of the eventual nonlinear properties of tissues. Once HIFU lesions were created no significant contrast was observed in the elastograms between the coagulated cancer and coagulated normal tissues, i.e. the tumors were no longer visible inside the HIFU lesion. The eventual presence of residual cancer could not be detected in the elastograms. The average strain in the prostate was shown to decrease by 54±26% after treatment on 29 patients. A small decrease after treatment may indicate incomplete treatment and a risk of local recurrence. The correlation between the relative value of the average strain after treatment and the treatment outcome needs to be investigated to determine if this value can be used as an indicator of treatment efficiency. It is interesting to note that the presence of stable hyper-echoic gas bubbles induced by HIFU therapy enhanced the sonographic SNRs and was therefore beneficial to the quality of the elastograms acquired after HIFU in the proximal area. Unstable cavitation bubbles are likely to induce significant decorrelation if RF data were acquired during heating. In certain elastograms the distal margins of the HIFU lesions were not visible because of the diminished SNRs behind the proximal bubbles. In these cases, the corresponding sonograms showed a sharp transition between a proximal hyper-echoic area and a distal hypo-echoic zone, and the distal edge of the hyper-echoic area often corresponded with the expected depth of the lesion. Figure 4.4: HIFU lesion volume measured from elastogram (vertical axis) vs. volume measured from MRI (horizontal axis). The dotted line corresponds to perfect equality. These results are too preliminary to draw any conclusion. Visualization of HIFU lesions in this study was facilitated by the large size of the lesions, but the size of the smallest detectable lesion is not known yet. On-going work is being conducted to determine if statistically significant correlation is found between the position, shape and volume of the HIFU lesions measured from the elastograms and from MRI (Souchon and Curiel 2003). Comparison between the volume of the HIFU lesion measured from

Chapter 4. HIFU Lesions

86

elastograms and from MRI have been analyzed for 7 patients only so far (fig. 4.4), and are therefore too preliminary to draw any conclusion. 4.1.5 Conclusion

Elastograms were acquired in three imaging planes (apical, middle, basal) on a total of 69 patients undergoing HIFU therapy of localized prostate cancer. In the presence of untreated tissues, HIFU lesions were visible in the elastograms during and immediately after therapy as low strain areas, even in the presence of cavitation-induced gas bubbles as seen on the sonograms. The hyper-echoic appearance of the bubbles increased the sonographic SNRs and was therefore beneficial for the quality of the elastograms. The size and position of the HIFU lesions visually determined from the elastograms matched the expected lesion, and were confirmed by MRI on two patients. On these patients, a strain contrast ratio between 1.6 and 3.2 was measured between the HIFU lesion and normal tissues. In 29 patients the whole prostate was treated and the average strain inside the prostate was shown to decrease by 50±10% when compared to the average strain measured at the same location before treatment. Variations in this parameter may be related to the severity of the lesion.

4.2 Monitoring of HIFU lesion formation by passive elastography in vitro30

4.2.1 Objectives

In the previous section elastography was shown to be capable of visualizing large HIFU lesions once they have been created. However monitoring HIFU lesions during their formation would be of considerable interest: this would ensure that coagulation effectively occurs, that targeting is accurate, and that surrounding tissues are spared. If provided in real time, this information could be used as feedback to adapt treatment parameters such as shot duration, acoustic intensity, or targeting. Hynynen et al. (1996) showed that MRI is able to measure temperature changes during HIFU application. Weidensteiner et al. (2003) were able to monitor temperature variations using MRI during thermal ablation of rabbit liver in vivo, with a temporal resolution of 3 s per slice, a 1.4x1.9x5 mm3 resolution, and a 1-3 °C standard deviation. An ultrasound-based method would be able to minimize the cost of the procedure and to improve temporal resolution. Recent work was reported on ultrasonic temperature estimation: temperature-related variations in the speed of sound induce a variation in the time of flight of ultrasonic signals, therefore a time delay in the backscattered RF signal received by a transducer. The variation in speed of sound between two RF signals acquired before and after heating can thus be estimated either from the shift of the central frequency (due to a “compression” of the RF signal in the time domain, similar to a frequency modulation), or from time delay estimates (as used in standard elastography). The variation in temperature can be estimated from the 30 Publications based on, or including part of, this section: Bouchoux Souchon et al. 2003, Souchon Bouchoux et al. 2003 (submitted).

Chapter 4. HIFU Lesions

87

variation in speed of sound using calibration tables (Maass-Moreno et al. 1996a). This method assumes (1) that thermal expansion is negligible, (2) that a calibration table is available for the tissues being investigated, and (3) that the initial temperature and initial speed of sound are known. Validation in vitro for temperature variations within the temperature range 20-40°C was shown by Maass-Moreno et al. (1996b). In various tissue types, as temperature increases above 40°C speed of sound was shown to reach a plateau with a maximum around 65°C, and decreases for higher temperatures (Bamber et al. 1979, Lu et al. 1996, Techavipoo et al. 2002, Worthington et al. 2002). Thus the method has low sensitivity around the maximum, and the correspondence between speed of sound and temperature is no longer unique after the maximum was reached. Surprisingly, no significant variation in speed of sound was observed during and after the coagulation of canine tissues (Techivapoo et al. 2002), whereas significant stiffening was observed (Kallel and Stafford 1999, Righetti et al. 1999, Shi et al. 1999, Wu et al. 2001). To our knowledge, no data are available in the literature regarding thermal expansion of biological tissues within the temperature range achieved during HIFU (37-90°C), but it is likely to be significant in such a large temperature range. As a reference, the volume of a constant mass of pure water increases by approximately 2.9% when heated from 37 to 90°C. The normalized cross-correlation function is a quantitative measurement of the similarity between two signals: high correlation (~1.0) is expected for highly similar RF signals, and decreasing correlation is expected with increasing differences between the RF signals. The onset of coagulation alters the tissue structure and is therefore expected to deform the backscattered signal from the treated region. As a consequence, the coagulation front is expected to be detectable as a decrease in the normalized cross-correlation coefficient. The coagulation boundary was shown to be detectable during laser heating of canine liver samples in vitro using the correlation between RF lines acquired at 5 s interval (Sun et al. 1999, Sun et al. 2001). Both temperature-related variations in speed of sound and thermal expansion are expected to induce an apparent strain in elastograms calculated between RF frames acquired before and after heating. Although it may not be possible to differentiate between speed of sound and thermal expansion, we hypothesized that the apparent strain elastogram and the correlation image (which is obtained during the calculation of an elastogram) obtained from RF frames acquired at high frame rate during HIFU application may provide information about the position and size of the HIFU lesion. This technique is referred to as passive elastography in this thesis because no external mechanical compression is applied during the acquisition. The theory underlying passive elastography is detailed below. Simulations of the apparent strain were performed using a finite elements model (FEM). Then experiments were carried out in vitro on porcine and bovine liver to compare apparent strain elastograms and correlation images with the HIFU lesion determined from gross pathology (Bouchoux et al. 2003a, Bouchoux 2003b). 4.2.2 Theory

The combined effects of variations in speed of sound and tissue thermal expansion on backscattered RF signals are detailed in this section.

Chapter 4. HIFU Lesions

88

Let P be a scatterer located at a distance z along the axis of a transducer. Let cTo(l) be the speed of sound at distance l from the transducer at temperature To. For an ultrasonic pulse to reach point P and to return to the transducer, the return-trip time of flight is:

∫=z

ToTo lc

dlzt0 )(

2)( Eq.4-1

After heating by ∆T = T-To, the temperature-dependent speed of sound changes. The variation in speed of sound δc(l) may also be related to structural changes in the tissues after heating, for example due to coagulation. Let δc(l) denote the temperature variation of speed of sound at location l:

δc(l) = cT(l) – cTo(l) Eq.4-2 Heat also induces thermal expansion of the tissues: the tissues deform and undergo internal strains. Considering that lTo is the initial length of a small portion of the medium, and δl is the variation in length of this portion, the local strain component in the direction of the ultrasonic beam is:

ε(l) = δl / lTo Eq.4-3 After heating, the return-trip time of flight of the ultrasonic pulse becomes:

( )∫ +

+=

z

ToT dl

lclclzt

0 )()()(12)(

δε Eq.4-4

The time delay induced by heating from temperature To to T at point P is :

)()()( ztztz ToT −=τ Eq.4-5 Assuming that the variation δc(l) in speed of sound is very small as compared to the initial speed of sound cTo(l), the following approximation can be written:

⎟⎟⎠

⎞⎜⎜⎝

⎛−≈

+ )()(1.

)(1

)()(1

lclc

lclclc ToToTo

δδ

Eq.4-6

The expression for the time delay can be simplified:

∫ ⎟⎟⎠

⎞⎜⎜⎝

⎛−≈

z

ToTo

dllclcl

lcz

0 )()()(

)(12)( δετ Eq.4-7

The apparent strain s(z) in the backscattered signal is proportional to the gradient of the time delays in the direction of the ultrasonic beam (z). Assuming a uniform speed of sound cTo(l) = c0 before heating, the apparent strain is given by:

0

)()()(.2

)()( 0

czczz

zzc

zs T δετ−≈

∂∂

= Eq.4-8

The apparent strain s(z) that can be measured from the RF signals at position z is therefore equal to the effective strain ε(z) minus the relative variation in speed of sound δc(z)/c0. When no a-priori knowledge is available, the individual contribution of these two effects cannot be determined. However the combined effects can be measured by passive elastography.

Chapter 4. HIFU Lesions

89

Figure 4.5 shows speed of sound vs. temperature in canine liver in vitro, as measured by Techavipoo et al. (2002). Speed of sound vs. temperature measured in canine liver, prostate, muscle and kidney were shown to possess similar parabolic shapes.

20 40 60 80 1001570

1575

1580

1585

1590

1595

1600Speed of sound in canine liver

Temperature (°C)

Spee

d of

sou

nd (m

/s)

Raw data c = -0.0195 T2 +2.28 T +1532

Fig 4.5: Speed of sound vs. temperature in canine liver. A parabolic approximation was superposed on the raw data. (Source : Techavipoo et al. 2002) For a simple 1D model (thin rod) of solid isotropic homogeneous material undergoing uniform heating, the thermal strain ε(Τ )Το is uniform when heating from temperature T0 to temperature T. It can be derived from the specific volume v (the inverse of the density ρ) of the medium. The solid being isotropic, its linear expansion is 1/3 of its volume expansion. Thus the thermal strain is given by the following equation:

⎟⎟⎠

⎞⎜⎜⎝

⎛−=⎟⎟

⎠

⎞⎜⎜⎝

⎛−= 1

)()(

311

)()(

31)( 0

00 T

TTvTvT T ρ

ρε Eq.4-9

In a biological material, thermal expansion may not be isotropic. Such a case could be modeled by changing the 1/3 coefficient in Eq.4-9 to a value ranging from 0 (no thermal expansion in the direction being considered) to 1 (the material expands only along the direction being considered). This 1D model is illustrated for a biological material that possesses the thermal expansion of water (Lemmon et al. 2003) and the speed of sound of canine liver in vitro (Techavipoo et al. 2002). Because of their large water content, thermal expansion of water is likely to be a reasonable approximation for biological tissues. A closed form expression (Eq.4-10) was obtained for the specific volume of water as a function of temperature (given in °C) at 105 kPa pressure using a 2nd order polynomial fit to the data obtained from the literature (Fig.4.6).

9993.010.47.610.82.3)( 526 ++≈ −− TTTv Eq.4-10

Chapter 4. HIFU Lesions

90

Fig 4.6: Specific volume of pure water vs. temperature at 105 kPa pressure (Lemmon et al. 2003). The parabolic

fit (Eq.4-10) superposed in red on the raw data is a good approximation for temperatures above 10°C. Fig. 4.7 shows the thermal strain, the apparent strain due to changes in the speed of sound, and their combined effect on the apparent strain measured from RF signals as a function of temperature in the simulated rod. Fig 4.7 : Individual contributions of thermal expansion (black) and of speed of sound (blue) on apparent strain in a thin rod (1D model) uniformly heated from To=20°C up to 100°C. The resulting apparent strain (red) is the sum of both contributions. The material was supposed to possess the specific volume of water at 105 kPa pressure (Lemmon et al. 2003) and the speed of sound of canine liver in vitro (Techavipoo et al. 2002). From fig 4.7, negative apparent strains (apparent compression) are expected for temperatures in the range 30-55 °C, where temperature-dependent variations in speed of sound dominate over thermal expansion. The most highly negative strain is expected around 55°C. Above 55 °C, apparent strain is expected to increase. Zero strain is reached at 75 °C, where thermal expansion and variation in speed of sound compensate each other. For temperatures above 75 °C, positive apparent strain (apparent expansion) is expected because speed of sound is close to its initial value and thermal expansion becomes the dominant effect. These thresholds depend on the initial temperature: in vivo, the apparent strain is zero at 37 °C.

