Automated Quantification in Echocardiography · 2019-05-25 · cardiac magnetic resonance imaging....

J A C C : C A R D I O V A S C U L A R I M A G I N G VO L . 1 2 , N O . 6 , 2 0 1 9

ª 2 0 1 9 B Y T H E AM E R I C A N C O L L E G E O F C A R D I O L O G Y F O UN DA T I O N

P U B L I S H E D B Y E L S E V I E R

STATE-OF-THE-ART REVIEW

Automated Quantification inEchocardiography
Mark T. Nolan, MBBS, Paaladinesh Thavendiranathan, MD, SM
ABSTRACT

ISS

Fro

Un

of

thi

Ma

Echocardiography remains the predominant modality for cardiac imaging. Recent technological advances have led to the

availability of new echocardiographic techniques for more accurate quantification of volumes, function, myocardial

mechanics, and valvular heart disease. However, in our opinion, the real-world clinical uptake of these techniques has

been poor due to limited awareness and familiarity, associated time burden, and issues of variability. Automation rep-

resents a potential solution to these issues and has already made routine myocardial strain measurements and 2- and

3-dimensional left ventricular ejection fraction measurements a clinical reality. Further enhancements in automation and

data in understudied populations are likely to assist in the uptake of these new quantitative echocardiographic techniques

in routine clinical practice. This review discusses current automated quantification techniques in echocardiography and

their limitations and describes how these techniques can be incorporated into echocardiography laboratories.

(J Am Coll Cardiol Img 2019;12:1073–92) © 2019 by the American College of Cardiology Foundation.

E chocardiography is the primary method fornoninvasive imaging of the heart with itscost comprising 11% of all U.S. Medicare

spending on imaging services in 2010 (1). Approxi-mately 20% of Medicare enrollees receive at least 1echocardiogram annually, accounting for 7.1 millionechocardiograms in the United States each year (2).Due to the progress in echocardiography over time(Central Illustration) and the availability of multiplenew quantitative parameters, the routine clinical ex-amination has become longer. A good example ofthis scenario is the patient undergoing cancer therapyin which the examination has evolved from bi-planeleft ventricular ejection fraction (LVEF) only to nowincluding 3-dimensional (3D) LVEF, diastolic functionassessment, and myocardial strain analysis. Thisapproach has added stress to echocardiography labo-ratories, increased time for studies to be performedand reported, and potentially results in delays in

N 1936-878X/$36.00

m the Division of Cardiology, Peter Munk Cardiac Centre, Division o

iversity Health Network, Toronto, Ontario, Canada. Dr. Thavendiranatha

Health Research New Investigator Award. Dr. Nolan has reported that

s paper to disclose.

nuscript received May 7, 2018; revised manuscript received November 25

diagnosis. Time restraints combined with multiplemanual measurements affect accuracy and introducea source of variability between readers and successivetests (3,4).

Automation of echocardiographic measurementshas the ability to change the workflow in echocardi-ography laboratories. Potential benefits of automa-tion include time and cost savings associated withstreamlining of image acquisition, rapid analysis andreporting, and greater accuracy and reproducibility ofmeasurements. The present review discusses some ofthe promising advances with automated quantifica-tion in echocardiography, their limitations, and howthese methods could be incorporated into echocar-diography laboratories. We have used the term “semi-automated” to describe quantification methodswhere the user needs to define regions of interestor anatomic landmarks, but the measurement processis automated and “fully automated” to describe

https://doi.org/10.1016/j.jcmg.2018.11.038

f Cardiology and Department of Medical Imaging,

n (147814) is supported by the Canadian Institutes

he has no relationships relevant to the contents of

, 2018, accepted November 29, 2018.

https://doi.org/10.1016/j.jcmg.2018.11.038

http://crossmark.crossref.org/dialog/?doi=10.1016/j.jcmg.2018.11.038&domain=pdf

CENTRAL ILLUSTR

PipeLinedImage

ProcessingEngine(PIPE)

Col

AcousticQuantification

119941985

Nolan, M.T. et al. J Am Coll

Description of the progression in

LVEF ¼ left ventricular ejection

ABBR EV I A T I ON S

AND ACRONYMS

2D = 2-dimensional

3D = 3-dimensional

CMR = cardiovascular magnetic

resonance

GLS = global longitudinal

strain

ICC = intraclass correlation

coefficient

LVEF = left ventricular ejection

fraction

MDCT = multi-detector

computed tomography

PISA = proximal isovelocity

surface area

TEE = transesophageal

echocardiography

Nolan and Thavendiranathan J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 2 , N O . 6 , 2 0 1 9

Quantification in Echocardiography J U N E 2 0 1 9 : 1 0 7 3 – 9 2

1074

quantification methods where the softwareautomatically identifies the landmarks andinitiates measurements. However, even inthe latter scenario, manual adjustments canbe made by the user if necessary.

CHAMBER AND

DOPPLER QUANTIFICATION

A complete echocardiography study involvesmultiple measurements of chamber di-mensions and Doppler spectra. In most cases,a single measure is obtained. The lack ofroutine use of an average of multiple mea-surements despite current guideline recom-mendations (5) is due to the limited time inbusy echocardiography laboratories. Regard-less of underlying rhythm, there is likely an

added benefit to using an average of multiple mea-surements for determining chamber size or quantifi-cation of velocities, gradients, or stroke volumes.Automation makes it possible to rapidly obtain

ATION Temporal Progression in Automated

or KinesisAutomated

2D-LVEFMeasurements

AutomatedDoppler and

ChamberDimension

Measurements

Digital EchoQuantification

3D-LVAutomaLV Anal

AutomatedStrain

Measurements

202009200520021997996

Cardiol Img. 2019;12(6):1073–92.

automated quantification in echocardiography over the past 33 ye

fraction; PISA ¼ proximal isovelocity surface area.

multiple measurements (6,7) of chamber dimensionsor Doppler spectra from multiple heartbeats. Thesemeasurements have been shown to be accurate (R2 ¼0.90 to 0.98) (6) compared with expert annotations;they reduce analysis time and contribute to improvedreproducibility of the measurements. Unfortunately,these automated measurements are not uniformlyavailable with all echocardiography vendors, hencelimiting their widespread use.

ASSESSMENT OF MYOCARDIAL FUNCTION

AUTOMATED MEASUREMENTS OF 2-DIMENSIONAL LEFT

VENTRICULAR VOLUMES AND FUNCTION.

Noninvasive measurement of LVEF was first per-formed by one-dimensional (M-mode) measurementin the late 1960s, but the computer technology at thetime was insufficient to automatically analyze imageson a frame-by-frame basis. Contemporary automatedsoftware uses artificial intelligence technology with a“knowledge base” gained from large datasets ofechocardiographic images to improve detection of

Quantification in Echocardiography

AutomatedMitral ValveParametric

Measurements

Multi-chamberAutomatedLV Analysis

AutomatedMitral

Regurgitant3D-PISA

Measurement

EFtedysis

AutomatedStroke and

RegurgitantVolume

Measurement

AutomatedAortic Root

Measurement

20162014201310

ars. 2D ¼ 2-dimensional; 3D ¼ 3-dimensional; LV ¼ left ventricular;

TABLE 1 Studies Assessing Automated Echocardiographic Measurements of 2D LVEF

First Author,Year (Ref. #) N

Software(Company) Feasibility Comparator

Correlation/Agreement(Bias � LOA) Time Cost

InterobserverVariability

Cannessonet al.,2007 (16)

218 (165 withabnormal LV)

Auto EF(Siemens)

92% Manual Simpsonbiplane 2D LVEFand CMR LVEF

Manual 2D LVEF:r ¼ 0.98/0 � 2.9%

CMR LVEF: r ¼ 0.95/�0.3� 6.0%

Automated 2D LVEF:48 � 26 s

Manual 2D LVEF:102 � 21 s

Automated 2DLVEF: 1.3� 1.7%Manual 2DLVEF: 2.9� 2.1%

Maret et al.,2008 (8)

60 Auto EF(Siemens)

100% Manual Simpsonbiplane 2D LVEF

Uncorrected Auto EF:r ¼ 0.81/2.2 � 12.1%

Corrected Auto EF:r ¼ 0.89/0.8 �10.5%

Uncorrected Auto EF:79 � 5 s

Corrected Auto EF:159 � 46 s

Manuel 2D LVEF:177 � 66 s

Uncorrected AutoEF: novariability

Corrected Auto EF:ICC ¼ 0.88

Manual EF:ICC ¼ 0.74

Rahmouniet al.,2008 (9)

92 Auto EF(Siemens)

100% Manual Simpsonbiplane 2D LVEFand CMR LVEF

Manual 2D LVEF:r ¼ 0.64/7.0 � 13.0%

CMR: r ¼ 0.64/3.7 �13.8%

NA NA

Szulik et al.,2011 (13)