0 20 40 60 80 1001

1.01

1.02

1.03

1.04

1.05

Temperature (°C)

Spec

ific

volu

me

(dm3 /k

g)

Water at 105 kPa

Lemmon et al. 20032nd order fit (R2=0.9997)

20 40 60 80 100-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

Temperature (°C)

App

aren

t str

ain

(%)

Speed of soundThermal expansionCombined

Chapter 4. HIFU Lesions

91

4.2.3 Materials and Methods

4.2.3.1 Simulations

For HIFU therapy, the transducer can be either focused, creating lesions that can be designed to avoid surface burns (Gelet et al. 2001), or unfocused, creating lesions that begin to extend from the surface of the tissues (Lafon et al. 2000). In the present study, an unfocused transducer was used. A 8x15 mm2 unfocused therapy transducer operating at 10 MHz was simulated, using a 20 s 15 W/cm2 application. The temperature elevation and the lesion inside a liver sample during HIFU application was simulated using the bio-heat transfer equation (BHTE) (Pennes et al. 1948, Lafon et al. 2000). The simulated sample was a 40-mm diameter, 40 mm thick cylinder. Attenuation was set to 0.75 dB/cm/MHz (liver in vitro), specific heat 3639 J/°C/kg, thermal conductivity 0.56 W/m/°C, density 1050 kg/m3, speed of sound was set to 1540 m/s and assumed constant for the estimation of the acoustic pressure field and of temperature. In surrounding water speed of sound was set to 1540 m/s, attenuation 2.2.10-3 dB/cm/MHz2, and water temperature was supposed to be constant (24°C) due to an external cooling system. The thermal dose (Sapareto et al. 1984) was determined locally as a function of time from the simulated temperature distribution:

∫ −° =

t TC dRtD

0

43)(43 )( ττ

Eq.4-11

T(τ) is the instantaneous temperature at time τ expressed in °C, R is a dimensionless scalar equal to 2 if T>43°C or to 4 otherwise. The thermal dose represents the duration (in seconds) of a 43°C exposure that would have an equivalent heating effect on the tissues. Tissues that received a thermal dose equivalent to 2 hours at 43°C or higher were considered as coagulated. Young’s modulus was set to 2.0 kPa in normal liver and 20.0 kPa in coagulated liver (Shi et al. 1999). The temperature distribution was assumed to be unaffected by thermally-induced displacements and by thermal lens effects (variations in the speed of sound). Thermal strain depends on local temperature elevation, tissue elasticity, thermo-elastic properties of the tissues, and on the stress applied by surrounding tissues. Using the simulated temperature distribution, thermal displacements and strains were determined from the theory of thermo-elasticity using Navier’s equation (Nowinski 1978, Jellab 1999):

( ) 0)(21

)1(221

12 =+∆∇−

+−⋅∇∇

−+∇

µβ

νν

νfTuu Eq.4-12

ur is the displacement, ν is the Poisson’s ratio, β is the linear expansion coefficient per Kelvin (K-1), µ is the shear modulus, ∆T is the temperature variation and f

r is the force acting on the

elementary volume being considered, 2∇ is the Laplacian operator, ∇r

is the gradient operator, and ur

r⋅∇ is the divergence of vector ur . Rigid confinement with perfect slip condition was

imposed on the edges of the tissue sample. The strain tensor was obtained from the displacement vector. Thermal expansion of liver was assumed to be equal to thermal expansion of water.

Chapter 4. HIFU Lesions

92

Equation 4-12 assumes that thermal expansion is linear with temperature, but this assumption does not hold for water over the temperature range investigated in this study (20-100°C). To minimize the errors due to the linear approximation in Eq. 4-12, each element in the model was assigned a specific coefficient β, which depended on local temperature T. Its value β (T) was obtained from the mean value of the expansion coefficient in the temperature range T0 to T:

( ) ⎟⎟⎠

⎞⎜⎜⎝

⎛−

−= 1

)()(

31)(

00 TvTv

TTTβ Eq.4-13

The distribution of speed of sound in the sample was estimated from the temperature distribution using speed of sound in canine liver in vitro (Techavipoo et al. 2002). The apparent strain distribution due to speed of sound was directly obtained from the local relative variation in speed of sound. Finally, both thermal strain and speed of sound-dependent apparent strain were added to form the simulated apparent strain elastogram. 4.2.3.2 Experiments

Thermal expansion of liver samples was measured using the setup shown in fig 4.7. Liver samples (approx. 60 ml) were immersed in water inside a waterproof glass container equipped with a volumetric pipette. The initial volume of the liver sample Vl(T0) at room temperature T0 and the initial volume of water Vw(T0) used to fill the container were measured. A thin thermocouple was inserted into the tissue sample to measure the temperature of the liver. The container was placed inside a heated water tank. Liver temperature was increased from room temperature up to 85 °C. The total variation of the volume ∆V(T) inside the container was measured using the pipette at T=37°C, 48°C, 65°C and 85°C. The variation of the volume of water ∆Vw(T) was estimated from the parabolic approximation of the specific volume of water (Eq.4-10) using the relation:

)(.1)()()( 0

0

TVTvTvTV ww ⎟⎟

⎠

⎞⎜⎜⎝

⎛−=∆ Eq.4-14

The variation of the volume of the liver sample ∆Vl(T) was given by the difference between the measured total variation ∆V(T) and the estimated variation of the volume of water ∆Vw(T):

)()()( TVTVTV wl ∆−∆=∆ Eq.4-15 Passive elastography experiments were conducted in vitro in bovine and porcine liver. Liver samples were immersed in a 9 g/L saline solution and degassed at –80 kPa for 30 min. Then, they were set in a PVC cylindrical sample holder (diameter 40 mm, thickness 20 mm). Liver samples were held in the sample holder by a latex envelope. The sample holder was immersed in a tank filled with of degassed water (temperature 24°C). The sample holder stood on one side of the imaging probe at a distance of 4 cm. The therapy transducer was set on the other side of the holder at a distance of 1-3 mm from the latex envelope (Fig 4.8).

Chapter 4. HIFU Lesions

93

Heating

120ms 120ms … 120ms 120ms 120ms

RF acquisition

1min 1min

Time t0=0 Cooling

Imaging scanner Generator

Imaging probe

Therapy transducer

Sample holderTank

Degassed water

Imaging plane

Synchronisation

Fig 4.8: Experimental setup The imaging system was based on a standard ultrasound scanner (Kretz Combison 311, Austria) equipped with a RF output. The imaging probe was a 5 MHz rotating transducer focused at 5 cm capable of providing 8 frames per second. RF data were digitized using a 14-bit 100 MHz A/D converter (Gage CompuScope 14100, Canada) with 256 MB on-board memory and stored for off-line processing. The high intensity ultrasound beam was generated by a 10 MHz flat (unfocused) 8x15 mm² transducer. In order to avoid interference between the imaging beam and the heating beam, the emission was interrupted for 10 ms during the acquisition of the RF data (diagram shown in Fig 4.9). Given that the acquisition frame rate was 8 fps (120 ms between frames), the resulting duty cycle of the emission was 92%. Experiments were carried out with sequence S1 (5 s, 10 W/cm² emissions) designed to induce temporary (reversible) temperature elevation without thermal damage, and with sequence S2 (20 s, 15 W/cm² emissions) designed to induce irreversible coagulation.

Fig 4.9: Acquisition procedure for synchronized heating & imaging. Instantaneous apparent displacements were estimated from time delays between consecutive RF frames using cross-correlation at zero lag with 1 mm window length and 0.5 mm window shift and parabolic interpolation. Instantaneous apparent strain was obtained using the gradient of the displacement. The maximum value of the cross-correlation function for each window was also interpolated and stored to create a correlation image. Cumulative apparent displacements at time t were obtained using a summation of the instantaneous apparent displacements starting from the beginning of heating (time t0=0). Cumulative apparent strain at time t was determined from the axial gradient of the cumulative apparent displacements.

Chapter 4. HIFU Lesions

94

4.2.4 Results

4.2.4.1 Simulations

Fig. 4.10 shows (a) the expected temperature distribution in the liver sample, (b) the expected distribution of speed of sound, (c) the expected mechanical strain due to thermo-elastic behavior (thermal expansion), (d) the expected apparent strain due to changes in speed of sound, (e) the expected apparent strain due to combined effects of thermal expansion and of changes in speed of sound, and (f) the expected lesion based on thermal dose estimation. This simulation was performed for heating sequence S2. The edges of the lesion, defined from fig. 4.10f, are outlined in all figures. The expected lesion (Fig. 4.10f) was approximately conical, 8-mm long, 8-mm wide near the transducer and only 4-5 mm wide near its apex . The 'saw-tooth' aspect of the lesion margins are an artifact due to poor spatial sampling; the true margins are expected to be continuous. The expected temperature (Fig. 4.10a) reached up to 95°C inside the lesion. A steep temperature gradient was predicted close to the liver/water interface (position x=0, top of images) due to water cooling. Speed of sound (Fig. 4.10b) ranged from ~1580 m.s-1 far from the lesion up to ~1600 m.s-1 near the edges of the lesion, and decreased down to ~1575 in the very heart of the lesion. Interestingly, the area of maximum speed of sound showed good correspondence with the lesion margin. Thermal strain due to thermal expansion (Fig. 4.10c) was mostly positive (expansive), and reached up to 1.5% inside the lesion. The negative (compressive) thermal strain area at the liver/water interface is due to the presence of tissue still not coagulated because of water cooling. Apparent strain due to speed of sound (Fig. 4.10d) was minimal (-1.3%) near the edges of the expected lesion (corresponding to the maximum of speed of sound), decayed slowly to zero outside the lesion, and reached up to 0-0.2% inside the lesion.

Chapter 4. HIFU Lesions

95

Fig 4.10: Simulated (a) temperature distribution, and corresponding (b) speed of sound, (c) axial strain due to thermal expansion only, (d) apparent strain due to speed of sound only, (e) total apparent strain due to combined effects, and (f) lesion surface (gray) estimated from the thermal dose. In the total apparent strain distribution (Fig. 4.10e), an apparent expansion area (red) was visible in the heart of the HIFU lesion, while a 'V'-shaped apparent compression area (blue) developed distal from the therapy transducer. The surface of the lesion included the apparent expansion area and a 1- to 2-mm peripheral area, which approximately corresponded to a –0.5% threshold in apparent strain on the lateral edges and to a –1.5% apparent strain threshold on its distal edge (bottom of figure).

(a) (b)

(c) (d)

(e) (f)

Chapter 4. HIFU Lesions

96

20 30 40 50 60 70 80 90-0.01

0

0.01

0.02

0.03

0.04

Temperature (°C)VoumeExpansondmensoness Water

Liver

4.2.4.2 Experiments

Fig. 4.11 shows the relative volume expansion of porcine liver (i.e. ∆Vl(T) / Vl(T0) ), with reference temperature T0=24°C, measured from N=6 samples, with corresponding error bars. The dotted line corresponds to reference data for pure water. The use of the thermal expansion coefficient of water for liver in the simulation was justified by the good experimental correspondence between liver and water.

Fig. 4.11: Experimental measurement of volume expansion of porcine liver, relative to its initial volume at reference temperature T0=24°C.

The lesions observed from gross pathology in the liver samples are shown in Fig. 4.12. Images (a)-(c) correspond to a lesion in bovine liver for exposures of (a) 10 s, (b) 16 s, and (c) 20 s, in a plane transverse to the imaging plane shown in Fig. 4.10. Figure 4.12d shows a lesion in porcine liver for a 20-s exposure, in the imaging plane.

Fig. 4.12: Gross pathology examination of the HIFU lesion in bovine liver for a 15 W/cm2 exposure of (a) t=10 s, (b) t=16 s, (c) t=20 s in a plane transverse to the imaging plane shown in Fig. 4.10, and (d) in the imaging plane in porcine liver for a 20-s, 15-W/cm2 exposure. The lesion contours visually identified from gross pathology are outlined. The scale is identical in the 4 pictures. The depth of the lesions measured from pathology and from BHTE simulations for a 15 W/cm2 exposure are shown in Table 4.1. Good correspondence was observed for 10-, 16- and

(a) t =10 s (b) t =16 s (c) t =20s (d) t = 20 s

Therapy transducer

Chapter 4. HIFU Lesions

97

20-s exposures. As experimental measurements were obtained after cooling, the lesion depth for a 20-s long exposure should be compared with the dimension of the simulated lesion after cooling (at time t>20 s). These results validated the BHTE model, and allowed for comparison between correlation images and/or apparent strain elastograms with the simulated lesions in the following paragraphs.

Exposure duration (s) 10 14 16 18 20 25 30 Lesion depth in experiments (mm) 2-3 X 5-7 X 7-9 X X Lesion depth in simulation (mm) 3 6 6.5 7 8 8.5 9 Lesion width in simulation (mm) 6 7.5 7.5 8 8 8.5 9 Table 4.1: Comparison of the depth of the lesion in BHTE simulations and in experiments for HIFU sequence S2, and largest simulated lesion width in the imaging plane. HIFU application was turned off after t=20 s, but the lesion continued to expand between 20 and 30 s due to heat diffusion. Experimental measurements were not available for 14-s, 18-s, 25-s and 30-s long exposures (X). Fig. 4.13 shows the correlation of RF frame acquired t0=0 (referred to as 'frame 0' in the following text) with RF frames obtained at times t=5 s, t=5 min and t=10 min in the absence of heating. The notation ρ (t1,t2) is used in the following paragraphs to denote the correlation between RF frames acquired at times t1 and t2. These acquisitions were performed to ensure that stable correlation was obtained during the experiment. A slow decrease of the average correlation with time was observed, probably due to tissue decay. The maximum duration of subsequent heating experiments was restricted to 5 min maximum to ensure that eventual decorrelation was mainly due to heating and not to unavoidable decay with time.

Fig. 4.13: Correlation of frame 0 with RF frames acquired at (a) t=5 s, (b) t=5 min, (c) t=10 min in the absence of any heating. A slow decay of correlation with time was observed. The decay was faster in the absence of saline solution. For heating sequence S1, Fig. 4.14 shows (a) the correlation ρ (0,5 s) immediately after 5s heating, and (b) the correlation ρ (0,5 min) after 5 min cooling. A temporary decrease in correlation was observed immediately after heating in the target area. High correlation (0.95-1.0) was almost fully recovered after cooling.

Liver-water interfaces

(a) (b) (c)

Chapter 4. HIFU Lesions

98

Fig. 4.14: Correlation images for a 5 s 10 W/cm² ultrasound emission: (a) immediately after heating, (b) after 5

min cooling.

Fig. 4.15 shows similar correlation images for heating sequence S2. Immediately after heating a large decorrelation area was observed. It was larger than the 8x8 mm2 coagulation lesion expected from simulations (solid line). After 5 min cooling the decorrelation area was slightly shrunk but was still clearly visible, showing that permanent alteration was induced in the tissues. The lesion surface was still largely overestimated after cooling.

Fig. 4.15: Correlation image for a 20 s 15 W/cm² ultrasound emission: (a) immediately after emission, (b) after 5 min cooling. The lesion expected from simulation is outlined.

Fig. 4.16 shows a M-mode image of the experimental correlation ρ (t-0.5, t) (RF frames t and t-0.5 s) averaged in a 8-mm wide region corresponding to the expected lesion width. The therapy transducer is at the top of the image. The depth of the simulated lesion is shown by an arrow on the right end side. Four areas can be observed in the M-mode. In area A (0-14 s), moderate decorrelation (light red, ~0.95) was observed in a 8-mm deep region between 0 and 14 s. Significant decorrelation (dark blue ~0.60) was visible in the target area between 14 and 15 s (area B), and low correlation (yellow-light blue, ~0.85) was maintained between 15 and 20 s (area C). Heating was interrupted after 20 s and moderate decorrelation (~0.90) was observed between 20 and 22.5 s (area D).

Therapy transducer

Liver/water interface

Imaging direction

(a) (b)

(a) (b)

Chapter 4. HIFU Lesions

99

Simulated lesion depth

8 mm Dep

th (m

m)

Time (s)

A B C D

Figure 4.16: Experimental M-mode image of correlation ρ (t-0.5, t) average over a 8-mm wide area corresponding to the expected lesion width. The time interval between RF frames used for correlation was 0.5 s. The depth of the decorrelation area showed good correspondence with the expected depth of the lesion. The moderate decorrelation observed in area A was probably due to the progressive increase in temperature. The lowest correlation (area B, 14-15 s) may correspond to the coagulation threshold, when tissue structure is changing. In area C, it is likely that most tissues were already coagulated but eventual structural changes as well as continued temperature rise might have occurred. In area D, heat diffusion was likely to keep inducing changes in the RF signals although heating was off. Fig. 4.17: Cumulative apparent strain elastograms acquired at (a) t=10 s, (b) t=14 s, (c) t=16 s, (d) t=18 s, and (e) t=20 s during the formation of the lesion for heating sequence S2. The color map ranges from –10% (apparent compression) to +10% (apparent expansion). The images correspond to a 16x30 mm2 area. The simulated lesions are outlined.

Fig. 4.17 shows the cumulative apparent strain elastograms at different times during heating. Low apparent compression was observed in the target area during the first 10 s of emission. Then an apparent expansion zone appeared near the transducer and expanded quickly between 14 and 18 s, while the apparent compression zone moved further from the transducer. After 20 s, the final elastogram showed an apparent expansion zone with good correspondence with the expected lesion, and a distal apparent contraction area. The experiment was repeated on 3 liver samples. The cumulative apparent strain elastograms measured in porcine liver and in bovine liver were similar.

-5 0 5-5 0 5-5 0 5-5 0 5-5 0 5

50

55

60

65

70

Water-liver interface

Therapy transducer

(a) 10 s (b) 14 s (c) 16 s (d) 18 s (e) 20 s

Expected lesion

-10

-5

0

5

10

Chapter 4. HIFU Lesions

100

The apparent strain was plotted vs. time in arbitrarily chosen locations. Fig. 4.18 shows typical apparent strain profile vs. time (a) in the apparent expansion zone corresponding to the expected lesion, and (b) the apparent compression zone. Inside the expected lesion, apparent strain slowly decreased from 0 to–2% (compressive strain) during the first 7-8 s, then it increased slowly up to +2% (expansive strain) until approximately 15 s. Then a sudden inflection was observed in the profile, and apparent strain increased rapidly up to approximately +15% for time t=20 s, and continued to increase up to +25% between 20 s and 25 s although the therapy transducer was turned off. Inside the apparent compression area, the apparent strain profile was slowly decreased from 0 to –2% during the first 14-15 s, then it decreased rapidly from –2% down to –7% for time t=20 s. After heating was turned off, apparent strain continued to decrease down to–9.5% between 20 and 25 s. The apparent strain decayed back to zero after 5 min cooling for temporary heating (sequence S1), but not for therapeutic sequence S2.

Figure 4.18: Apparent strain vs. time (a) in location A, inside the apparent expansion zone, (b) in location B,

inside the apparent compression zone. 4.2.5 Discussion

From the theory (Fig. 4.7) and the simulation (Fig 4.10), apparent expansion was expected inside the lesion because of the dominant effect of thermal expansion. This was confirmed in the experiment (Fig. 4.17). Apparent compression was expected and experimentally observed in areas where temperature was below ~75°C because of the dominant effect of speed of sound. The apparent compression was enhanced at the deep edge of the lesion by the mechanical compression exerted by the expanding tissues, as visible from the negative (true) strain below the lesion in Fig. 4.10. The apparent compression area had a typical 'V' shape in this area, which was also observed experimentally. The pattern in the experimental cumulative elastograms showed very good correspondence with apparent strain simulations (compare Fig. 4.17e with Fig. 4.10e). In both cases the edges of the simulated lesion corresponded to a rim of negative apparent strain surrounding an apparent expansion area. However a significant discrepancy was observed in the amplitude of apparent strain, as experimental apparent strain ranged from –7% to +15% while simulated apparent strain ranged from –2% to +2%. The discrepancy may arise from an thermo-elastic expansion phenomenon occurring with coagulation, or from a possible systematic build-up of bias errors in the experimental cumulative elastograms. The amplitude of the RF signals (Alam and Ophir 2000) and the errors in parabolic approximation of the time delays

Chapter 4. HIFU Lesions

101

(Céspedes et al. 1995) were reported to induce such bias (see chapter 1.3.3.7). No significant bias was experimentally observed in areas located far from the target area, where cumulative displacements and apparent strain were close to zero, but the bias due to parabolic interpolation is known to occur for non-zero displacements (Céspedes et al. 1995). No significant hypo- or hyper-echoic area was observed in the target area that would allow to suspect an underline artifact due to a variation in RF signal amplitude. Interestingly, the inflexion observed in Fig. 4.18 in both apparent strain profiles corresponded to the time (14-15 s) when the lowest correlation was observed in the M-mode correlation image (zone B in Fig. 4.16). This instant was likely to correspond to the coagulation threshold. The large variations in apparent strain in the next 5 s following the end of HIFU application were not expected, and may have been induced by continued thermal expansion in the presence of heat diffusion, as suggested by the continued increase in simulated lesion dimensions at time t=25 s and t=30s (Table 4.1). Correlation images obtained from RF frames acquired before and after heating were not representative of the lesion. The depth of the lesion was visible in the M-mode correlation image obtained from consecutive RF frames (fig. 4.16) using a spatial averaging of correlation on adjacent lines, therefore at the cost of resolution. Nevertheless, during the first 10 s of HIFU application, the depth of the affected area in the M-mode correlation image (~8 mm) was not representative of the depth of the lesion at that particular time (2-3 mm). The value of the correlation coefficient was probably affected by structural changes in the tissues during coagulation, but it also depends on many other parameters such as sonographic signal-to-noise ratio, intra-window (apparent) strain, and amplitude of eventual lateral and/or elevational motion. The major limitation in using correlation images was that it was not possible to clearly identify coagulation-induced decorrelation from decorrelation induced by others parameters. These parameters influence the precision of the strain estimates, but not their accuracy. As mentioned previously, for the sake of simplicity the unfocused transducer and lesion it generated have been referred to as HIFU in this section, although historically HIFU at first denoted focused transducers only. The basic principles underlying the formation of the lesion are the same with both transducer types, so the results found in this chapter are also expected to apply to focused transducers. There exist some differences however. First, for an unfocused transducer the lesion is initiated near the tissue-water interface and it propagates deeper as ultrasound application is maintained. For a focused transducer, the lesion is initiated at the focus of the transducer, and it propagates towards the transducer (Chapelon et al. 1992). This propagation is expected to be visible in cumulative apparent strain elastograms. Second, focused transducers are highly likely to induce cavitation because of the very high acoustic pressure generated at their focus, whereas unfocused transducers are usually driven with acoustic power levels that are below the cavitation threshold. In our experiment, no cavitation was expected because of the high frequency (10 MHz) and of the low acoustic intensity (15 W/cm2). No hyper-echoic area was visible in the target area after ultrasound application that would have suggested that cavitation bubbles appeared. If cavitation were to occur, the phenomenon would very likely induce significant changes in the backscattered RF signals used in passive elastography: in the first milliseconds of ultrasound application, microbubbles present in the acoustic pressure field grow and eventually reach their resonant size. Then two different types of cavitation are to be distinguished. During inertial (transient) cavitation, bubbles collapse and are destroyed, thus generating new cavitation germs that may or may not grow to form new bubbles. In non-inertial (stable) cavitation, bubbles oscillate non-linearly

Chapter 4. HIFU Lesions

102

around their resonance frequency. The creation of bubbles from the cavitation germs and destruction of bubbles by inertial cavitation are highly likely to induce significant variations in backscattered RF signals, while bubble oscillation during non-inertial cavitation can be expected to be a stable process that would leave the bubbles unchanged between two imaging acquisitions. Therefore, RF signal decorrelation can be expected in passive elastography in the presence of inertial cavitation, but may not occur in the presence of non-inertial cavitation. Experiments need to be carried out to corroborate these assumptions and to determine if passive elastography is able to show the creation of HIFU lesions in the presence of cavitation. The ripple artifact previously reported by several groups during ultrasonic temperature imaging experiments was not observed (Simon et al. 1997, Le Floch et al. 1997, Alaniz et al. 2002). If present, it should have been visible as a typical decorrelation area behind the HIFU lesion, i.e. between the lesion and the therapy transducer. This decorrelation artifact arises from the deviation (refraction) of the ultrasonic imaging beam (due to temperature-related local variation of speed of sound) between two consecutive RF signal acquisitions, which is known as the thermal lens. The deviation angle directly depends on the variation in speed of sound, i.e. on the variation in temperature. From an elastographic point of view, the deviation of the imaging beam by the thermal lens is equivalent to a lateral (or elevational) motion of the scatterers. Correlation between the RF signals is expected to decrease with increased lateral motion (beam deviation), therefore with increased temperature variation. In our experiments, high acquisition frame rate (8 fps) ensured that temperature variation between two acquisitions was small, so that refraction of the beam (lateral motion) was expected to be small enough not to have a significant impact on correlation and on time-delay estimation. This is similar to handling lateral (and/or elevational) motion problems with high acquisition frame rates in conventional elastography, as discussed in chapter 3.2.3. During thermal expansion, the tissues also undergo lateral and elevational strains. In the future, lateral apparent strain elastograms may also be acquired to determine if they can provide some information about the creation of the lesion. Low elastographic signal-to-noise ratio (SNRe) is expected in lateral strain elastograms because of poor lateral motion tracking accuracy, but this problem may be minimized using cumulative (multi-compression) displacements estimates. The lateral apparent strain is expected to be dependent on the mechanical lateral strain due to thermal expansion, and on the refraction angle due to the thermal lens. It may be possible to set the frame rate high enough to ensure that refraction is negligible. The next step will be to determine if passive elastography is able to show the formation of the HIFU lesion in vivo. Cumulative apparent strain elastograms presented in this study strongly rely on the stability of the experiment. In vivo, uncontrollable tissue motion will prevent cumulative displacements to be tracked over a period as long as 20 s. An alternative would be to develop a robust registration algorithm, but using a standard ultrasound scanner the algorithm would be ultimately limited to tracking displacements in the imaging plane, while out-of-plane displacements will be unrecoverable. A more realistic option is to use the instantaneous apparent strain rate elastogram (calculated using RF frames N and N+1, then N+1 and N+2, …), eventually integrated over a short time interval (cumulated over RF frames N to N+p, then N+1 to N+p+1, etc… where p is the integration duration expressed in number of RF frames) to improve SNRe. Longer time intervals would provide better SNRe at the cost of higher sensitivity to uncontrolled motion. The corresponding elastograms would show the apparent strain rate, i.e. the expected coagulation front, instead of the cumulative apparent

Chapter 4. HIFU Lesions

103

strain. An example of such an imaging sequence is shown in Fig. 4.19, where elastograms show the instantaneous apparent strain rate integrated over a 1 s long duration for HIFU sequence S2. Fig. 4.19: Instantaneous apparent strain rate elastograms acquired at (a) t=10 s, (b) t=14 s, (c) t=16 s, (d) t=18 s, and (e) t=20 s during the formation of the lesion using cumulative elastograms between time t and time t+δ t. The color map ranges from –1.5%/s (apparent compression) to +1.5%/s (apparent expansion). The images correspond to a 16x30 mm2 area. The simulated lesions are outlined. Note the correspondence between the distal (lower) edge of the lesion and the apparent expansion zone (red). 4.2.6 Conclusion

Cumulative apparent strain elastograms were acquired during HIFU application in three porcine and one bovine liver samples in vitro. Apparent strain was induced in the backscattered RF signals received from the heated tissues because of the combined effects of tissue expansion and temperature-dependent speed of sound. No external compression was exerted during HIFU application, thus the technique was referred to as passive elastography (or thermally-induced elastography). A HIFU lesion was induced using an 8x15 mm2 unfocused therapy transducer operated at 10 MHz frequency and 15 W/cm2 acoustic power during 20 s. Imaging was performed at 8 fps with a 5 MHz transducer while heating. The HIFU transducer was briefly switched off during the acquisition of each RF frame to avoid interference between the imaging and the therapy beams. In correlation images, the decorrelation area largely overestimated the lesion area. The lesion was visible in the elastograms as an apparent expansion area, whereas surrounding tissues exhibited apparent compression. The apparent expansion area showed good correspondence with the lesion size predicted from simulations. Good correspondence in lesion size and shape was also found with gross pathology. Apparent strain distributions were simulated based on the theory of thermo-elasticity and on speed of sound measurements in vitro. The simulated apparent strain distributions corresponded in size and amplitude with experimental data before coagulation occurred. After coagulation, the size and amplitude of the apparent expansion area was much larger in experiments than in the simulated elastograms. However the size of the apparent expansion area showed good correspondence with the simulated lesion size. Cumulative apparent strains as high as +15% (inside the lesion) and as low as –7% (outside the lesion) were measured. We hypothesized that unexpected but significant tissue expansion occurred with tissue coagulation, which made the lesion detectable by passive elastography during its formation.