81 (hospitalized) Auto EF (GE) 90% Average of 4 visualLVEFmeasurementsand 2 CMR imagereads

Manual 2D LVEF:r ¼ 0.80/�0.66 �10.4%

Auto EF: r ¼ 0.77/�1.88�11.6%

Auto LVEF: 54 � 22 sManual LVEF:104 � 22 s

Auto EF: r ¼ 0.80Manual EF:

r ¼ 0.52

Aurich et al.,2014 (18)

47 Auto EF (GE) 100% Manual Simpsonbiplane 2D LVEF

CMR LVEF

2D LVEF: r ¼ 0.85/3 �12%

CMR: r ¼ 0.74/9 � 17%

Auto EF: 74 � 18 sManual 2D LVEF:113 � 30 s

CMR: 139 � 18 s

Auto LVEF:CoV ¼ 12%

Manual 2D LVEF:NA

Knackstedtet al., 2015(10)

255 (multicenter) AutoLV(TOMTEC)

98% Manual Simpsonbiplane 2D LVEF;visual LVEF

2D LVEF, Simpson:r ¼ 0.83/0.3 � 18.7%

2D LVEF, visual: r ¼ 0.83/2.2 � 17.4%

AutoLV: 8 � 1 sManual LVEF: NA

AutoLV: ICC ¼ 1.002DLVEF, Simpson:

ICC ¼ 0.782D LVEF, visual:

ICC ¼ 0.87

Frederiksenet al., 2015(11)

102 Auto EF (GE) 83% Manual Simpsonbiplane 2D LVEF,visual LVEF

2D LVEF, Simpson:r ¼ 0.82/NA

2D LVEF, visual: r ¼ 0.82/0 � 19%

Automated: 41 � 5 sManual: 98 � 8 s

NA

Hovnanianset al., 2017(12)

184 A2DQ (Philips) 100% Manual Simpsonbiplane 2D LVEF

ICC ¼ 0.93/0.4 � 15.3 Automated: 116� 57 sManual: 217 � 69 s

ICC ¼ 0.96

Abazid et al.,2018 (14)

268 Auto EF (GE) 89.5% Visual LVEFM-mode LVEF

2D LVEF, visual:r ¼ 0.92/�0.3 � 4.5%

M-mode LVEF: r ¼ 0.77/�2.4 � 8.4%

NA NA

2D ¼ 2-dimensional; CMR ¼ cardiovascular magnetic resonance; CoV ¼ coefficient of variation; GE ¼ General Electric; ICC ¼ intraclass correlation coefficient; LOA ¼ limits of agreement (2SD); LV ¼ leftventricle; LVEF ¼ left ventricular ejection fraction; NA ¼ not available.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 2 , N O . 6 , 2 0 1 9 Nolan and ThavendiranathanJ U N E 2 0 1 9 : 1 0 7 3 – 9 2 Quantification in Echocardiography

1075

endocardial borders. Multiple imaging companies arenow marketing automated 2-dimensional (2D) LVEFmeasurement software that is gain-independent andaccurate (e.g., eSie LVA by Siemens [Munich, Ger-many], AutoEF by GE Healthcare [Chicago, Illinois],AutoLV by TOMTEC [Munich, Germany], and a2DQAI

by Philips [Best, the Netherlands]). In early studies(Table 1) of automated analysis (8,9), correlation with2D LVEF and accuracy for grading cardiomyopathyseverity were suboptimal (8). However, multiplestudies using software from several vendorsinvolving >1,300 patients have now shown highfeasibility (83% to 100%) for bi-plane 2D LVEF mea-surements and excellent agreement with core labo-ratory measurements or other external reference

standards such as cardiovascular magnetic resonance(CMR) imaging (8,10–16).

The fully automated methods have no interob-server variability (10,17) when repeated on the sameechocardiographic images, whereas semi-automatedmethods have better reproducibility than manualmethods. Furthermore, automated methods reducethe gap in reproducibility between expert and novicereaders and improve accuracy and reproducibility inmulticenter settings for 2D LVEF measurements(8,10). The analysis duration is shortened by >50%,with fully automated analysis providing the greatesttime savings (as short as 10 s [10]) comparedwith semi-automated (8), manual, or even CMR(11–13,16,18) methods. Therefore, given the

FIGURE 1 Automated 3D Volumes and Ejection Fraction Quantification

A

B

LV

LA

400 10 20 30 40 50 60 70 80 90

50

60

70

80

90

100

110

120

130

Volume (mL)

Volume

Frame number

Volume EF: 66.20% ED 0 (128.0mL) ES 9 (43.3mL)

(A) Quantification using eSie LVA with measurements from 3 consecutive beats. (B) Quantification using HeartModel in which all 4 chambers are automatically modeled.

LA ¼ left atrium; LV ¼ left ventricle.



1076

existing data and availability of automatedmethods with most vendors, routine use of auto-mated 2D LVEF measurements in echocardiogra-phy laboratories should be feasible, especially inpatients with good acoustic windows and sinusrhythm.

AUTOMATED MEASUREMENT OF 3D LEFT VENTRICULAR

VOLUMES AND FUNCTION. 3D LVEF with echocardi-ography is a more accurate and reproducible methodof measuring left ventricular function than 2D LVEF(19), with test–retest variability approximating one-half that of 2D LVEF (3). However, acquisition andmanual post-processing take 3 to 5 min per study(20,21). Despite recommendations for routine use of3D LVEF in contemporary guidelines (22,23), work-flow limitations and need for analysis expertise haveremained bottlenecks for widespread uptake. Auto-mated analysis removes some of these limitations andcan promote uptake of 3D LVEF measurements inbusy echocardiography laboratories.

Several vendors have produced fully automatedsoftware (e.g., eSie LVA from Siemens, HeartModelfrom Philips) that requires minimal user interaction

other than initiation of the analysis package. eSie LVAoffers single-click analysis of the left ventricle,whereas HeartModel provides single-click analysis ofall 4 cardiac chambers, although only the left-sidedanalysis is clinically available (Figure 1). Moststudies have reported >90% feasibility for fully orsemi-automated 3D LVEF measurements (Table 2)(15,20,21,24–26) with single-beat fully automatedmeasurements taking <30 s (21,27,28) and 3 to 5consecutive beats taking 30 to 60 s depending onvolume rates (26,29,30). Using HeartModel, thequantification of 3D LVEF and atrial volumes for asingle volume takes <30 s (21). Therefore, fullyautomated methods can reduce analysis times by>75% compared with semi-automated or manualmethods (20,21,26,31). For a busy echocardiographicdepartment that reports 80 to 100 studies daily, >4 to5 h (assuming 3 min per study) may be spent per-forming manual 3D LVEF measurements, and fullautomation could potentially minimize this post-processing time.

The agreement of the automated methods for 3DLVEF measurements have ranged from r ¼ 0.75 to0.98 compared with manual 3D methods or CMR (32).

TABLE 2 Studies Assessing Automated Echocardiographic Measurements of 3D LVEF



Correlation / Agreement(Bias � LOA) Time Cost Interobserver Variability

Muraru et al.,2010 (25)

103 4D AutoLVQ(GE)

100% Manual 3D LVEF/CMR LVEF

Corrected contours:3D LVEF (manual): r ¼

0.95/–0.1 � 4.9%CMR: r ¼ 0.85/2.9 � 8.4Uncorrected contours:3D LVEF (manual): r ¼

0.75/–4.4 � 11.6%CMR: r ¼ 0.64/–1.5 �

12.5%

Automatic: 142 � 30 sManual: 226 � 114 s

4D AutoLVQ: r ¼ 0.98,6.5 � 6.0%

3D LVEF (manual):r¼ 0.98, 4.6� 4.3%

Barbosa et al.,2013 (24)

24 BEAS algorithm 100% Manual 3D LVEF r ¼ 0.91/–1.0 � 9.8% Automatic: 30.7 � 7.5 s Semi-automated:3.6 � 4.3%

3D LVEF (manual):7.0 � 5.5%

Thavendiranathanet al., 2012 (30)

91 (67 SR,24 AF)

eSie LVA(Siemens)

71.1% LVEF as measured byCMR imaging forSR, and manual2D LVEF for AF

SR: r ¼ 0.98/0.3 � 2.5%AF: r ¼ 0.91/–2.0 � 4.0%

Automatic: 30–60 s eSie LVA: 0.4 � 4.5%CMR: 1.0 � 2.0% Test–

retest: r ¼ 0.98,0.4 � 2.8%

Shibayama et al.,2013 (20)

44 eSie LVA(Siemens)

93.2% Manual Simpsonbiplane 2D LVEF;CMR LVEF

2D LVEF, Simpson: r ¼0.89/–5.5 � 15.4%

CMR: r ¼ 0.90/–1.0 �15.1%

Fully automated:37 � 8 s

Semi-automated:371 � 116 s

Fully automated:ICC ¼ 0.95, 0.6�7.0%

Semi-automated:ICC ¼ 0.92, 3.6�11.5%

Aurich et al.,2014 (18)