-5 0 5

50

55

60

65

70

-5 0 5 -5 0 5 -5 0 5 -5 0 5 (a) 10 s (b) 14 s (c) 16 s (d) 18 s (e) 20 s

Water-liver interface

Therapy transducer

Expected lesion

-1.5

-1

-0.5

0

0.5

1

1.5

Conclusion

104

Discussion Before a final conclusion is drawn to this thesis, this chapter is meant to provide an insight into questions that remain unanswered and to future directions for research. What is the contribution of this work over the work of Ermert et al. ? Ermert et al. showed that elastography is able to detect prostate cancer in vivo. Using handheld compression by the imaging probe and real-time feedback, they reported a sensitivity of 76% and a specificity of 84% in a preliminary study in 170 patients. As compared to their work, our study: • Confirmed that some prostate cancers can be detected by elastography in vivo, using a

different acquisition system, based on a balloon compressor, and a different data processing algorithm.

• Demonstrated that elastography is also able to show the zonal anatomy of the prostate and to see benign prostatic tumors. In order to minimize the risk of false positive during diagnostic, future studies are needed to find criteria that can differentiate between benign and malignant tumors.

• Showed that there exist a natural strain contrast between the transition zone and a softer peripheral zone. As cancer is known to be stiff and mostly found in the peripheral zone, this contrast may be advantageous for cancer detection. This finding seems to contradict the results of Krouskop et al. (1998) that showed no significant modulus contrast between ‘anterior’ and ‘posterior’ tissues. However in this article it is not clear what ‘posterior’ and ‘anterior’ refer to, and especially if these notions correspond to the pathologist’s definition of peripheral and transition zones.

Is there a fundamental difference between fixed displacement and fixed stress imaging ? Our acquisition system was based on fixed stress imaging (uniform pressure), whereas most groups used fixed displacement imaging with a rigid compression plate. A similar design was proposed for intravascular elastography (de Korte, Brusseau, Ermert). In either case, strain elastograms show the behavior of tissues under a given load. A strain contrast observed between different areas may be either due to a stiffness contrast (soft vs. stiff) or to a non-uniform load (more or less compression). As a results, stiffness contrast can be ascertained only if both strain distribution and stress distribution are known, i.e. as the solution of an inverse problem. In a real situation stress cannot be measured. Using fixed displacement compression, the applied stress is non-uniform and the interpretation of strain elastograms can be complex and at worst misleading. This was identified as a potential problem for fixed displacement imaging, and displacement apodization was proposed to minimize strain artifacts (Konofagou et al. 1996). However apodization requires some a-priori knowledge of the geometry of the observed organ, which may not be available or reliable in patients. Using fixed stress compression, stress non-uniformities are minimized, resulting in strain being estimated under a uniform load. However some fundamental limitations still exist, that also apply to fixed displacement imaging. Uniform stress can be applied at the compressor interface, but inside a heterogeneous medium the uniformity of the stress distribution cannot be ensured. And even in a homogeneous medium, strain and stress are likely to be depth-dependent.

Conclusion

105

From our experience using a balloon compressor in patients can be difficult because the behavior of the balloon depends on boundary conditions such as patient anatomy and how the balloon is attached to the imaging probe. On the other hand, it is easy to ensure stable positioning of the imaging probe when using the balloon compressor. Fixed displacement vs. fixed stress can therefore be thought of as a tradeoff between surface stress uniformity, organ accessibility, practical implementation issues and clinical acceptation. In the future it is worth investigating the effective benefits and drawbacks of using the balloon compressor, compared to simple handheld compression. Is there an optimal grayscale allocation for the blind reading of elastograms ? So far being able to see a cancer in an elastogram has been sufficient for most feasibility studies. But in the near future clinical evaluation of elastography will be conducted and the results may vary depending on the chosen grayscale. As in any other imaging modality, the choice of the grayscale dynamic range can have a significant influence on the detectability of cancer or other structures of interest. The dynamic range of the measured strains needs to be matched to the dynamic range of colors that is perceivable by human vision. There is a controversy about whether to use pseudo-colors or a gray scale. Pseudo-colors yield a wide dynamic range but induce possibly misleading artificial segmentation which is highly dependent on the color allocation. Grayscale does not induce segmentation but has narrower dynamic range. During our study we proposed a composite grayscale that allocates 70% of the gray levels to low strains and 30% only to high strains, based on the assumption that we were looking for stiff cancer and stiff HIFU lesions. Additional information can be used to chose the strain dynamic range to be displayed. For example the normalized cross-correlation value is a quantitative indicator of the quality and reliability of the estimate, and unrealistic strain estimates can be detected based on the elastographic dynamic range, which is given by the strain filter theory. Investigating the display options and their influence on the interpretation of the elastograms should be a future topic of investigation. How fast should the acquisition be ? Increasing the acquisition frame rate up to 8 fps resulted in a significant improvement in the quality of the elastograms in vivo, because undesirable inter-frame motion was minimized. But increasing the frame rate requires expensive hardware and eventually long processing times if multi-compression is used. On an engineering point of view it is important to determine where is the point where there are no longer observable improvement with speed. There might also exist a limit to speed because of the visco-elastic properties of soft tissues. It is important to clarify the influence of the acquisition frame rate, the compression rate, the rate of natural motion, and of the number of compression steps on image quality (SNRe, CNRe). How to optimize cancer detection ? Detectability is determined by the size to resolution ratio and by CNRe. The key factors are therefore the resolution, the strain contrast, and the noise level in each portion of the elastogram. Axial resolution is given by the window length and window shift, and lateral resolution by the beam width. For an elastic material the strain contrast is directly proportional to the applied strain, but in general it can be modified by the nonlinear behavior of some tissues, and/or by their visco-elastic (time-dependent) behavior. The noise level results from a complex interaction between the local strain, ultrasonic parameters (bandwidth, central frequency), processing algorithm and parameters, non-axial motion, and can be predicted by the strain filter theory. The data available in the literature suggest that, in the range of strains from 2% to 4%, stiff prostate cancer is highly nonlinear whereas soft normal tissues behave linearly (Krouskop et

Conclusion

106

al. 1998). In the presence of nonlinear behavior, pre-compression has a major role in determining the strain contrast. Detectability could be improved by nonlinearity provided that the gain in strain contrast is not compensated by a loss in image quality (increased noise). This may be achievable using high frame rate multi-compression, which is likely to provide a high quality sequence of elastograms during continuously varying pre-compression. Why didn’t we observe strain decay in vivo ? Low strain was measured at large depth in phantoms and in vitro because of strain decay, and resulted in low SNRe at large depth. However this phenomenon was not observed in vivo. It is possible that strain decay was minimized because of the presence of the public bone, that acted as a compressor located on the anterior face of the prostate. The presence of a second compressor plate behind the observed object was previously reported to induce stress decay and re-establishment with depth (Ponnekanti et al. 1992). What is the effect of post-HIFU bubbles on the elastograms ? The presence of hyper-echoic bubbles increases SNRs, and can therefore be beneficial to SNRe. But if bubbles become the dominant source of scattering then time-delay estimators will track bubble motion instead of tissue motion. Because of their fragility, bubbles may collapse during elastographic compression, resulting in scatterers disappearing/changing between frames and therefore to a loss in correlation. For example, loss of correlation (LOC) imaging uses bubble destruction as its signal source for non-quantitative flow assessment. Moreover increased backscattering may be associated with increased attenuation at large depth, which in turn results in lower SNRe. The investigation of these complex interactions may be useful to determine if elastograms should be acquired immediately after HIFU, while hyper-echoic bubbles are numerous, or to wait a few minutes until they faded. Such a study may open the way to an eventual use of contrasts agents in elastography.

Conclusion

107

Conclusion (Français) Cette étude est dédiée à l’application de l’élastographie à la détection du cancer de la prostate et à la visualisation des lésions HIFU. Un système d’imagerie par élastographie a été conçu et réalisé pour une utilisation in vivo. Il est basé sur une sonde échographique endorectale sectorielle travaillant à une fréquence centrale de 5 MHz. La sonde est couverte par un ballon en latex qui peut être gonflé afin d’induire une contrainte radiale uniforme sur les tissus environnants. Ce système peut sembler un peu plus difficile à utiliser qu’une simple compression manuelle par la sonde, mais il présente trois avantages par rapport à celle-ci : (1) Si l’on s’assure que la sonde d’imagerie et le ballon sont centrés, alors la composante principale des déplacements induits et la direction de propagation de l’onde ultrasonore sont colinéaires, (2) une pression uniforme est appliquée sur la surface des tissus, (3) la sonde peut être montée sur une table motorisée pour obtenir un positionnement stable. Le système est compatible avec un appareil HIFU de façon à permettre l’évaluation du traitement pendant l’opération. Le système a été validé sur des fantômes cylindriques homogènes et sur des fantômes contenant une inclusion rigide. Ces fantômes étaient conçus pour simuler un examen endorectal. La déformation appliquée et le rapport signal-sur-bruit élastographique (SNRe) a été estimés expérimentalement sur un fantôme homogène volontairement non confiné afin d’inclure l’effet néfaste des déplacements latéraux dus à la compression, qui sont inévitables in vivo. Le plus haut SNRe a ainsi été obtenu par une technique d’acquisition en multi-compression, consistant à cumuler plusieurs petits pas de compression, pour un pas de déplacement du ballon inférieur à 0.5 mm (déformation maximale 1.5%). Pour une taille de fenêtre de 2 mm et sans chevauchement entre fenêtres (résolution axiale estimée ~3 mm) le SNRe expérimental maximal était d’environ 4 pour un seul pas de compression et de 6 pour une déformation maximale de 6% obtenue par 8 pas de multi-compression. Ces valeurs sont approximatives et sont en pratique sujettes à de larges variations en fonction de la taille de fenêtres utilisées et de l’amplitude des déplacement non axiaux, mais permettent de quantifier les performances du système dans des conditions d’utilisation proches de la réalité. Ces essais ont aussi montré que la qualité d’image était très sensible aux déplacement indésirables (latéraux ou élévationels). Afin d’évaluer les possibilités de l’élastographie pour le diagnostic du cancer de la prostate, les élastogrammes acquis in vitro sur 13 prostates ont été comparés aux résultats anatomopathologiques. Les élastogrammes obtenus in vitro présentent des contours nets et un contraste important. L’anatomie zonale de la prostate était clairement visible sur toutes les prostates. Les nodules d’hyperplasie glandulaires bénins (BPH) sont apparus sous la forme de zones ellipsoïdales rigides aux contours nets (4 cas). Les nodules myomateux ont été vus soit sous la forme de zones peu ou pas contrastées mais clairement démarquées par des contours nets de fortes déformations, soit sous la forme de zones de décorrélation liées à leur nature hypo-échogène. Quatre tumeur malignes sur 10 ont été détectées par lecture en aveugle dans la zone périphérique, et une sur 6 dans la zone de transition. Les tumeurs malignes sont généralement apparues comme des zones dures, à faible contraste, et présentant des contours flous. Ceci suggère qu’une étude serait nécessaire pour déterminer si les critères de contraste

Conclusion

108

et de netteté des contours peuvent permettre de différencier l’adénocarcinome des tumeurs bénignes. Aucun foyer tumoral de volume inférieur à 0.5 cm3 n’était visible dans les élastogrammes. Bien que ces résultats puissent apparaître décevants, il est important de noter que le dispositif expérimental était conçu pour une étude de faisabilité mais n’était pas optimisé pour la détection du cancer. Dans le futur, une telle optimisation doit inclure la possibilité d’exercer une compression bien plus importante sur la prostate afin d’exploiter l’augmentation de contraste liée à l’augmentation des déformations, et éventuellement le comportement mécanique non linéaire du cancer. D’un point de vue pratique, on peut estimer qu’une compression allant jusqu’à environ 15-20% est réalisable en clinique, alors que notre dispositif in vitro était limité à 2-3% à cause de la fragilité du gel. Afin de détecter la cancer de la prostate in vivo, 69 patients ont été examinés par élastographie. Malgré le nombre limité d’acquisitions (3 par patient), des foyers cancéreux confirmés par biopsies et échographie ont été trouvés dans les élastogrammes chez 5 patients. Les foyers tumoraux sont apparus comme des zones dures dans la zone périphérique de la prostate, présentant un fort contraste de déformation (de 3.8 à 9.5) avec les tissus sains pour une précompression initiale de l’ordre de 15-20%. Le fréquent manque de correspondance entre les biopsies et les échographies a fortement limité le nombre de cas pour lesquels les élastogrammes ont pu être interprété avec certitude, ce qui a conduit à arrêter cette étude. Il sera dans le futur nécessaire d’acquérir les élastogrammes sur des patients devant subir une prostatectomie radicale afin de pouvoir comparer l’élastographie avec les résultats anatomopathologiques. L’étude a cependant montré qu’une haute cadence d’acquisition était nécessaire, mais non suffisante, pour acquérir des élastogrammes exploitables du point de vue clinique. A une cadence d’acquisition de 8 images par seconde (im/s), la prostate était clairement visible dans 71% des élastogrammes. Les 29% restants étaient généralement dus à une compression non uniforme induisant des déplacements latéraux ou hors-plan significatifs (se traduisant par une image de déplacements fortement asymmétrique et/ou une décorrélation importante de toute l’image) ou à un rapport signal-sur-bruit échographique faible (décorrélation dans les zones hypoéchogènes). De fait, on peut s’attendre à obtenir des élastogrammes de meilleure qualité en augmentant encore la cadence d’acquisition, en contrôlant mieux la directivité de la compression, et en augmentant le gain de transmission pour mieux « voir » dans les plages hypoéchogènes. Un affichage des élastogrammes en temps réel permettrait au radiologue de sélectionner les meilleures images, et de faire suffisamment d’acquisitions pour « explorer » la prostate de façon adéquate en vue du diagnostic. Alors seulement il sera possible de tirer des conclusions sur la sensibilité et la spécificité de l’élastographie pour la détection du cancer de la prostate. Néanmoins les résultats de cette étude ont confirmé qu’il est possible de détecter le cancer de la prostate in vivo par élastographie avec notre dispositif. Afin de déterminer s’il est possible de visualiser les lésions HIFU in vivo par élastographie, des élastogrammes ont été acquis sur 69 patients ayant subi un traitement HIFU pour un cancer localisé de la prostate. Trois acquisition ont été faîtes à mi-parcours du traitement (i.e. juste un lobe prostatique traité), puis immédiatement (3-5 min) après traitement. Les lésions HIFU sont apparues comme des zones dures (faible déformation) nettement discernables lorsque des tissus non traités étaient présents pour avoir un contraste, ceci même en présence de bulles de cavitation visibles sur l’échographe. Les élastogrammes dépourvus de contraste laissent à penser que la totalité du volume prostatique était traité. L’hyper-échogènicité due aux bulles accroît le rapport signal-sur-bruit échographique dans la prostate et est donc un facteur favorable pour l’élastographie. La taille et la position des lésions HIFU dans les élastogrammes ont montré une bonne correspondance avec les zones ciblées, et ont été