268 3D AutoLVQ(GE)

100% Manual Simpsonbiplane 2D LVEF

CMR LVEF

2D LVEF: r ¼ 0.79/3.0 �14%

CMR: r ¼ 0.73/9.0 � 17

3D AutoLVQ (withcontour adjustment);

261 � 93 s;2D Bi-plane EF 113 � 30 s;

CMR 139 � 18 s

3D AutoLVQ:COV ¼ 11%

Tsang et al.,2016 (21)

159 HeartModel(Philips)

90.5% Manual 3D LVEF(94 patients),CMR LVEF

(65 patients)

Manual 3D LVEFNo contour adjustment

r ¼ 0.87/–6 � 16%Contour adjustment

r ¼ 0.92/–4 � 12%CMR: No contour

adjustment r ¼ 0.85/–2 � 18%

Contour adjustmentr ¼ 0.91/–2 � 16%

Fully automated:26 � 2 s

Semi-automated:76 � 6 s

Manual 3D LVEF:144 � 32 s

Fully automated:0 � 0%

Semi-automated:9 � 6%

Test–retest:Fully automated,8 � 9%; semi-automated,

8 � 8%

Otani et al.,2016 (35)

88 withAF

HeartModel(Philips)

92.4% LVEF as measured bymanual 3D LVEF

r ¼ 0.91/–0.3 � 10.4% Automated: 5 minManual: 27 min

Test–retest variability:CoV ¼ 3.8%;ICC ¼ 0.99

Levy et al.,2017 (29)


85.7% CMR LVEF r ¼ 0.91/0.7 � 7.0% NA r ¼ 0.90; CoV ¼ 5%Test–retest: r ¼ 0.91,

CoV ¼ 6%

Medvedofskyet al., 2017 (32)


90% 3D guided bi-planeLVEF

r ¼ 0.79/0.0 � 19%(manual correctionused as necessary)

r ¼ 0.94/–0.1 � 10%(when poor imagequality datasetremoved)

NA Adjustment ofautomatedcontours notadvisable forunexperiencedreaders

Tamborini et al.,2017 (28)


94.5% 3D LVEF semi-automaticmethod;

CMR LVEF

3D LVEF: r ¼ 0.88/7.3 �12.9%

CMR: r ¼ 0.79/4 � 20%

HeartModel: 29 � 10 s3D LVEF: 160 � 30 s

For LVEDV: CoV ¼5.0%; ICC ¼ 0.98

For LVESV: CoV ¼10.7%, ICC ¼ 0.97

Spitzer et al., 2017(31)


93% Manual 3D LVEF Correlation: r ¼ 0.84 Automated: 0.5 minAutomated with manual

corrections: 2.5 minManual: 7.0 min

Automated corrected:CoV ¼ 11.9%

Test–retest variability:Automated without

corrections:CoV ¼ 6.9%

Medvedofskyet al., 2018(33)


90% LVEF as measured bymanual 3D LVEF

Fully automated: r ¼0.88/–2.0 � 15%

Semi-automated: r ¼0.90/1 � 14%

NA HeartModel:13 � 12%, manual:

17 � 14%Test–retest:Fully automated:

9 � 11%, semi-automated:14 � 9%

Continued on the next page


1077

TABLE 2 Continued



Correlation / Agreement(Bias � LOA) Time Cost Interobserver Variability

Sun et al., 2018(34)


94.2% Manual 3D LVEF Fully automated: r ¼0.96/–1 � 4%

Semi-automated: r ¼0.97/–1 � 3%

Automated: 1.1 � 0.3 minManual: 4.9 � 2.4 min

Fully automated:7 � 4%

Semi-automated:7 � 4%; 3D LVEF(manual): 7 � 3%

Muraru et al., 2018(26)

92 AutoLVQ (GE),3DQ ADV(Philips),eSie LVA(Siemens)

Auto LVQ90%;3DQ ADVNA; eSieLVA96%

CMR LVEF(n ¼ 35)

AutoLVQ: r ¼ 0.80/–2.0 � 12%

3DQ ADV: r ¼ 0.79/–1.6 � 14%

eSieLVA: r ¼ 0.77/–0.4 � 14%

AutoLVQ : 224 � 29 s3DQ ADV: 358 � 36 s eSie

LVA: 135 � 24Automated: 30 � 10 sSemi-automated:187 � 46 s

Semi-automated:AutoLVQ ¼ 1 � 10%;

3DQ ADV ¼ 0 �10% eSieLVA ¼ 3 �10%

Automated:AutoLVQ ¼ 0 � 4%

eSie LVA ¼ 0 � 4%

3D ¼ 3-dimensional; 4D¼ 4-dimensional; AF ¼ atrial fibrillation; BEAS¼ B-Spline Explicit Active Surfaces; EF ¼ ejection fraction; LVEDV¼ left ventricular end-diastolic volume; LVESV¼ left ventricular end-systolic volume; SR ¼ sinus rhythm; other abbreviations as in Table 1.



1078

Interestingly, there seems to be small improvementsin agreement with reference method with manualadjustment of automated contours made by expertreaders, but not novice readers, compared with fullyautomated measurements (21,25,32–34). However,

FIGURE 2 Automated Quantification in Atrial Fibrillation

0 200 4

260

250

240

230

220

210

200

190

VolumemL

A4C A2C

Multiple 3-dimensional datasets from consecutive beats measured by usi

EDV ¼ end-diastolic volume; ESV ¼ end-systolic volume; SV ¼ stroke v

fully automated 3D LVEF measurements offer theadvantage of significantly reducing or eliminatingmeasurement variability, as measurements aredeterministic and outcomes are invariable for a givenstudy regardless of the user experience (32). This

00 600 800% of heart cycle

SAX

ng eSie LVA in a patient with atrial fibrillation. EF ¼ ejection fraction;

olume.

TABLE 3 Studies Assessing Automated Echocardiographic Measurements of Myocardial Strain

First Author,Year (Ref. #) Patients Software Feasibility Comparator Correlation/Agreement Time Cost Interobserver Variability

Ingul et al.,2005 (40)

60 (30 prior MI, 30healthy controlsubjects)

GcMat (GEVingmedUltrasound)

81.5% ofsegments(GLS)

Manual strainanalysis

Accuracy specktrackingmethod ¼ 95.8%,manual method ¼96.2%

Automated:2 min

Manual: 11 min

CoV Automated 15%;manual 30%

Delgado et al.,2008 (43)

222 (CAD) AFI (GE) 100% (GLS) 2D LVEF r ¼ 0.83 NA ICC ¼ 0.92; –0.2 �2.6%

Belghiti et al.,2008 (45)

65 (referred forcatheterization)

AFI (GE) 97% (GLS) LVEF asmeasured by2D LVEF

r ¼ 0.87 (experiencedid not affectcorrelation), r ¼0.96 betweenexpert andbeginner (LOA3.4%)

<60 s Variability: 7.1%experienced reader,8.7% inexperiencedreader

2D LVEF 14.5%

Brown et al.,2009 (42)

62 (prior MI) 4D analysis(TOMTEC)

93% ofsegments(GLS)

LVEF asmeasured byCMR

r ¼ –0.69 Strain: 132 � 30 sCMR: 630 � 60 s

8.6%

Villanueva-Fernandezet al., 2012(44)

59 QLab (Philips) 91% amongexpert, 89%among non-experts (GLS)

NA NA NA �0.09 � 4.45(experts)

0.66 � 7.68 (expert vs.nonexpert)

Knackstedtet al., 2015(10)

255 AutoLV(TOMTEC)

98% (GLS) Manual biplanelongitudinalstrain

ICC ¼ 0.83/0.7(95% CI, 0.1-1.3)

NA Automated ICC ¼ 0,LOA 0

Manual ICC ¼ 0.88,LOA 9.6%

Medvedofskyet al., 2017(32)

30 EchoInsight(Epsilon)

93% (GLS) Manual biplaneLVEF

NA GLS: 1 min2D LVEF: 2 min

GLS r ¼ 0.98, 2D EFr ¼ 0.91 Minimalimpact of readerexperiences on GLS,EF ICC 0.89, GLS0.98

AFI ¼ automated function imaging; CAD ¼ coronary artery disease; GLS ¼ global longitudinal strain; MI ¼ myocardial infarction; NA ¼ not reported; other abbreviations as in Table 1.

FIGURE 3 Automated 3D Volume Color Doppler Technique for Stroke Volume Quantification

20

ml/s

ms20001000

Oo Oc

Io Ic Io Ic Io Ic

Oo Oc Oo Oc

0

Stroke Volume

InflowOutflow

10

0

10

-20

The left panel shows flow through the mitral annulus (yellow arrow) along with the sampling plane, the middle panel shows the flow through the left ventricular

outflow tract and the sampling plane (orange arrow), and the right panel shows flow curves and the inflow and outflow stroke volumes (SVs). In this patient with mild

mitral regurgitation, the discrepancy in SV was 11.7 ml (i.e., regurgitant volume). 3D ¼ 3-dimensional; CO ¼ cardiac output; ERO ¼ effective orifice area; MR ¼ mitral

regurgitation; PISA ¼ proximal isovelocity surface area; RF ¼ regurgitant fraction; VTI ¼ velocity-time integral.


1079

TABLE 4 Studies Assessing Automated Echocardiographic Measurements of Stroke Volumes and Regurgitant Volumes by Using 3D Color Doppler

Echocardiography

First Author, Year(Ref. #) N Software (Company) Feasibility Comparator

Correlation/Agreement(Bias � LOA) Time Cost

InterobserverVariability

Matthews et al.,2010 (52)

15 (post-cardiacsurgery)

Unnamed 100% CO as measured byPAcatheterization

Cardiac output; R2 ¼0.71/0.09 � 1.3 L

NA Not reported

Thavendiranathanet al., 2012 (49)

44 patients(referred forassessmentof cardiacfunction)

Unnamed(Siemens)

100% PC-CMR, 2D manualSV method

CMR vs Auto 3D SV:MV: r ¼ 0.91/1.1� 18.9 ml; AV SV:r ¼ 0.93/–0.7 �17.8 ml

CMR vs 2D SV: MVr ¼ 0.66/10.6 �36.0 ml; AV0.6/10.6� 40.3 ml

Unadjusted:20–40 s;Adjustments:40–60 s2D manual ¼ 4–6 min

Automated 3D MV:r ¼ 0.97, AV0.95

2D manual MV:r ¼ 0.79, AV0.92

Son et al., 2013(53)

32 patients$moderateMR

Unnamed(Siemens)

93.8% PC-CMR for MRseverity

RV: r ¼ 0.85/–5.7 �33.6 ml

NA ICC ¼ 0.89

Thavendiranathanet al., 2013(54)

30 (functionalMR)

Unnamed(Siemens)

100% PC-CMR for MRseverity

RV: r ¼ 0.91/–1.6 �17.0 ml

RF: r ¼ 0.92/–0.3 �14.6%

30-60 s RV 0.9 � 11.5 ml,CCC 0.96

RF 0.2 � 10.9%;CCC 0.93

Test–retest: RV, 1.2� 8.8 ml; RF,1.6 � 9.7%

Gruner et al., 2015(57)

27 (post-mitralclip for MR)

Unnamed(Siemens)

89% Visual MRclassification

3D Automated colorDoppler betteragreement withvisual MR than 2Dmanual method

NA ICC 088 for MV SV0.86 of AV SV

Choi et al., 2015(55)

32 (moderate-to-severe AR)

Unnamed(Siemens)

93.8% PC-CMR for ARseverity

AR RV: r ¼ 0.93, LOA9.5 ml, ARseverityagreementk ¼ 0.94

5.6 � 2.0 min RV ICC ¼ 0.96

Heo et al., 2017(56)

152 (MR) Unnamed(Siemens)

97.4% PC-CMR for MRseverity (37patients)

MR RV volume:r ¼ 0.94,2D Volumetricmethod r ¼ 0.56

Automated:4.3 � 2.2 min

ICC ¼ 0.87automated,2D manualICC 0.93

Kato et al., 2018(51)

34 children (29ASD, 3 hearttransplant, 1arterial duct)

Unnamed(Siemens)

92% for AV97% for MV80% for PV92% for TV

Qp:Qs calculated viaFick method

PV/AV ratio: r ¼ 0.84TV/MV ratio: r ¼ 0.87

Automated:3–5 min per valve

MV: CoV 12.6%TV: CoV 8.9%AV: CoV 13.2%PV: CoV 8.2%

ASD ¼ atrial septal defect; AR¼ aortic regurgitation; AV ¼ aortic valve; CCC ¼ concordance correlation coefficient; CO ¼ cardiac output; MR ¼mitral regurgitation; MV ¼mitral valve; PA¼ pulmonary artery;PC-CMR ¼ phase contrast cardiovascular magnetic resonance; PV ¼ pulmonary valve; Qp:Qs ¼ pulmonary-systemic flow ratio; RV ¼ regurgitant volume; RF ¼ regurgitant fraction; SV ¼ stroke volume; TV ¼tricuspid valve; other abbreviations as in Tables 1 and 2.



1080

results in decreased interobserver and test–retestvariability (20,21,26,31,33), likely leading to moreuniform clinical decision-making.

Another contemporary challenge for accurate 3DLVEF measurement is in patients with irregularrhythms due to substantial beat-to-beat variation intheir LVEF. Current guidelines recommend averagingof sequential LVEF measurements over 5 beats (23).Use of multiple single-beat 3D acquisitions may allow3D LVEF quantifications in the context of arrhyth-mias. Automated analysis with eSie LVA using 3Ddatasets from multiple consecutive heartbeats isfeasible with accuracy similar to individual-beat 2Dmeasurements (Figure 2) (30). In addition, automated3D LVEF measurements from single index beats

correlate with manually measured averaged beatswith absolute reductions of time to analysis of 22 minper study (35). Therefore, automated analysis ofsingle-beat 3D LVEF data may become a reasonableapproach in patients with atrial fibrillation and otherarrhythmias.

Given the data on the accuracy and reproducibilityof automated 3D LVEF measurements and the avail-ability of semi-automated or fully automated quan-tification packages with most vendors, routine 3DLVEF measurement should be feasible in patientswith good 3D image quality.AUTOMATED STRAIN MEASUREMENTS. Assessment ofmyocardial mechanics, referred to as strain imaging,has many strengths, including detection of

FIGURE 4 Automated 3D PISA Quantification

The 3D PISA model (green) is shown on the 2-dimensional and 3D image planes. The calculated effective regurgitant orifice area was 0.63 cm2,

and regurgitant volume was 64.3 ml. Abbreviations as in Figure 3.


1081

subclinical cardiac dysfunction (36,37), and has su-perior ability to predict mortality compared with 2DLVEF measurements (36,38). Although strain imagingwas introduced 20 years ago (39), clinical uptake wasinitially delayed by the complexity of the analysis andsuboptimal reproducibility of manual measurements.Advancements enabling high frame rate 2D imagesallowed successful implementation of semi-automated strain measurements using speckletracking methods. This approach achieved accuracysimilar to manual measurements and reduced anal-ysis time by up to 82%, positioning semi-automated2D speckle strain for better clinical uptake (40).Given that 2D speckle tracking–based strain analysiswas introduced clinically as a semi-automated tech-nique and was validated by using sonomicrometry(41), there is limited published data comparing theaccuracy and reproducibility of automated versusmanual methods (Table 3) (40,42–44).

Currently, strain measurements are made by usingvendor-specific software (AFI [General Electric] andQLAB [Philips]) and vendor-neutral software (Auto-STRAIN [TOMTECT], VVI [Siemens], and EchoInsight

[Epsilon Imaging, Ann Arbor, Michigan]). These pro-grams require manual identification of certain ven-tricular landmarks followed by specification of thewidth of the region of interest, which is then auto-matically tracked through the cardiac cycle (45).Although some degree of manual input may intro-duce variability, contemporary analysis packageshave shown excellent interobserver reproducibilityfor semi-automated global longitudinal strain (GLS)measurements compared with manual methods(Table 3) (10,43). For example, 1 study of 546 patientsanalyzed with AFI reported an interobserver intra-class correlation coefficient (ICC) of 0.92 for GLScompared with an ICC of 0.80 for 2D LVEF (38).Similarly, in a multicenter setting, semi-automatedmeasurements of GLS have been shown to be morereproducible than 2D LVEF (46). As opposed to GLS,regional longitudinal strain measurements are lessreproducible (41) and currently not encouraged forroutine clinical practice. More recently, fully auto-mated GLS strain measurements are now possiblewith some vendors (10). Whether this availability willfurther improve the reproducibility of the global and

FIGURE 5 Quantification of Regurgitant Volume by Using 3D Integrated PISA

MR Regurgitant Volume by 3D Integrated PISA = 42.3 ml

0.63 cm2 12.1 ml 0.31 cm2 6.0 ml 0.28 cm2 5.5 ml 0.12 cm2 2.2 ml 0.2 cm2 3.9 ml 0.21 cm2 4.0 ml 0.45 cm2 8.6 ml

The 3D PISA on each frame is automatically quantified and regurgitant volume is calculated. Individual volumes are added together to obtain the total regurgitant

volume. Abbreviations as in Figure 3.

TABLE 5 Studies As

First Author,Year (Ref. #)

Thavendiranathanet al., 2013(54)

3

De Agustin et al.,2012 (78)

3

De Agustin et al.,2013 (64)

9

Schmidt et al.,2014 (62)

9

Choi et al.,2014 (63)

2

EROA ¼ estimated regurgi



1082

segmental strain measurements and optimize work-flow remains to be seen.