Conclusion

109

confirmées par IRM pour deux patients31. Sur ces patients, un contraste de déformation allant de 1,6 à 3,2 a été mesuré entre la lésion HIFU et les tissus sains. Lorsque la totalité de la glande a été traitée (29 patients), la déformation moyenne dans la prostate a baissé de 54±26% par rapport à la moyenne avant traitement, mettent ainsi en évidence la rigidification des tissus. Les variations de ce paramètre sont peut-être reliées à l’intensité de la lésion. Enfin une étude de faisabilité de la visualisation de la formation d’une lésion HIFU élémentaire par une nouvelle possibilité d’imagerie, que nous avons dénommée élastographie passive, a été menée. Le principe repose sur l’imagerie des déformations apparentes dans les signaux RF acquis pendant l’élévation de température produite par HIFU. Aucune excitation mécanique externe n’est nécessaire, d’où le terme d’imagerie « passive ». La température induit une expansion thermique, visible en tant que déformation réelle des tissus, ainsi qu’une variation de la vitesse du son, qui se traduit par une déformation apparente se superposant à la déformation réelle. Les équations reliant la déformation mesurée, la déformation réelle et l’effet de la vitesse du son ont été établies. Ensuite le phénomène a été modélisé en se basant sur l’équation de transfert de bio-chaleur (BHTE) pour évaluer la température, sur la dose thermique pour évaluer l’état de coagulation des tissus, et sur les équations de thermoélasticité pour évaluer la dilatation thermique. La vitesse du son a été prise dans la littérature. Enfin les images de déformation apparente et la la taille des lésions simulées ont été comparées aux images obtenues expérimentalement sur 3 échantillons de foie de porc et de bœuf32. Expérimentalement, les élastogrammes ont été acquis à une fréquence de 5 MHz et une cadence de 8 im/s, puis cumulés à partir du début de l’échauffement. L’acquisition des données RF et l’émission HIFU ont été synchronisées afin d’éliminer les interférences entre les deux transducteurs. Les images de maximum d’intercorrélation obtenues par un algorithme d’élastographie standard ont montré une zone de décorrélation qui persista après le refroidissement des tissus mais qui surestimait largement la zone coagulée. Dans les élastogrammes simulés, une zone de dilatation apparente était visible au cœur de la lésion mais sous-estimait la lésion. Dans les élastogrammes expérimentaux la lésion est apparue comme une zone d’expansion apparente, alors qu’une compression apparente était observée dans les tissus voisins. Cette zone correspondait bien avec la taille et la position des lésions simulées, ainsi qu’avec les lésions observées macroscopiquement après découpe des échantillons. Avant coagulation les déformations apparentes simulées et expérimentales concordaient, mais les après coagulation les valeurs expérimentales sont devenues bien plus grandes que celles attendues, atteignant jusqu’à +15% dans la lésion et jusqu’à –7% en dehors. Nous supposons aujourd’hui que la coagulation induit un changement structurel des tissus qui se traduit par une expansion importante qui rend la lésion visible pendant sa formation par cette technique. La poursuite de cette étude inclura la mesure (non ultrasonore) du volume des tissus en fonction de la température, la faisabilité en présence de cavitation, qui sera vraisemblablement un facteur défavorable, puis sur de multiples lésions élémentaires adjacentes. En conclusion, ce travail a mis en évidence que l’élastographie est capable de montrer in vitro l’anatomie zonale de la prostate ainsi que divers types de tumeurs bénignes ou malignes, et de montrer in vivo le cancer et les larges (~2x2 cm2) lésions HIFU. L’élastographie passive permet de visualiser in vitro une lésion HIFU élémentaire pendant sa formation. Il est intéressant de noter que ces résultats ont été obtenus avec une sonde ultrasonore de faible bande passante, et peuvent donc facilement être améliorés en utilisant une sonde large bande 31 Depuis la rédaction de ce chapitre, plus de 61 patients supplémentaires ont été examinés par élastographie et IRM, et la correspondance des surfaces et volumes est en cours d’évaluation. 32 Depuis la rédaction de ce chapitre, la faisabilité a été évaluée sur 15 échantillons supplémentaires.

Conclusion

110

aujourd’hui disponibles sur le marché. Il a en effet été montré dans la littérature que l’augmentation de la bande passante améliore le SNRe ainsi que la résolution. L’étude a aussi permis d’identifier les facteurs capables d’améliorer la qualité, la reproductibilité et l’utilité clinique des images acquises in vivo : une haute cadence d’acquisition, un haut rapport signal-sur-bruit échographique, et l’amélioration de la reproductibilité de la poussée exercée par le ballon, par exemple en utilisant un ballon spécialement conçu pour cette application. Le calcul des élastogrammes en temps réel permettrait au radiologue de pratiquer une exploration interactive d’un nombre suffisant d’images pour couvrir la totalité du volume prostatique, ce qui est susceptible d’affiner le diagnostic.

Conclusion

111

Conclusion (English) Ultrasonic elastography applied to prostate cancer detection and to high-intensity focused ultrasound (HIFU) therapy monitoring was investigated during this thesis. An elastographic imaging system was developed based on a mechanical sector-scan imaging probe covered by a balloon. The balloon was inflated using a coupling liquid in order to induce a radial compression of the surrounding tissues. Compared to a hand-held setup, this system may be somewhat more difficult to use but has three advantages: (1) the main component of the compression is aligned with the ultrasonic beam, minimizing undesirable non-axial displacements, (2) a uniform pressure is applied at the surface, minimizing possible artifacts due to stress nonuniformities, (3) the probe can be attached to a motorized table for a stable positioning, minimizing undesirable probe motion. The setup was compatible with a HIFU machine so that treatment follow-up could be investigated. A slightly modified version of the cross-correlation function was used to ensure that the effective window length remained constant and independent of the measured delay. This method was shown to induce fewer false peaks in time delay estimates than the standard cross-correlation approach, at a slight cost in calculation time. System validation and performance estimation were carried out in gel phantoms designed to mimic transrectal examination. These experiments showed that multi-compression with a displacement step ≤0.45 mm at the balloon interface provided maximal elastographic signal-to-noise ratio (SNRe) compared to single-step compression. Multi-compression also permits the use of high applied strains that would not be measurable with single-step compression. SNRe remained relatively low in phantom experiments, with an experimental maximum of 6 for 6% applied strain, and presented a decay with depth. However it is to be noted that this result was obtained in disadvantageous conditions: the pulse bandwidth B was relatively narrow (B/f0≈40% at 5.5 MHz center frequency), the phantom was not laterally confined, resulting in detrimental lateral motion, and short window length (1 mm) was used. In vitro, prostate elastograms with sharp edges and significant contrast were obtained. The zonal anatomy of the prostate (peripheral zone, transition zone, verumontanum) was clearly visible in the elastograms in all 13 cases. Five malignant tumors and six benign (glandular and/or fibrous) tumors were seen as stiff areas in the elastograms. We noted that benign tumors presented sharp edges, whereas cancer had diffuse margins. This criterion may help differentiate between malignant and benign, but it needs to be evaluated on a larger scale. Small cancers (≤ 0.5 cm3) were not visible in the elastograms. 4 out of 10 cancers were found by blind reading in the peripheral zone, and only 1/6 in the transition zone where cancer detection is usually considered more difficult. Although these performances may at first seem disappointing, it is to be noted that this experiment was designed for a feasibility study and was not optimized for cancer detection: only small pre-compressions were applied (0-2%) so that no advantage was taken from the nonlinear behavior of some malignant tissues, and the low temperature (~10°C) of the embedding gel may have modified the stiffness contrast inside the prostate. Because of these limitations, it is too early to estimate the sensitivity and specificity of prostate elastography.

Conclusion

112

Prostate elastograms were acquired in vivo in 69 patients. In spite of the limited number of imaging planes (3 per patient), prostate cancer confirmed by biopsies and sonograms was found in the elastograms in 5 cases. It was seen as a stiff area in the peripheral zone of the prostate. Using a pre-compression of 15-20%, cancer presented a high strain contrast ratio (3.8 to 9.5) with the surrounding tissues. The results of this study confirmed the feasibility of elastographic detection of prostate cancer in vivo. For patients undergoing HIFU therapy of prostate cancer, HIFU lesions were visible in the elastograms immediately after treatment. It was possible to pause the treatment, to acquire an elastogram to see the lesion, and to resume the treatment. The size and position of the HIFU lesions seen on the elastograms matched the expected lesion, and were confirmed by MRI in two patients. Elastography is therefore likely to become an effective imaging modality for HIFU treatment evaluation and follow-up. One step further was to visualize thermal lesions during their formation. Thermally-induced elastography, called passive elastography in this study because it uses the heat deposition due to HIFU, was able to visualize the formation of an elementary HIFU lesion in vitro. No compressor is required. The technique is similar to ultrasonic temperature estimation; it relies on a combination of tissue expansion with temperature-dependent speed of sound to form an apparent strain elastogram. Tissue temperature could not be determined from the resulting images, but the coagulation area was visible. High tensile strains measured inside the lesion suggested that coagulation induces tissue expansion, probably through a change in the tissue structure. A few considerations on image quality are necessary to conclude. The bandwidth is fundamental to good performances in elastography: axial resolution Ra improves with bandwidth (Ra α B -1), SNRe increases proportionally to B 3/2, and CNRe increases as B 3. In clinical practice today, prostate scanning is performed at 7-9 MHz with 60%-80% bandwidth transducers. With such a probe, axial resolution would improve by a 2.5 factor, SNRe by 4, and CNRe by 16. High acquisition frame rate was critical to acquire high quality elastograms in vivo because of uncontrollable patient and/or organ motion. The detrimental influence of undesirable motion on the average inter-frame cross-correlation, which is an estimator of elastogram quality, was shown even when no compression was applied (natural motion). Our scanner was limited to 8 fps, whereas standard actual scanners operating at 30-50 fps are likely to improve elastogram quality further. In vivo, we sometimes observed that compression may be ineffective (i.e. the balloon may elongate toward the colon rather than inflate) and/or non-uniform (more effective on one side of the prostate than on the other). Real-time feedback was not available; it would have allowed to detect and re-acquire these low-quality elastograms. It would also have allowed the radiologist to focus on suspicious areas. In spite of the important bandwidth and frame rate limitations of the ultrasound scanner, the feasibility of using elastography for prostate cancer detection and for HIFU treatment monitoring was demonstrated.

References

113

References 1. ALAM SK, OPHIR J, KONOFAGOU E. An adaptive strain estimator for elastography. IEEE Trans

Ultrason Ferroelectr Freq Control , 1998, vol.45, n°2, pp.461-472 2. ALAM SK, OPHIR J. The effect of nonlinear signal transformations on bias errors in elastography. IEEE

Trans Ultrason Ferroelectr Freq Control, 2000, vol. 47, n° 1, pp. 297-303 3. ALANIZ A, KALLEL F, HUNGERFORD E, OPHIR J. Variational method for estimating the effects of

continuously varying lenses in HIFU, sonography and sonography-based cross-correlation methods. J Acoust Soc Am, 2002, vol.111, n°1 part 1, pp.468-474

4. ALLARD JF. Propagation of sound in porous media. London: Elsevier Applied Science, 1993, 284 p. 5. BAI J, DING C, LUO J, HE P. Estimation and reduction of decorrelation effect due to tissue lateral

displacement in elastography. IEEE Trans Ultrason Ferroelectr Freq Control, 2002, vol.49, n°5, pp.541-549 6. BAMBER JC, HILL CR. Ultrasonic attenuation and propagation speed in mammalian tissues as a function

of temperature. Ultrasound Med. Biol., 1979, vol.5, pp.149-157. 7. BANGMA CH, HULAND H, SCHRODER FH, VAN CANGH PJ. Early diagnosis and treatment of

localized prostate cancer. European Urology, 2001, vol.40, n°3, pp.C1-C10. 8. BERCOFF J, CHAFFAL S, TANTER M, SANDRIN L, CATHELINE S, FINK M, GENISSON JL,

MEUNIER M. In vivo breast tumor detection using transient elastography. Ultrasound Med Biol, 2003, vol.29, n°10, pp.1387-1396.