Currently, clinical strain measurements arelimited to systolic GLS and are obtained by usingmultiple 2D left ventricular apical images. Small

sessing Automated Echocardiographic Measurements of 3D PISA

N Software Feasibility ComparatorCorre

0 withfunctionalMR

eSie PISA(Siemens)

85.6% RV measured byCMR RV

RV byr ¼

3 (25 primaryMR; 8secondaryMR)

eSie PISA(Siemens)

100% 3D TEE planimetry AutomEROA;

0.1RV: r ¼Manua

0.9RV: r ¼

0 (chronic TR) eSie PISA(Siemens)

100% 3D VCA Vs VCA0.0PIS0.2

3 (17 primary;76 secondaryMR)

eSie PISA(Siemens)

100% Manual 3D PISAmeasurement

RV: r ¼EROA:

21 (111 primary;110secondaryMR)

eSie PISA(Siemens)

95.5% PC-CMR RV: SePIS13.8PIS19.

tant orifice Area; PISA ¼ proximal isovelocity surface area; TEE ¼ transesophageal echocard

studies have shown that semi-automated 3D leftventricular strain is feasible and reproducible (47),but larger studies are required to quantify clinicalbenefit. Therefore, the large amount of data on thevalue of automated GLS measurements in many

lation/Agreement(Bias � LOA) Time Cost Interobserver Variability

CMR:0.92/–1.4 � 18

Automated IntegratedPISA:

1.7 � 0.7 min;Automated Peak PISA:

15 � 4 sCMR: 6–8 min

RV: 0.9 � 11.5 mlRF: 0.2 � 10.9%Test–retest:RV: 0.0 � 13.7 ml

ated 3D PISA:r ¼ 0.99/0.0 �cm2

0.99l 2D PISA: r ¼3/0.2 � 0.4 cm2

0.95

Automated 3D PISA:4–5 min

Manual 2D PISA: NA

3D PISA ICC: 0.922D PISA ICC: NA

3D PISA r ¼ 0.97/1 � 0.12 cm2, 2DA r ¼ 0.89/0.1 �7 cm2

2–3 min for 3D PISA,3–4 min for 2D PISA

3D PISA ICC: 0.882D PISA ICC: 0.79

0.91r ¼ 0.93

Not reported Not reported

mi-automated 3DA; r ¼ 0.97/–0.9 �ml; Manual 2D

A r ¼ 0.84/10.4 �4 ml

Automated:4.6 � 2.0 min

3D PISA ICC: 0.972D PISA ICC: 0.95

iography; VCA ¼ vena contracta area; other abbreviations as in Tables 1, 2, and 4.

TABLE 6 Studies Assessing Automated Echocardiographic Measurements of Aortic Annulus and Root

First Author,Year (Ref. #) N Software Feasibility Comparator

Correlation/Agreement(Bias � LOA) Time Cost Interobserver Variability

Calleja et al.,2013 (66)

69 (20 healthyvolunteers, 14severe AS, 35dilated aortic root� AR)

eSie Valve (Siemens) 100% MDCT, automatedannular diametermeasurement

0.5 � 5.87 mm Automated:1.1–3.4 min

ICC ¼ 0.90 to 0.93

Garcia-Martinet al., 2016(68)

31 (referred forTAVR)

eSie Valve (Siemens) 88.6% Manual 3D TEEmeasurements

Aortic annulardiameter:ICC 0.85

Aortic annulararea: ICC 0.74

NA Aortic annulardiameter: ICC ¼0.94

Aortic annular area:ICC ¼ 0.95

Medirattaet al., 2017(65)

52(severe AS)

Unnamed (Philips) 90% MDCT Aortic annular area:r ¼ 0.92

Aortic annularperimeter:r ¼ 0.91

NA Aortic annular area: ICC0.70;

Aortic annularperimeter: ICC 0.72

Prihadi et al.,2018 (67)

150 (severe AS) Aortic ValveNavigator(Philips)

100% for aorticannulardimensions;89% foraortic rootdimensions

MDCT Aortic annular area:r ¼ 0.91

Aortic annularperimeter:r ¼ 0.83

Automated aorticrootmeasurements:

4.2 � 1.0 min

ICC 0.93

Queiros et al.,2018 (79)

101 Speqle3D (Universityof Leuven)

92.1% MDCT Fully automated:ICC 0.78;

Contour correction:ICC 0.83

Automated:19.0 � 1.9 s

ICC 0.94CoV 5.6%

Podlesnikaret al., 2018(80)

83 (severe ASseparated intohigh and low AVC

4D Auto AVG(GE-Vingmed)

97.6% MDCT For selectingprosthesis size

Low AVC: K ¼ 0.93;High AVC: K ¼ 0.71

NA Perimeter ICC 0.96;Area: ICC 0.97

Kato et al.,2018 (84)

43 (severe AS) eSieValves(Siemens) 93.5% MDCT Automated annulararea: r ¼ 0.86

Semi-automatedannular area:r ¼ 0.94

Manual annular area:r ¼ 0.93

Automated 3D TEE:30.1 � 5.8 s

Semi-automated 3DTEE: 74.1 � 15 s

Manual 3D TEE: 81.8� 18.5 s

Automated; ICC 0.99,LOA –28.6 to 26.7

Semi-automated;ICC 0.96

LOA –22.7 to 66.8Manual; ICC ¼ 0.95LOA –47.0 to 81

AS ¼ aortic stenosis; AVC ¼ aortic valve calcium; MDCT ¼ multi detector computed tomography; TAVR ¼ transcatheter aortic valve implantation; other abbreviations as in Tables 1, 2, and 5.


1083

clinical situations and its availability with mostvendors makes it attractive for routine clinicalapplication. It is likely that further advances inautomation and efforts in standardization of strainmeasurements between vendors may allow GLS tosupersede LVEF as the primary measure of leftventricular systolic function and enhance its use inpoint-of-care imaging.

ASSESSMENT OF VALVULAR HEART DISEASE

AUTOMATED MEASUREMENT OF STROKE VOLUME

AND REGURGITANT VOLUME. The AmericanSociety of Echocardiography’s Valvular RegurgitationGuidelines recommend the use of volumetric tech-niques to quantify severity of valvular regurgitation(48). However, this technique is not routinely per-formed due to concerns of wide interobserver mea-surement variability from squaring of measurements

leading to exaggeration of imprecisions and workflowdemands with multiple measurements taking up to 4to 6 min (49).

Semi-automated measurement of stroke volume byusing real-time 3D color Doppler datasets allowsmeasurement of velocities over the entire orifice(e.g., mitral or tricuspid annulus) of interest to pro-vide a measure of stroke volume (Figure 3) (49,50).The benefits of this approach are avoidance of as-sumptions of orifice geometry, rapid simultaneousstroke volume measurements at multiple valves,and improved accuracy and reproducibility (Table 4)(49–54). Stroke volume calculations are determinedby automated detection of the relevant landmarks byusing an expert-annotated database of sample im-ages, followed by placement of hemispheric velocity-sampling planes and automated de-aliasing forDoppler velocity ambiguity. Measurements of strokevolume have been shown to be feasible in >85% of

FIGURE 6 Automated Modeling of the Aortic Root and Aortic Cusps

Three-dimensional Color Doppler data are superimposed on the images. Annular, ostial, and valve orifice areas are automatically measured

and displayed.



1084

the patients with good agreement with CMR phase-contrast imaging (49), and they take <60 s withoutmanual adjustments (49).

A promising application of the automated 3D colorDoppler stroke volume technique is quantification ofvalvular regurgitation. Mitral or aortic regurgitantvolume can be calculated indirectly as difference inmitral inflow and aortic outflow stroke volumes usingthe same cardiac cycle (54,55). For functional mitralregurgitation, for example, the time required toobtain automated mitral and aortic stroke volume for3 to 5 consecutive cardiac cycles is reportedly 30 to 60s, and the calculated regurgitant volume had goodcorrelation and agreement with CMR and was supe-rior to that for 2D manual methods (54). The value ofthe automated 3D color Doppler technique has nowbeen consistently illustrated in several studies,including for the assessment of aortic regurgitation(55–57). The strength of this technique is particularlynotable in patients with eccentric and multiple jets.However, because published studies excluded pa-tients with multivalvular disease, significant

arrhythmia, and poor images, it is likely that accu-racy, reproducibility, and time efficiency may belower during routine clinical application.

AUTOMATED PROXIMAL ISOVELOCITY SURFACE

AREAS MEASUREMENTS. Semi-automated analysisof the 3D proximal isovelocity surface area (PISA) toestimate effective regurgitant orifice area and regur-gitant volume offers a new approach to directlyquantify regurgitation severity. Manual quantifica-tion of effective regurgitant orifice area and regur-gitant volume by using 2D PISA makes assumptionsregarding the shape of the proximal flow convergenceregion, resulting in significant interobserverdisagreement for severity classification (58). The useof manual 3D PISA avoids the need to make specificshape assumptions, offering improved accuracycompared with the 2D PISA technique (59,60), butthis method is laborious and impractical (59).

Semi-automated software can use 3D colorDoppler datasets to quantify the 3D PISA (Figure 4)(54) over the entire cardiac cycle (“integrated

FIGURE 7 Automated Modeling of the Mitral Annuls and Leaflets

Color Doppler data are superimposed on the model. Various automated measurements are displayed.


1085

PISA”), accounting for the dynamic nature of theregurgitant orifice (Figure 5) (61). This has beenshown to be accurate and reproducible in bothin vitro models (54) and when applied to patientswith functional mitral regurgitation (Table 5). Thesemultiple measurements of the 3D PISA over thecardiac cycle are only practical with semi-automated methods. The reduced variability ineffective regurgitant orifice area measurement byusing 3D PISA compared with traditional 2D PISAmethods has been replicated in 2 other largerstudies of >300 patients with both organic andfunctional mitral regurgitation (62,63). Similar re-sults have been observed for quantification oftricuspid regurgitation severity (64).

GUIDANCE FOR INTERVENTION

AUTOMATED MEASUREMENT OF AORTIC ANNULUS

AND ROOT. Clinical outcomes of transcatheter ther-apies for aortic valvular disease depend on accuratemeasurements of aortic valve landmarks. Measure-ments of aortic annulus size by using 2D trans-esophageal echocardiography (TEE) underestimate

aortic annular area due to incorrect geometric as-sumptions. 3D TEE–based measurements offer incre-mental improvement compared with multidetectorcomputed tomography (MDCT) imaging (65).Currently, MDCT scanning is the recommended im-aging modality for accurate aortic annulus and rootdimensions before transcatheter aortic valve im-plantation; it is limited, however, by risk of nephro-toxicity from contrast use, radiation exposure,limited temporal resolution, and susceptibility tocalcium blooming artifacts obscuring the true annularborder. Automated 3D TEE measurements may avoidthese undesirable effects and lead to safer, accurate,and reproducible measurements.

Several vendors have now produced automatedsoftware specifically designed for measuring aorticannular and root measurements from 3D TEE data-sets with good feasibility, accuracy, and reproduc-ibility (Table 6, Figure 6) (66,67). This softwareallows fully automated analysis but also allows formanual adjustments (66). For transcatheter aorticvalve replacement assessment, the bias of auto-mated 3D TEE measurements for annular diameter,perimeter, and area appears small (approximately

TABLE 7 Studies Assessing Automated Echocardiographic Measurements for MV Anatomy

First Author,Year (Ref. #) N Software Feasibility Comparator Correlation/Agreement Time Cost Interobserver Variability

Grewal et al.,2010 (72)

32 (moderate-to-severeprimary MR)

Q-Lab MVR(Philips)

100% Direct surgicalmeasurement

Mean difference inaortic annulusmeasurement:0.1 � 0.1 mm,95% CI � 4.4 mm

NA Antero-posterior diameterCoV 5.7%

BA 95% CI � 5.1 mm; Inter-commissural distanceCoV 4.3%; BA 95% CI �2.3 mm

Pouch et al.,2014 (81)

20 pre-cardiacsurgery (6normal MV, 6mild MR, 8severeischemic MR)

ITK-SNAP (open-source)(automated)

100% Manual 3D TEEmeasurements

Mean distancebetween manualand automaticsegment: diastole0.8 � 0.2 mm;systole 0.6 �0.2 mm

NA NA

Calleja et al.,2015 (76)

94, booked forMV surgery

Pending (semi-automated)

100% Length of surgicallyimplantedannuloplasty ring

Correlation betweenpredicted andimplantedannuloplastyband length:r ¼ 0.74

Automated, nocontourcorrection:

<60 s; Automated,contourcorrection:

8 min

Annular area:ICC ¼ 0.99

Jin et al., 2015(83)

55 (33 MVP, 11functionalMR, 11normal)

Mitral ValveNavigator(Philips)


Annuluscircumference:r ¼ 0.95

Antero-posteriordiameter:r ¼ 0.96

Automated:144 � 24 s;Manual: 770 � 89 s

Annulus circumference:ICC ¼ 0.98Antero-posterior diameter:

ICC ¼ 0.97

Kagiyamaet al., 2016(82)

74 (15 functionalMR; 32 MVP;27 normal)

Mitral ValveNavigator(Philips)


Agreement usingCronbach’s alpha:3D annuluscircumference:a ¼ 0.88

Antero-posteriordiameter:a ¼ 0.90

Automated: 260 �65 s;

Manual: 381 � 68 s

Interobserver agreementusing Cronbach’s alpha:3D annuluscircumference:

a ¼ 0.97; Antero-posteriordiameter:

a ¼ 0.96

Aquila et al.,2016 (85)

36, referred for3D TEE forany reason

eSie Valves(Siemens)


Mitral annular area:r ¼ 0.94

Inter-commissuraldistance: r ¼ 0.84

Not reported Mitral annular area:ICC ¼ 0.96Inter-commissural

diameter: ICC ¼ 0.96

BA ¼ Bland-Altman; MVP ¼ mitral valve prolapse; other abbreviations as in Tables 1, 2, 4, and 5.



1086

�1.2% to �2.2%), and agreement with MDCT imag-ing for selection of device size has been excellent(k ¼ 0.90) (67). Time required for automatedaortic root measurements from 3D TEE datasetshas ranged from 2.3 � 0.6 min (66) to 4.2 � 1.0 min(67), with longer times required if excessivecalcium or poor image quality is present. Smallerstudies have shown similar promising results interms of accuracy compared with MDCT imaging(65,68,69).

Beyond the annulus, automated modeling of aorticroot and valve is also feasible using software based onlearned-pattern recognition (66), with cross-sectionalmeasurement root diameters and leaflet parametersdisplaying good agreement with MDCT imaging (66).In addition, several published studies display repro-ducibility similar to MDCT measurements with ICCsfor intraobserver and interobserver variability

ranging from 0.91 to 0.98 (Table 6). This reproduc-ibility is particularly important in longitudinal follow-up to determine timing of surgical intervention(66,67,70).AUTOMATED MEASUREMENT OF MITRAL VALVE

ANATOMY. The use of 3D echocardiography hassignificantly enhanced our understanding of themitral valve and annular anatomy. Specifically,the ability to generate parametric maps has improvedthe accuracy of identifying mitral valve pathology,has been validated surgically, and has the potential toguide surgical planning (71–75). 3D mitral valveparametric maps can now be generated by usingautomated software from several vendors (Figure 7)and can be either fully automated or semi-automated(Table 7). Semi-automated analysis in patients withdegenerative mitral valve disease has shown thatquantitative annular circumference is associated with

TABLE 8 Recommendations and Assessment of Evidence Base for Use of Automated Echocardiography

2D LVEF Moderate number of studies with moderate number of patients with uniform direction of effect� Use of automated 2D LVEF measurements is reasonable for assessing left ventricular systolic function for purposes of

improved time efficiency and reproducibility� Use of software that has been validated in peer-reviewed published reports is recommended� Tracking quality should be visually assessed, with contour adjustment as necessary� TTE report should specify that automated LVEF measurement was used� Caution is advised in cases of limited image quality, significant arrhythmia, left ventricular aneurysm, and congenital

heart disease, as few published studies recruited these patients

3D LVEF Moderate number of studies with moderate number of patients with uniform direction of effect� Use of automated 3D LVEF measurements is reasonable for assessing left ventricular systolic function for purposes of

improved time efficiency and reproducibility� If available, preference for automated 3D LVEF should be given over automated 2D LVEF� All other recommendations for automated 2D LVEF apply to automated 3D LVEF

GLS Moderate number of studies demonstrating clinical use� It is reasonable that automated GLS measurements be used� Commercial package that has been validated in peer-reviewed study should be used� Manual GLS strain measurements are not recommended for clinical use

SV and RV Small number of single-center studies suggestive of clinical use� Use of automated SV/RV measurements for quantification of mitral RV and transmitral SV needs further validation

before routine clinical use

PISA Small number of single-center studies that included relatively small number of patients� Use of automated PISA measurements for quantification of mitral RV and EROA needs further validation before routine

clinical use

Aortic annulusand root

Small number of single-center studies that included a relatively small number of patients� It would be reasonable to use automated aortic root and annulus measurements for purposes of improved time

efficiency and reproducibility; however, further experience is needed before routine clinical use� Automated measurements should only be taken using datasets obtained by 3D TEE at present� Caution should be exercised in cases of reduced image quality and high-grade annular calcification as manual contour

adjustments are more likely to be necessary

MV anatomy Small number of single-center studies with small number of patients that did not include variety of MV conditions� Use of automated mitral valvular dimensional measurements requires further validation before clinical use� Automated measurements should only be taken using datasets obtained by 3D TEE at present

TTE ¼ transthoracic echocardiogram; other abbreviations as in Tables 1, 4, and 5.