9. BILGEN M. Target detectability in acoustic elastography. IEEE Trans Ultrason Ferroelectr Freq Control, 1999, vol.46, n°5, pp.1128-1133

10. BOUCHOUX G, SOUCHON R, LAFON C, CATHIGNOL D, CHAPELON JY. Ultrasound monitoring of thermal lesions by cross- correlation of back-scattered RF lines. In: CHAPELON JY, LAFON C Eds. Proc. 3rd Int. Symp. Therapeutic Ultrasound (ISTU3), June 2003, Lyon, France. Lyon: INSERM, 2003, pp.173-179.

11. BOUCHOUX G. Imagerie ultrasonore de déformation apparente et de corrélation pour la visualisation des lésions thermiques : Rapport de stage de DEA. DEA Images et Systèmes. Lyon : INSERM Unité 556, 2003, 43 p. [in French]

12. BRAY F, SANKILA R, FERLAY J, PARKIN DM. Estimates of cancer incidence and mortality in Europe in 1995. European Journal of Cancer, 2002, vol.38, pp. 99-166.

13. BRUSSEAU E, FROMAGEAU J, FINET G, DELACHARTRE P, VRAY D. Axial strain imaging of intravascular data : Results on polyvinyl alcohol cryogel phantoms and carotid artery . Ultrasound Med Biol, 2001, vol.27, n°12, pp. 1631-1642

14. BUSH NL, RIVENS I, TER HAAR GR, BAMBER JC. Acoustic properties of lesions generated with an ultrasound therapy system. Ultrasound Med Biol., 1993, vol.19, n°9, pp.789-801.

15. CESPEDES I, OPHIR J. Reduction of image noise in elastography. Ultrasonic Imaging, 1993, vol.15, pp.89-102.

16. CESPEDES I, HUANG Y, OPHIR J, SPRATT S. Methods for estimation of subsample time delays of digitized echo signals. Ultrasonic Imaging, 1995, vol. 17, pp. 142-171

17. CESPEDES I, OPHIR J, ALAM SK. The combined effect of signal decorrelation and random noise on the variance of time delay estimation. IEEE Trans Ultrason Ferroelectr Freq Control, 1997, vol.44, n°1, pp.220-225.

18. CESPEDES I, DE KORTE CL, VAN DER STEEN AFW. Echo decorrelation from displacement gradients in elasticity and velocity estimation. IEEE Trans Ultrason Ferroelectr Freq Control, 1999, vol.46, n°4, pp.791-801

19. CHAPELON JY, MARGONARI J, VERNIER F, GORRY F, ECOCHARD R, GELET A. In vivo effects of high-intensity ultrasound on prostatic adenocarcinoma Dunning R3327. Cancer Res., 1992, vol.52, n°22, pp.6353-6357.

20. CHATURDEVI P, INSANA MF, HALL TJ. 2-D companding for noise reduction in strain imaging. IEEE Trans Ultrason Ferroelectr Freq Control, 1998, vol.45, n°1, pp.179-191.

References

114

21. CHRISTENSEN DA. Ultrasonic Bioinstrumentation. New York: John Wiley & Sons Eds, 1988, 256 p. 22. CORNUD F, OYEN R. Role of imaging in the diagnosis and staging of prostatic adenocarcinomas. J

Radiol., 2002, vol.83, n°6 Pt 2, pp.863-882 [in French] 23. CRANDALL SH, DAHL NC. An introduction to the mechanics of solids. New York: McGraw-Hill Book

Company, 1959, 444 p. 24. DE KORTE CL, CESPEDES EI, VAN DER STEEN AFW. Influence of catheter position on estimated

strain in intravascular elastography. IEEE Trans Ultrason Ferroelectr Freq Control, 1999, vol.46, n°3, pp.616-625.

25. DETTI V. Elastographie de la prostate : Rapport de post-doctorat. Lyon : INSERM Unité 556, Décembre 2002, 48 p.

26. EL-GABRY EA, HALPERN EJ, STRUP SE ,GOMELLA LG. Imaging prostate cancer: Current and future applications. Oncology, 2001, vol.15, n°3, pp.325-336

27. ERKAMP RQ, EMELIANOV SY, SKOVORODA AR, CHEN X, O'DONNELL M. Exploiting strain-hardening of tissue to increase contrast in elasticity imaging. Proc IEEE Ultrasonics Symp, 2000, vol.2, pp.1833-1836

28. FATEMI M, GREENLEAF JF. Ultrasound-stimulated vibro-acoustic spectrography. Science, 1998, vol.280, pp.82-85.

29. FOSTER RS, BIHRLE R, SANGHVI NT, FRY FJ, DONOHUE JP. High-intensity focused ultrasound in the treatment of prostatic disease. Eur Urol., 1993, suppl.1, vol.23, pp.29-33

30. FRY WJ, BARNARD JW, FRY EJ,BRENNAN JF. Ultrasonically produced localized selective lesions in the central nervous system. Am J Phys Med, 1955, vol.34, n°3, pp.413-423

31. GARRA BS, CESPEDES EI, OPHIR J, SPRATT SR, ZUURBIER RA, MAGNANT CM, PENNANEN MF. Elastography of breast lesions: initial clinical results. Radiology, 1997, vol.202, pp.79-86.

32. GELET A, CHAPELON JY, BOUVIER R, ROUVIERE O, LYONNET D, DUBERNARD JM. Transrectal high intensity focused ultrasound for the treatment of localized prostate cancer: factors influencing the outcome. Eur Urol., 2001, vol.40, n°2, pp.124-129

33. GOODIER JN. Concentration of stress around spherical and cylindrical inclusions and flaws. Trans ASME (later called J. Applied Mech., Vol. 1), 1933, vol. 55, pp.39-44.

34. GREENLEAF JF, FATEMI M, INSANA M. Selected methods for imaging elastic properties of biological tissues. Annu Rev Biomed Eng, 2003, vol.5, pp.57-78.

35. HALL T, ZHU Y, SPALDING CS. In vivo real-time freehand palpation imaging. Ultrasound Med Biol, 2003, vol.29, n°3, pp.427-435.

36. HARRIS RP, LOHR KN, BECK R, FINK K, GODLEY P, BUNTON A. Screening for Prostate Cancer. File Inventory, Systematic Evidence Review Number 16. October 2002. Agency for Healthcare Research and Quality, Rockville, MD. [online]. Available on: http://www.ahrq.gov/clinic/prev/prostinv.htm (downloaded 01-Aug-2003)

37. HAYDEN HW, MOFFATT WG, WULFF J. The structure and properties of materials. Volume III: Mechanical behavior. New York: John Wiley & Sons Eds, 1965 (247 p)

38. HEDRICK WR, HYKES DL, STARCHMAN DE. Ultrasound Physics and Instrumentation. 3rd edition. Saint Louis, U.S.A.: Mosby Inc., 1995, 382 p.

39. HYNYNEN K, FREUND WR, CLINE HE, CHUNG AH, WATKINS RD, VETRO JP, JOLESZ FA. A clinical, noninvasive, MR imaging-monitored ultrasound surgery method. Radiographics, 1996, vol.16, n°1, pp.185-195

40. INSANA MF, CHATURDEVI P, HALL TJ, BILGEN M. 3-D companding using linear arrays for improved strain imaging. Proc IEEE Ultrasonics Symp, 1997, vol.2, pp.1435-1438.

41. JELLAB H. Modélisation thermoélastique des tissus mous par éléments finis : Mémoire de Maîtrise. Maîtrise ès Sciences Appliquées de Génie Biomédical. Montréal : Ecole Polytechnique de Montréal, 1999, 122 p.

42. KALLEL F, BERTRAND M. Tissue elasticity reconstruction using linear perturbation method. IEEE Trans Med Imaging, 1996, vol.15, n°3, pp.299-313

43. KALLEL F, OPHIR J. A least-squares strain estimator for elastography. Ultrasonic Imaging, 1997, vol.19, pp.195-208

44. KALLEL F, VARGHESE T, OPHIR J, BILGEN M. The nonstationary strain filter in elastography, Part II Lateral and elevational decorrelation. Ultrasound Med Biol, 1997, vol.23, n°9, pp.1357-1370.

References

115

45. KALLEL F, PRICE RE, KONOFAGOU EE, OPHIR J. Elastographic imaging of the normal canine prostate in-vitro. Ultrasonic Imaging, 1999, vol.21, pp.201-215

46. KALLEL F, STAFFORD RJ, PRICE RE, RIGHETTI R, OPHIR J, HAZLE JD. The feasibility of elastographic visualization of HIFU-induced thermal lesions in soft tissues. Ultrasound Med. Biol., 1999, vol.25, n°4, pp.641-647

47. KALLEL F, PRIHODA CD, OPHIR J. Contrast-transfer efficiency for continuously varying tissue moduli: Simulation and phantom validation. Ultrasound Med Biol, 2001, vol.27, n°8, pp.1115-1126.

48. KONOFAGOU E, DUTTA P, OPHIR J, CESPEDES I. Reduction of stress nonuniformities by apodization of compressor displacement in elastography. Ultrasound Med Biol, 1996, vol.22, n°9, pp.1229-1236.

49. KONOFAGOU E, OPHIR J. A new elastographic method for estimation and imaging of lateral displacements, lateral strains, corrected axial strains and Poisson's ratios in tissues. Ultrasound Med Biol, 1998, vol.24, n°8, pp.1183-1199.

50. KONOFAGOU E, VARGHESE T, OPHIR J, ALAM SK. Power spectral strain estimators in elastography. Ultrasound Med Biol, 1999, vol.25, n°7, pp.1115-1129

51. KREMKAU FW. Diagnostic Ultrasound. 6th edition. Philadelphia: W.B. Saunders Company, 2002, 428 p. 52. KROUSKOP TA, DOUGHERTY DR, VINSON FS. A pulsed Doppler ultrasonic system for making

noninvasive measurements of the mechanical properties of soft tissues. Journal of Rehabilitation Research, 1987, vol.24, n°2, pp.1-8.

53. KROUSKOP TA, WHEELER TM, KALLEL F, GARRA BS, HALL T. Elastic moduli of breast and prostate tissues under compression. Ultrason Imaging, 1998, vol.20, pp.260-274

54. LAFON C, PRAT F, CHAPELON JY, GORRY F, MARGONARI J, THEILLERE Y, CATHIGNOL D. Cylindrical thermal coagulation necrosis using an interstitial applicator with a plane ultrasonic transducer: in vitro and in vivo experiments versus computer simulations. Int. J. Hyperthermia, 2000, vol. 16, n° 6, pp. 508-522.

55. LE FLOCH C, FINK M. Ultrasonic mapping of temperature in hyperthermia: The thermal lens effect. Proc IEEE Ultrasonics Symp, 1997, vol.2, pp.1301-1304.

56. LEMMON EW, MCLINDEN MO, FRIEND DG. Thermophysical Properties of Fluid Systems. In: P.J. Linstrom and W.G. Mallard Eds, NIST Chemistry WebBook, NIST Standard Reference Database Number 69, Gaithersburg MD, 2003 (http://webbook.nist.gov)

57. LERNER RM, HUANG SR, PARKER KJ. Sonoelasticity images derived from ultrasound signals in mechanically vibrated tissues. Ultrasound Med Biol, 1990, vol.16, n°3, pp.231-239

58. LIZZI FL, MURATORE R, DENG CX, KETTERLING JA, ALAM SK, MIKAELIAN S, KALISZ A. Radiation-force technique to monitor lesions during ultrasonic therapy. Ultrasound Med Biol, 2003, vol.29, n°11, pp.1593-1605.

59. LORENZ A, SOMMERFELD HJ, GARCIA-SCHURMANN M, PHILIPPOU S, SENGE T, ERMERT H. A new system for the acquisition of ultrasonic multicompression strain images of the human prostate in vivo. IEEE Trans Ultrason Ferroelectr Freq Control, 1999, vol.46, n°5, pp.1147-1153

60. LU J, YING H, SUN Z, MOTAMEDI M, BELL B, SHEPPARD LC. In vitro measurement of speed of sound during coagulate tissue heating. IEEE Ultrasonics Symp. Proc., 1996, vol.2, pp.1299-1302.

61. LYNN JG, ZWEMMER RL, CHICK AJ et al. A new method for the generation and use of focused ultrasound in experimental biology. J Gen Physiol, 1942, vol.26, pp.179-193

62. MADERBACHER S, PEDEVILLA M, VINGERS L, SUSANI M, MARBERGER M. Effect of high-intensity focused ultrasound on human prostate cancer in vivo. Cancer Res., 1995, vol.55, n°15, pp.3346-3351.

63. MAASS-MORENO R, DAMIANOU CA. Noninvasive temperature estimation in tissue via ultrasound echo-shifts. Part I. Analytical model., J. Acoust. Soc. Am., 1996, vol.100, n°4, pp.2514-2521.

64. MAASS-MORENO R, DAMIANOU CA, SANGHVI NT. Noninvasive temperature estimation in tissue via ultrasound echo-shifts. Part II. In vitro study., J. Acoust. Soc. Am., 1996, vol.100, n°4, pp.2522-2530.

65. MC NEAL JE. The zonal anatomy of the prostate. Prostate, 1981, vol.2, n°1, pp.35-49. 66. MC NEAL JE. Prostate. In: STERNBERG SS. Histology for Pathologists. New York: Raven Press, 1992,

pp.749-763. 67. MENEGOZ F, BLACK RJ, ARVEUX P, MAGNE V, FERLAY J, BUEMI A, CARLI PM, CHAPELAIN

G, FAIVRE J, GIGNOUX M, GROSCLAUDE P, MACE-LESEC'H J, RAVERDY N, SCHAFFER P. Cancer incidence and mortality in France in 1975-95. Eur J Cancer Prev., 1997, vol.6, n°5, pp.442-466.