1087

implanted annuloplasty band length, whereas poste-rior mitral valve leaflet segment-2 length and areawas associated with performance of intraoperativeleaflet resection (72,76). Similar to automation inother areas of echocardiography, intraobserver andinterobserver variability for semi-automated mea-surements are reportedly good, with ICCs rangingfrom 0.83 to 0.99 depending on the structure beingmeasured (76). In addition, fully automated methodsresult in >75% reduction in the time for analysiscompared with manual methods.

CHALLENGES AND KNOWLEDGE GAPS IN

AUTOMATED QUANTIFICATION

Despite the enthusiasm for automated measurements,these techniques have not been widely embraced byechocardiography laboratories due to several limita-tions. First, it is challenging to generalize the existingdata to routine clinical practice in which patients withpoor to modest image quality, arrhythmia, congenitalheart disease, abnormal cardiac chamber configura-tion, and multivalvular disease are common. There-fore, for all automated methods, pragmatic studies

that include patient cohorts not enrolled in previousstudies yet routinely seen in clinical practice areneeded. This approach would allow clear demonstra-tion of where automated techniques have the greatestaccuracy and reproducibility. Second, several of theautomated quantification techniques described earlier(e.g., automated stroke volume, 3D PISA, aorticmodeling techniques) have only been assessed in smallstudies with selected populations and require vendor-specific software for analysis. These factors limitwidespread application. Third, several knowledgegaps in automated quantification need to be addressedvia further research. For automated 2D LVEF, 3D LVEF,and GLS measurements, studies to determine inter-vendor agreement and multicenter reproducibility areneeded. This is critical for multivendor laboratories sothat clinicians can confidentially monitor longitudinalchanges in function when measurements are notpossible using the same vendor or even at the sameinstitution at every visit. For stroke volume, PISA, andmitral and aortic valve/root analysis, further addi-tional data on accuracy and reproducibility in largercohort of patients (especially in multicenter settings)are necessary. In addition, the hypothesis of workflow

FIGURE 8 Incorporation of Automated Techniques in Routine Echocardiography

1) Choose vendor software, preferably one withpeer-reviewed evidence base

3) Initiate physician and sonographer training on echocardiographicdatasets before going live (at least 20-30 individual patientdatasets recommended)

4) Adjust echo report template settings to allow for descriptionof automated measurements

5) If sequential measurements over time are required, werecommend using same vendor software for each study

6) Make automated measurements part of local qualityassurance (QA) program, including:• Routine storage of automated cine images as part of echo study• At quarterly QA meetings, assess institutional LVEF variability• Update vendor software regularly as needed

2) Establish local practice guidelines for following considerations:• Patient subsets that should be excluded from automated imaging (e.g. arrhythmia, complex congenital heart disease)• Determination of whether sonographer or physician should initiate automation process• Billing protocol for novel technology

Flowchart of steps required for echocardiography laboratories to incorporate automated

measurements in routine practice. LVEF ¼ left ventricular ejection fraction.



1088

improvement with automated quantification must beprospectively assessed in busy echocardiography lab-oratories. Finally, cost-effectiveness data wouldprompt organizations and funding bodies to encouragethe implementation of these automated algorithms inroutine clinical practice.

Another crucial consideration with automatedquantification is the balance between reproducibilityand accuracy. The primary benefit of automatedquantification is its reproducibility. Although thedescribed studies have shown modest to good accu-racy at the group level, at an individual patient levelaccuracy may be different. Studies comparing manualversus automated quantification on individual pa-tient management decisions and clinical outcomesare therefore needed to provide the strongest evi-dence for its adoption. Until such studies becomeavailable, when an experienced clinician believes that

automated measurements are inaccurate, a carefulassessment of validity of the data is needed byexamining the image quality, frame rate, contours,and automated tracking. Furthermore, data fromautomated quantification should be considered in thecontext of all other available data and the clinicalcontext. Finally, if clinical uncertainty remains, analternative cardiac imaging modality should beconsidered.

ADOPTION OF AUTOMATED METHODS IN

ROUTINE ECHOCARDIOGRAPHY

Table 8 summarizes the evidence for the use ofautomated techniques and provides recommenda-tions for its use. Despite the challenges and knowl-edge gaps described earlier, we believe that there isadequate evidence to support the clinical use of semi-automated or fully automated analysis of 2D LVEF, 3DLVEF, and measurements of GLS in populationssimilar to those in published studies (Tables 1 to 3).Before use of automated 2D LVEF or GLS measure-ments, it is prudent to ensure adequate imagequality, including adequate endocardial definition,non-foreshortened views, minimal to no dropout,or artifacts. For GLS measurements, acquisition of all3 apical views sequentially with similar heart ratesand frame rates (>40 frames/s) is essential. Forautomated 3D LVEF, in addition to good image qual-ity, volume rates should ideally exceed 20 volumes/s.Once the automated algorithm is applied for LVEF orstrain measurements, it is important to visuallyassess tracking quality and make contour adjust-ments if necessary. However, unless there are majorerrors in fully automated contours and adjustmentsare made by experienced observers, contour adjust-ments should be minimized. Other automated quan-tification techniques described will becomeimportant in the assessment of valvular heart disease,especially with a growing need for reproduciblequantification of valvular disease severity and toguide percutaneous and minimally invasive proced-ures. However, a larger body of data in broader pa-tient populations is required before routine clinicalapplication.

An approach for echocardiography laboratories toincorporate automated techniques as part of theirworkflow has been outlined in Figure 8. The types ofautomated quantification techniques available willdepend on the vendor and the version of the echo-cardiography machines being used. Each laboratoryshould establish practice guidelines by identifyingpatient subsets that should be included and excluded

HIGHLIGHTS

� Automated quantification is a potentialavenue to improve accuracy and repro-ducibility of routine echocardiographymeasurements.

� Automation is possible for2-/3-dimensional left ventricular ejectionfraction measurements and valvular dis-ease quantification.

� Automated quantification consistentlyshows time efficiency of analysis andimproves reproducibility.

� Future studies need to assess intervendoragreement and performance in specialpatient populations.


1089

from automated quantification. This approach shouldbe based on patients enrolled in published studies.Analysis of the acquired data should ideally be per-formed immediately so that additional images couldbe obtained in case of poor tracking. Physicians andsonographers should be trained on 20 to 30 echocar-diographic datasets with the focus on what constitutesadequate automated analysis. Echocardiographytemplates may need to be modified to accommodatethe automated measurement. It is also important touse the same vendor’s automated techniques for lon-gitudinal follow-up of ejection fraction or strainmeasurements because there are limited data onintervendor comparisons with 2D LVEF, and there areknown vendor differences with GLS measurements(77). Interestingly, for 3D LVEF, a recent studycomparing 3 vendors by using semi-automated algo-rithms has reported good intervendor agreement (26).Therefore, intervendor differences may be less of anissue for 3D LVEF measurements. It is also importantto ensure that the echocardiography image data arestored in a format (e.g., uncompressed format) thatcould be re-analyzed by the reporting physician ifthere is disagreement with the automated analysis.Finally, laboratories should incorporate a local qualityassurance program to verify robust automated anal-ysis processes and to ensure that any software up-grades are carefully considered, especially if they willhave an impact on measurement accuracy andreproducibility.

FUTURE DIRECTIONS

Significant advances in automated quantification inechocardiography are anticipated over the nextdecade. First, current knowledge gaps are expected tobe addressed by ongoing studies. We anticipateimprovement in 3D and 3D color Doppler technologythat will better lend itself to automated quantifica-tion. This includes improvement in spatial andtemporal resolution, better contrast-to-noise andsignal-to-noise ratios, and enhancement in 3D imagevisualization. Virtual reality image visualization, forexample, may allow the user to better appreciate theadequacy of endocardial tracking of the entire 3Dvolumes. Simultaneously, automated algorithms areexpected to improve through the greater use of arti-ficial intelligence approaches. With these advances,automated quantification will likely extend beyond asingle-chamber analysis to provide simultaneousassessment of the size and function (ejection fractionand strain) of all cardiac chambers and great arteriesand blood flow simultaneously.

Furthermore, practice guidelines addressing theissues of automation and recommendations for stan-dardization of analysis techniques can possiblyreduce intervendor differences in measurements.Finally, with the expansion in the use of point-of-careultrasound techniques, automated quantification canmove this field from a visual assessment tool to onethat provides reproducible quantitative diagnosticdata to enhance rapid triaging of patients.