68. MEUNIER J, BERTRAND M. Ultrasonic texture motion analysis: theory and simulation. IEEE Trans Medical Imaging, 1995, vol.14, n°2, pp.293-300

References

116

69. MUSHKELISHVILI NI. Solution for regions bounded by a circle. Chapter 9. In: Some basic problems of the mathematical theory of elasticity, 4th ed. Groningen: P. Noordhoff Ltd., 1963.

70. NIGHTINGALE KR, PALMERI ML, NIGHTINGALE RW, TRAHEY GE. On the feasibility of remote palpation using acoustic radiation force. J. Acoust. Soc. Am., 2001, vol.110, n°1, pp.625-634.

71. NOWINSKI JL. Theory of thermoelasticity with applications. Groningen: Sijthoff & Noordhoff Int. Pub., 1978, 852 p.

72. O’DONNELL M, SKOVORODA AR, SHAPO BM, EMELIANOV SY. Internal displacement and strain imaging using ultrasonic speckle tracking. IEEE Trans Ultrason Ferroelectr Freq Control, 1994, vol.41, n°3, pp.314-325

73. OPHIR J, CESPEDES I, PONNEKANTI H, YAZDI Y, LI X. Elastography : a quantitative method for imaging the elasticity of biological tissues. Ultrason Imaging, 1991, vol.13, pp.111-134.

74. OPHIR J, CÉSPEDES I, GARRA B, PONNEKANTI H, HUANG Y, MAKLAD N. Elastography: Ultrasonic imaging of tissue strain and elastic modulus in vivo. Eur J Ultrasound, 1996, vol.3, pp.49-70

75. OPHIR J, CÉSPEDES I, MAKLAD NF, PONNEKANTI H. Elastography: A method for imaging the elastic properties of tissue in-vivo. Chapter 7. In: TANAKA M et al. Ultrasonic Tissue Characterization. Tokyo: Springer Verlag Publ, 1996, pp. 95-123

76. OPHIR J, ALAM SK, GARRA B, KALLEL F, KONOFAGOU E, KROUSKOP T, VARGHESE T. Elastography: ultrasonic estimation and imaging of the elastic properties of tissues. Proc Inst Mech Eng [H], 1999, vol.213, pp.203-233

77. PENNES HH. Analysis of tissue and arterial blood temperature in resting forearm. J. Appl. Physiol., 1948, vol. 1, pp. 93-122.

78. PESAVENTO A, PERREY C, KRUEGER M, ERMERT H. A time-efficient and accurate strain estimation concept for ultrasonic elastography using iterative phase zero estimation. IEEE Trans Ultrason Ferroelectr Freq Control, 1999, vol.46, n°5, pp.1057-1067

79. PESAVENTO A, LORENZ A. Real time strain imaging and in-vivo applications in prostate cancer. Proc IEEE Ultrasonics Symp, 2001, vol.2, pp.1647-1652.

80. PONNEKANTI H, OPHIR J, CESPEDES I. Axial stress distribution between coaxial compressors in elastography: an analytical model. Ultrasound Med Biol, 1992, vol.18, n°8, pp.667-673.

81. QUINLAN DM, PARTIN AW, WALSH PC. Can aggressive prostatic carcinomas be identified and can their natural history be altered by treatment ? Urology, 1995, vol.46, pp.77-82.

82. RIBAULT M, CHAPELON JY, CATHIGNOL D, GELET A. Differential attenuation imaging for the characterization of high intensity focused ultrasound lesions. Ultrason Imaging, 1998, vol.20, n°3, pp.160-177.

83. RIBAULT M. Caractérisation ultrasonore de lésions induites par ultrasons focalisés de haute intensité. PhD Thesis. Lyon : Université Lyon 1 Claude Bernard, 1999, 147 p. [in French]

84. RIGHETTI R, KALLEL F, STAFFORD RJ, PRICE RE, KROUSKOP TA, HAZLE JD, OPHIR J. Elastographic characterization of HIFU-induced lesions in canine livers. Ultrasound Med. Biol., 1999, vol.25, n°7, pp.1099-1113.

85. RIGHETTI R, OPHIR J, KTONAS P. Axial resolution in elastography. Ultrasound Med Biol, 2002, vol. 28, n°1, pp.101-113.

86. RIGHETTI R, SRINIVASAN S, OPHIR J. Lateral resolution in elastography. Ultrasound Med Biol, 2003, vol. 29, n°5, pp.695-704.

87. ROSSIGNOL G, RINALDI R. Imagerie de diffuseurs multiples dans les tissus biologiques par élastographie : Rapport de projet de fin d’études. Lyon : INSERM Unité 556, Février 2002, 45 p. [in French]

88. ROSSIGNOL G. Etude de faisabilité de la détection par élastographie des grains de Curiethérapie : Rapport de DEA. DEA de Génie Mécanique. Lyon : INSERM Unité 556, Septembre 2002, 92 p. [in French]

89. ROUVIERE O, LYONNET D, RAUDRANT A, COLIN-PANGAUD C, CHAPELON JY, BOUVIER R, DUBERNARD JM, GELET A. MRI appearance of prostate following transrectal HIFU ablation of localized cancer. Eur Urol, 2001, vol.40, pp.265-274.

90. RUBENS DJ, HADLEY MA, ALAM SK, GAO L, MAYER RD, PARKER KJ. Sonoelasticity imaging of prostate cancer: in vitro results. Radiology, 1995, vol.195, pp.379-383.

91. SANDRIN L, CATHELINE S, TANTER M, HENNEQUIN X, FINK M. Time resolved pulsed elastography with ultrafast ultrasonic imaging. Ultrason Imaging, 1999, vol.21, n°4, pp.259-272.

92. SAPARETO SA, DEWEY WC. Thermal dose determination in cancer therapy. Int. J. Radiat. Oncol. Biol. Phys., 1984, vol. 10, pp. 787-800.

References

117

93. SEDELAAR JPM, AARNINK RG, VAN LEENDERS GJLH, BEERLAGE HP, DEBRUYNE FMJ, WIJKSTRA H, DE LA ROSETTE JJMCH. The application of three-dimensional contrast-enhanced ultrasound to measure volume of affected tissue after HIFU treatment for localized prostate cancer. Eur Urol., 1999, vol.37, pp.559-568.

94. SIMON C, VANBAREN P, EBBINI E. Quantitative analysis and applications of non-invasive temperature estimation using diagnostic ultrasound. Proc IEEE Ultrasonics Symp, 1997, vol.2, pp.1319-1322.

95. SHAPO BM, CROWE JR, SKOVORODA AR, EBERLE MJ, COHN NA, O’DONNELL M. Ultrasonic displacement and strain imaging of coronary arteries with a catheter array. Proc IEEE Ultrasonics Symp, 1995, vol.2, pp.1511-1514

96. SHI X, MARTIN RW, ROUSEFF D, VAEZY S, CRUM LA. Detection of high-intensity focused ultrasound liver lesions using dynamic elastometry. Ultrasonic Imaging, 1999, vol.21, pp.107-126

97. SOBIN LH, WITTEKIND C. TNM Classification of malignant tumors, 6th edition. New York: Wiley, 2002, 202 p.

98. SOUCHON R, CHAPELON JY, BERTRAND M, KALLEL F, OPHIR J. Feasibility of monitoring HIFU prostate cancer therapy using elastography. Proc SPIE Medical Imaging Symp, 2001, vol.4325, pp.392-399.

99. SOUCHON R, SOUALMI L, BERTRAND M, CHAPELON JY, KALLEL F, OPHIR J. Ultrasonic elastography using sector scan imaging and a radial compression. Ultrasonics, 2002, vol.40, n°1-8, pp.867-871.

100. SOUCHON R, GELET A, ROUVIERE O, CHAPELON JY, KALLEL F, OPHIR J. Elastographie de la prostate in vivo, et application aux ultrasons focalisés de haute intensité (HIFU): Résultats initiaux. Actes 6ème Congrès Français d'Acoustique (CFA), April 2002, Lille. pp.30-33 [in French].

101. SOUCHON R, ROUVIERE O, GELET A, DETTI V, SRINIVASAN S, OPHIR J, CHAPELON JY. Visualisation of HIFU lesions using elastography of the human prostate in vivo: Preliminary results. Ultrasound Med Biol, 2003, vol.29, n°7, pp.1007-1015.

102. SOUCHON R, DETTI V, CURIEL L, OPHIR J, CHAPELON JY. L’élastographie pour guider la thérapie par ultrasons focalisés du cancer de la prostate in vivo. Proc 12ème Forum Jeunes Chercheurs, May 2003, Nantes. pp.174-175 [in French].

103. SOUCHON R, DETTI V, GELET A, OPHIR J, CHAPELON JY. Visualisation of high intensity focused ultrasound (HIFU)-induced lesions in the human prostate in vivo using elastography. In: CHAPELON JY, LAFON C Eds. Proc. 3rd Int. Symp. Therapeutic Ultrasound (ISTU3), June 2003, Lyon, France. Lyon: INSERM, 2003, pp. 211-216.

104. SOUCHON R, CURIEL L, ROUVIERE O, GELET A, OPHIR J, CHAPELON JY. Comparison of US elastograms and MR images of thermal lesions in the prostate. Proc World Congress Ultrasonics (WCU), September 2003, Paris. pp. 385-388.

105. SOUCHON R, HERVIEU V, GELET A, OPHIR J, CHAPELON JY. Human prostate elastography : an in vitro study. Proc IEEE Ultrasonics Symp 2003, vol.2, pp.1251-1253.

106. SOUCHON R, BOUCHOUX G, MACIEJKO E, LAFON C, BERTRAND M, CHAPELON JY. Monitoring of the formation of high intensity focused ultrasound (HIFU) lesions with passive elastography. Accepted in Ultrasound Med Biol, March 2004.

107. SRINIVASAN S, OPHIR J, ALAM SK. Elastographic imaging using staggered strain estimates. Ultrasonic Imaging, 2002, vol.24, pp.229-245.

108. SRINIVASAN S, KALLEL F, SOUCHON R, OPHIR J. Analysis of an adaptive strain estimation technique in elastography. Ultrasonic Imaging, 2002, vol.24, n°2, pp.109-118.

109. SRINIVASAN S, OPHIR J. A zero-crossing strain estimator for elastography. Ultrasound Med Biol, 2003, vol.29, n°2, pp.227-238.

110. SRINIVASAN S, RIGHETTI R, OPHIR J. Trade-offs between the axial resolution and the signal-to-noise ratio. Ultrasound Med Biol, 2003, vol.29, n°6, pp.847-866.

111. SUN Z, YING H, LU J, BELL B, COWAN DF, MOTAMEDI M. Automatic ultrasound determination of thermal coagulation front during laser tissue heating. IEEE Trans Ultrason Ferroelectr Freq Control, 1999, vol.46, n°5, pp.1134-1146

112. SUN Z, YING H, LU J. A non-invasive cross-correlation ultrasound technique for detecting spatial profile of laser-induced coagulation damage : an in-vitro study. IEEE Trans Biomed Eng, 2001, vol.48, n°2, pp.223-229.

113. TECHAVIPOO U, VARGHESE T, ZAGZEBSKI JA, STILES T, FRANK G. Temperature dependence of ultrasonic propagation speed and attenuation in canine tissue. Ultrasonic Imaging, 2002, vol.24, pp.246-260.

114. VANBAREN P, SIMON C, SEIP R, SOLF T, CAIN CA, EBBINI ES. Image-guided phased array system for ultrasound thermotherapy. Proc IEEE Ultrasonics Symp, 1996, vol.2, pp.1269-1272.

References

118

115. VARGA OH. Chapter II: Stress and elastic behavior in materials. In: Stress-strain behavior of elastic materials. Bristol: Interscience Publishers, 1966, 189 p.

116. VARGHESE T, OPHIR J, CESPEDES I. Noise reduction in elastograms using temporal stretching with multicompression averaging. Ultrasound Med Biol, 1996, vol.22, n°8, pp.1043-1052.

117. VARGHESE T, OPHIR J. A theoretical framework for performance characterization of elastography: The strain filter. IEEE Trans Ultrason Ferroelectr Freq Control, 1997, vol.44, n°1, pp.164-172.

118. VARGHESE T, KONOFAGOU EE, OPHIR J, ALAM SK, BILGEN M. Direct strain estimation in elastography using spectral cross-correlation. Ultrasound Med Biol, 2000, vol.26, n°9, pp.1525-1537

119. VARGHESE T, OPHIR J, KROUSKOP TA. Nonlinear stress-strain relationship in tissue and their effect on the contrast-to-noise ratio in elastograms. Ultrasound Med Biol, 2000, vol.26, n°5, pp.839-851.

120. WALKER WF, TRAHEY GR. A fundamental limit on delay estimation using partially correlated speckle signals. IEEE Trans Ultrason Ferroelectr Freq Control, 1995, vol.42, n°2, pp.301-308

121. WANG CT. Applied elasticity. New York : McGraw-Hill Book Company, 1953, 357 p. 122. WATERHOUSE RL, RESNICK MI. The use of transrectal prostatic ultrasonography in the evaluation of

patients with prostatic carcinoma. J Urol., 1989, vol.141, n°2, pp.233-239. 123. WEIDENSTEINER C, QUESSON B, CAIRE-GANA B, KERIOUI N, RULLIER A, TRILLAUD H,

MOONEN CT. Real-time MR temperature mapping of rabbit liver in vivo during thermal ablation. Magn Reson Med., 2003, vol.50, n°2, pp.322-330

124. WEINSTEIN E, WEISS AJ. Fundamental limitations in passive time delay estimation - Part II: Wide-band systems. IEEE Trans Acoustics Speech Signal Processing, 1984, vol.32, n°5, pp.1064-1078

125. WELLS PNT. Biomedical Ultrasonics. London: Academic Press, 1977, 635 p. 126. WORTHINGTON AE, TRACHTENBERG J, SHERAR MD. Ultrasound properties of human prostate

tissue during heating. Ultrasound Med Biol, 2002, vol.28, n°10, pp.1311-1318 127. WU T, FELMLEE JP, GREENLEAF JF, RIEDERER SJ, EHMAN RL. Assessment of thermal tissue

ablation with MR elastography. Magnetic Resonance in Medicine, 2001, vol.45, n°1, pp.80-87. 128. YAMAKOSHI Y, SATO J, SATO T. Ultrasonic imaging of internal vibration of soft tissue under forced

vibration. IEEE Trans Ultrason Ferroelectr Freq Control, 1990, vol.37, n°2, pp.45-53. 129. ZHU Y, HALL TJ. A modified block matching method for real-time freehand strain imaging. Ultrasonic

Imaging, 2002, vol. 24, n°3, pp.161-176

Annexes

Annexes Annexe A: List of the publications of the author

Articles 1. OPHIR J, ALAM SK, GARRA B, KALLEL F, KONOFAGOU E, KROUSKOP T, MERRITT CRB,

RIGHETTI R, SOUCHON R, SRINIVASAN S, VARGHESE T. Elastography: Imaging the elastic properties of soft tissues with ultrasound. Journal of Medical Ultrasonics, 2002, vol.29, pp.155-171.

2. SOUCHON R, ROUVIERE O, GELET A, DETTI V, SRINIVASAN S, OPHIR J, CHAPELON JY. Visualisation of HIFU lesions using elastography of the human prostate in vivo: Preliminary results. Ultrasound Med Biol, 2003, vol.29, n°7, pp.1007-1015.

3. SOUCHON R, BOUCHOUX G, MACIEJKO E, LAFON C, BERTRAND M, CHAPELON JY. Monitoring of the formation of high intensity focused ultrasound (HIFU) lesions with passive elastography. Accepted in Ultrasound Med Biol, March 2004.

4. SRINIVASAN S, KALLEL F, SOUCHON R, OPHIR J. Analysis of an adaptive strain estimation technique in elastography. Ultrasonic Imaging, 2002, vol.24, n°2, pp.109-118.

Proceedings 1. BOUCHOUX G, SOUCHON R, LAFON C, CATHIGNOL D, CHAPELON JY. Ultrasound monitoring of

thermal lesions by cross- correlation of back-scattered RF lines. In: CHAPELON JY, LAFON C Eds. Proc. 3rd Int. Symp. Therapeutic Ultrasound (ISTU3), June 2003, Lyon, France. Lyon: INSERM, 2003, pp. 173-179.

2. SOUCHON R, CHAPELON JY, BERTRAND M, KALLEL F, OPHIR J. Feasibility of monitoring HIFU prostate cancer therapy using elastography. Proc SPIE Medical Imaging Symp, 2001, vol.4325, pp.392-399.

3. SOUCHON R, SOUALMI L, BERTRAND M, CHAPELON JY, KALLEL F, OPHIR J. Ultrasonic elastography using sector scan imaging and a radial compression. Ultrasonics, 2002, vol.40, n°1-8, pp.867-871.

4. SOUCHON R, GELET A, ROUVIERE O, CHAPELON JY, KALLEL F, OPHIR J. Elastographie de la prostate in vivo, et application aux ultrasons focalisés de haute intensité (HIFU): Résultats initiaux. Actes 6ème Congrès Français d'Acoustique (CFA), April 2002, Lille. pp.30-33 [in French].

5. SOUCHON R, DETTI V, CURIEL L, OPHIR J, CHAPELON JY. L’élastographie pour guider la thérapie par ultrasons focalisés du cancer de la prostate in vivo. Proc 12ème Forum Jeunes Chercheurs, May 2003, Nantes. pp.174-175 [in French].

6. SOUCHON R, DETTI V, GELET A, OPHIR J, CHAPELON JY. Visualisation of high intensity focused ultrasound (HIFU)-induced lesions in the human prostate in vivo using elastography. In: CHAPELON JY, LAFON C Eds. Proc. 3rd Int. Symp. Therapeutic Ultrasound (ISTU3), June 2003, Lyon, France. Lyon: INSERM, 2003, pp. 211-216.

7. SOUCHON R, CURIEL L, ROUVIERE O, GELET A, OPHIR J, CHAPELON JY. Comparison of US elastograms and MR images of thermal lesions in the prostate. Proc World Congress Ultrasonics (WCU), September 2003, Paris. pp. 385-388.

8. SOUCHON R, HERVIEU V, GELET A, OPHIR J, CHAPELON JY. Human prostate elastography : an in vitro study. Proc IEEE Ultrasonics Symp 2003, vol.2, pp.1251-1253.

Abstracts 1. BERTRAND M, CHAPELON JY, SOUALMI L, GELET A, SOUCHON R. Proposed system for prostate

elastography during HIFU treatment. Ultrasound Med Biol, 2000, vol. 26, Suppl.. 2, p.A229. 2. BOUCHOUX G, SOUCHON R, LAFON C, CATHIGNOL D, CHAPELON JY. HIFU lesion formation

monitored by passive elastography in vitro. Proc. 2nd Conf. on Ultrasonic Measurement and Imaging of Tissue Elasticity, October 2003, Corpus Christi TX (USA), p.54.

Annexes

3. HE Z, GRIMM S, TRAINER TD, KALOF A, SOUCHON R, OPHIR J, WEAR KA, WAGNER RF, HUSTON D, WEISS LJ, GARRA BS. Integrating elastography with ultrasound backscatter and image texture features for prostate cancer detection: pathology––US data registration method and results. Ultrasound Med Biol, 2003, vol.29, n°5, Suppl.1, pp.S186-S187.

4. MAURICE R, SOUCHON R, BERTRAND M, CHAPELON JY. Jitter correction of digitized A-lines to improve speckle motion estimation. Ultrasonic Imaging, 2001, vol. 22, p.248.

5. ROSSIGNOL G, SOUCHON R, ANGEL Y, CHAPELON J-Y. Using elastography to detect brachytherapy seeds: a feasibility study. Proc 1st Conf. on Ultrasonic Measurement and Imaging of Tissue Elasticity, October 2002, Niagara Falls (Canada), p.24.

6. SOUALMI L, SOUCHON R, BERTRAND M, CHAPELON JY. Transrectal prostate elastography: numerical and experimental studies. Ultrasonic Imaging, 2001, vol. 22, pp.247-248.

7. SOUCHON R, CHAPELON JY, ROUVIÈRE O, GELET A, KALLEL F, OPHIR J. In-vivo elastography of the prostate and HIFU applications : initial results. Ultrasonic Imaging, 2001, vol.22, pp.244-245.

8. SOUCHON R, OPHIR, SRINIVASAN S, CHAPELON JY. Depth-dependent strain filter in radial compression elastography: Experimental measurements. Proc. 1st Conf on Ultrasonic Measurement and Imaging of Tissue Elasticity, October 2002, Niagara Falls (Canada), p.25.

9. SOUCHON R, OPHIR J, SRINIVASAN S, CHAPELON JY. Prostate elastography: causes of decorrelation in vivo. Proc. 1st Conf on Ultrasonic Measurement and Imaging of Tissue Elasticity, October 2002, Niagara Falls (Canada), p.86.

10. SOUCHON R, DETTI V, SRINIVASAN S, CHAPELON JY. A fast acquisition system for radial compression elastography of the prostate in-vivo. Proc. 1st Conf. on Ultrasonic Measurement and Imaging of Tissue Elasticity, October 2002, Niagara Falls (Canada), p.57.

11. SOUCHON R, BERA JC, POUSSE A, CHAPELON JY. Improvement in elastographic signal-to-noise ratio and resolution using coded excitation in elastography. Submitted to 75th Anniversary Meeting (147th Meeting) of the Acoustical Society of America, May 2004, New York.

12. SRINIVASAN S, SOUCHON R, OPHIR J. In-vivo results using a zero-crossing strain estimator for elastography. Proc 1st Conf. on Ultrasonic Measurement and Imaging of Tissue Elasticity, October 2002, Niagara Falls (Canada), p.91.

Publications anterior to this thesis (refereed articles only) 1. CURIEL L, CHAVRIER F, SOUCHON R, BIRER A, CHAPELON JY. 1.5-D high intensity focused

ultrasound array for non-invasive prostate cancer surgery. IEEE Trans Ultrason Ferroelectr Freq Control, 2002, vol.49, n°2, pp.231-242

2. KHOKHLOVA V, SOUCHON R, TAVAKKOLI J, SAPOZHNIKOV O, CATHIGNOL D. Numerical modeling of finite-amplitude sound beams: shock formation in the near field of a CW plane piston source. J. Acoust. Soc. Am., 2001, vol.110, n°1, pp.95-108

3. CHAPELON JY, RIBAULT M, BIRER A, VERNIER F, SOUCHON R, GELET A. Treatment of localised prostate cancer with transrectal high intensity focused ultrasound. Eur J Ultrasound, 1999, vol.9, pp.31-38

4. GELET A, CHAPELON JY, BOUVIER R, SOUCHON R, PANGAUD C. Treatment of prostate cancer with "Ablatherm" (transrectal focused ultrasound). Eur Urol, 1999, vol.33, suppl. 1, pp.173-173

5. TAVAKKOLI J, CATHIGNOL D, SOUCHON R, SAPOZHNIKOV O. Modeling of pulsed finite-amplitude focused sound beams in time domain. J Acoust Soc Am, 1998, vol.104, n°4, pp. 2061-2072

6. GELET A, CHAPELON JY, BOUVIER R, PANGAUD C, SOUCHON R, BLANC E, CATHIGNOL D, DUBERNARD JM. Résultats préliminaires du traitement de 44 patients porteurs de cancers localisés de la prostate par ultrasons focalisés transrectaux. Progrès en Urologie, 1998, vol.8, pp.68-77 [in French]

7. CHAPELON JY, GELET A, SOUCHON R, PANGAUD C, BLANC E. Therapy using ultrasound: application to localised prostate cancer. Journal d’Echographie et de Médecine par Ultrasons, 1998, vol.19, n°2-3, pp.260-264

8. PRAT F, CHAPELON JY, AREFIEV A, CATHIGNOL D, SOUCHON R, THEILLERE Y. High-intensity focused ultrasound transducers suitable for endoscopy: feasibility study in rabbits. Gastrointestinal Endoscopy, 1997, vol.46, n°4, pp.348-351

9. GELET A, CHAPELON JY, BOUVIER R, SOUCHON R, PANGAUD C, ABDELRAHIM AF, CATHIGNOL D, DUBERNARD JM. Treatment of prostate cancer with transrectal focused ultrasound: early clinical experience. Eur Urol, 1996, vol.29, n°2, pp.174-183

10. GELET A, CHAPELON JY, BOUVIER R, PANGAUD C, SOUCHON R, CATHIGNOL D, DUBERNARD JM. Traitement des cancers localisés de la prostate par ultrasons focalisés émis par voie endorectale (projet Ablatherm). Progrès en Urologie, 1995, vol.5, supp.1, pp.59A-59A [in French]

Annexes

11. GELET A, CHAPELON JY, MARGONARI J, THEILLERE Y, GORRY F, SOUCHON R, BOUVIER T. Termo-ablacion prostatica por ultrasonido transrectal focalizado. Urologia Panamericana, 1994, vol.6, n°3, pp.14-18 [in Spanish]

12. GELET A, CHAPELON JY, MARGONARI J, THEILLERE Y, GORRY F, SOUCHON R, BOUVIER R. High-intensity focused ultrasound experimentation on human benign prostatic hypertrophy. Eur Urol, 1993, vol.23, supp.1, pp.44-47

Annexe B: Teaching activities

1. Elastography. DEA Images et Systèmes & Ingénierie Médicale et Biologique. Lyon, 2001-2004.

2. Elastography. European School of Medical Physics (ESMP). Archamps, 2001-2004. 3. Tissue characterization. DEA Images et Systèmes & Ingénierie Médicale et Biologique.

Lyon, 2002-2004. 4. Medical Imaging with Ultrasound. Ecole Supérieure de Mécanique de Marseille (ESM2).

Marseille, 2003.

FOLIO ADMINISTRATIF

THESE SOUTENUE DEVANT L'INSTITUT NATIONAL DES SCIENCES APPLIQUEES DE LYON

NOM : SOUCHON DATE de SOUTENANCE : 15 mars 2004 (avec précision du nom de jeune fille, le cas échéant) Prénoms : Rémi TITRE : Application de l’élastographie à l’imagerie du cancer de la prostate et à sa thérapie par ultrasons focalisés NATURE : Doctorat Numéro d'ordre : 04 ISAL 0018 Ecole doctorale : Electronique, Electrotechnique et Automatique (EEA) Spécialité : Images et Systèmes Cote B.I.U. - Lyon : T 50/210/19 / et bis CLASSE : RESUME : Un système d’imagerie de la prostate par élastographie ultrasonore a été développé. Il utilise un ballon rempli d’un liquide de couplage ultrasonore pour comprimer la prostate. La faisabilité de visualiser l’anatomie prostatique ainsi que des tumeurs bénignes et malignes de la prostate in vitro est démontrée. L’importance de la rapidité de l’acquisition pour obtenir des images de qualité in vivo est ensuite mise en évidence. Il est alors montré que ce système permet de détecter le cancer et de visualiser les effets de la thérapie par ultrasons focalisés de haute intensité (HIFU) in vivo. Enfin il est montré in vitro qu’il est possible de visualiser la formation d’une lésion HIFU élémentaire en utilisant seulement l’élévation de température pour former l’image. MOTS-CLES : Elastographie, prostate, cancer, HIFU, ultrasons Laboratoire (s) de recherches : INSERM Unité 556, LYON Directeur de thèse: CHAPELON Jean-Yves Président de jury : Composition du jury : BRIDAL Lori, Rapporteuse CHAPELON Jean-Yves, Directeur de thèse FINK Mathias GELbert OPHIR Jonathan REMENIERAS Jean-Pierre, Rapporteur