CONCLUSIONS

Automated quantification in echocardiography hasthe potential to improve workflow in busy echocardi-ography laboratories and enhance the accuracy andreproducibility of quantitative parameters. However,despite the growing body of data, widespread appli-cation has been hampered by limited data on accuracyand reproducibility in patients seen in routine clinicalpractice (e.g., those with modest image quality orarrhythmia), vendor-specific availability of sometechniques, and several knowledge gaps. We antici-pate that much of these limitations will be addressedwith ongoing studies. However, given the existingdata, we believe that automated quantification of 2DLVEF, 3D LVEF, and GLS should be incorporated intoroutine echocardiography using a structured approachas we have suggested. We anticipate significant ad-vancements in automated quantification in the nextdecade that will allow clinicians to focus less on theprocess of measurements and shift their attention todata quality, data synthesis, and diagnosis. To fullyreach the potential of automated quantification,innovative partnerships between physicians,



1090

technologists, software engineers, and industry needto be encouraged and supported.

ACKNOWLEDGMENTS The authors acknowledgeDr. Shizhen Liu, Dr. Mani Vannan, Babitha Tham-pinathan, and Helene Houle for providing some ofthe images used in the manuscript.

ADDRESS FOR CORRESPONDENCE: Dr. PaaladineshThavendiranathan, Division of Cardiology, PeterMunk Cardiac Centre, Toronto General Hospital,4N-490, 200 Elizabeth Street, Toronto, Ontario M5G 2C4,Canada. E-mail: [email protected]: @dineshpmcc1.

RE F E RENCE S

1. Andrus BW, Welch HG. Medicare services pro-vided by cardiologists in the United States: 1999-2008. Circ Cardiovasc Qual Outcomes 2012;5:31–6.

2. Virnig BA, Shippee ND, O’Donnell B, Zeglin J,Parashuram S. Trends in the use of echocardiog-raphy, 2007 to 2011: Data Points #20. Data PointsPublication Series. Agency for Healthcare Researchand Quality (U.S.); Rockville, MD: 2011.

3. Thavendiranathan P, Grant AD, Negishi T,Plana JC, Popovic ZB, Marwick TH. Reproducibilityof echocardiographic techniques for sequentialassessment of left ventricular ejection fraction andvolumes: application to patients undergoing can-cer chemotherapy. J Am Coll Cardiol 2013;61:77–84.

4. Kaufmann BA, Min SY, Goetschalckx K, et al.How reliable are left ventricular ejection fractioncut offs assessed by echocardiography for clinicaldecision making in patients with heart failure? Int JCardiovasc Imaging 2013;29:581–8.

5. Lang RM, Badano LP, Mor-Avi V, et al. Recom-mendations for cardiac chamber quantification byechocardiography in adults: an update from theAmerican Society of Echocardiography and theEuropean Association of Cardiovascular Imaging.J Am Soc Echocardiogr 2015;28:1–39.e14.

6. Zolgharni M, Dhutia NM, Cole GD, et al. Auto-mated aortic Doppler flow tracing for reproducibleresearch and clinical measurements. IEEE TransMed Imaging 2014;33:1071–82.

7. Dhutia NM, Zolgharni M, Mielewczik M, et al.Open-source, vendor-independent, automatedmulti-beat tissue Doppler echocardiography anal-ysis. Int J Cardiovasc Imaging 2017;33:1135–48.

8. Maret E, Brudin L, Lindstrom L, Nylander E,Ohlsson JL, Engvall JE. Computer-assisted deter-mination of left ventricular endocardial bordersreduces variability in the echocardiographicassessment of ejection fraction. Cardiovasc Ultra-soun 2008;6:55.

9. Rahmouni HW, Ky B, Plappert T, et al. Clinicalutility of automated assessment of left ventricularejection fraction using artificial intelligence-assisted border detection. Am Heart J 2008;155:562–70.

10. Knackstedt C, Bekkers SC, Schummers G, et al.Fully automated versus standard tracking of leftventricular ejection fraction and longitudinalstrain: the FAST-EFs Multicenter Study. J Am CollCardiol 2015;66:1456–66.

11. Frederiksen CA, Juhl-Olsen P, Hermansen JF,Andersen NH, Sloth E. Clinical utility of semi-

automated estimation of ejection fraction at thepoint-of-care. Heart Lung Vessel 2015;7:208–16.

12. Hovnanians N, Win T, Makkiya M, Zheng Q,Taub C. Validity of automated measurement of leftventricular ejection fraction and volume using thePhilips EPIQ system. Echocardiography 2017;34:1575–83.

13. Szulik M, Pappas CJ, Jurcut R, et al. Clinicalvalidation of a novel speckle-tracking-basedejection fraction assessment method. J Am SocEchocardiogr 2011;24:1092–100.

14. Abazid RM, Abohamr SI, Smettei OA, et al.Visual versus fully automated assessment of leftventricular ejection fraction. Avicenna J Med 2018;8:41–5.

15. Aurich M, Keller M, Greiner S, et al. Left ven-tricular mechanics assessed by two-dimensionalechocardiography and cardiac magnetic reso-nance imaging: comparison of high-resolutionspeckle tracking and feature tracking. Eur Heart JCardiovasc Imaging 2016;17:1370–8.

16. Cannesson M, Tanabe M, Suffoletto MS, et al.A novel two-dimensional echocardiographic imageanalysis system using artificial intelligence-learned pattern recognition for rapid automatedejection fraction. J Am Coll Cardiol 2007;49:217–26.

17. Mantovani G, Madeddu C, Cadeddu C, et al.Persistence, up to 18 months of follow-up, ofepirubicin-induced myocardial dysfunction detec-ted early by serial tissue Doppler echocardiogra-phy: correlation with inflammatory and oxidativestress markers. Oncologist 2008;13:1296–305.

18. Aurich M, Andre F, Keller M, et al. Assessmentof left ventricular volumes with echocardiographyand cardiac magnetic resonance imaging: real-lifeevaluation of standard versus new semiautomaticmethods. J Am Soc Echocardiogr 2014;27:1017–24.

19. Dorosz JL, Lezotte DC, Weitzenkamp DA,Allen LA, Salcedo EE. Performance of 3-dimensional echocardiography in measuring leftventricular volumes and ejection fraction: a sys-tematic review and meta-analysis. J Am Coll Car-diol 2012;59:1799–808.

20. Shibayama K, Watanabe H, Iguchi N, et al.Evaluation of automated measurement of leftventricular volume by novel real-time 3-dimensional echocardiographic system: validationwith cardiac magnetic resonance imaging and 2-dimensional echocardiography. J Cardiol 2013;61:281–8.

21. Tsang W, Salgo IS, Medvedofsky D, et al.Transthoracic 3D echocardiographic left heart

chamber quantification using an automatedadaptive analytics algorithm. J Am Coll CardiolImg 2016;9:769–82.

22. Plana JC, Galderisi M, Barac A, et al. Expertconsensus for multimodality imaging evaluation ofadult patients during and after cancer therapy: areport from the American Society of Echocardi-ography and the European Association of Cardio-vascular Imaging. Eur Heart J Cardiovasc Imaging2014;15:1063–93.

23. Lang RM, Badano LP, Mor-Avi V, et al. Rec-ommendations for cardiac chamber quantificationby echocardiography in adults: an update fromthe American Society of Echocardiography andthe European Association of Cardiovascular Im-aging. Eur Heart J Cardiovasc Imaging 2015;16:233–70.

24. Barbosa D, Dietenbeck T, Heyde B, et al. Fastand fully automatic 3-d echocardiographic seg-mentation using B-spline explicit active surfaces:feasibility study and validation in a clinical setting.Ultrasound Med Biol 2013;39:89–101.

25. Muraru D, Badano LP, Piccoli G, et al. Valida-tion of a novel automated border-detection al-gorithm for rapid and accurate quantitation of leftventricular volumes based on three-dimensionalechocardiography. Eur J Echocardiogr 2010;11:359–68.

26. Muraru D, Cecchetto A, Cucchini U, et al.Intervendor consistency and accuracy of leftventricular volume measurements using three-dimensional echocardiography. J Am Soc Echo-cardiogr 2018;31:158–68.e1.

27. Spitzer E, Ren B, Zijlstra F, Mieghem NMV,Geleijnse ML. The role of automated 3d echocar-diography for left ventricular ejection fractionassessment. Card Fail Rev 2017;3:97–101.

28. Tamborini G, Piazzese C, Lang RM, et al.Feasibility and accuracy of automated software fortransthoracic three-dimensional left ventricularvolume and function analysis: comparisons withtwo-dimensional echocardiography, three-dimensional transthoracic manual method, andcardiac magnetic resonance imaging. J Am SocEchocardiogr 2017;30:1049–58.

29. Levy F, Dan Schouver E, Iacuzio L, et al. Per-formance of new automated transthoracic three-dimensional echocardiographic software for leftventricular volumes and function assessment inroutine clinical practice: comparison with 3 Teslacardiac magnetic resonance. Arch Cardiovasc Dis2017;110:580–9.

30. Thavendiranathan P, Liu S, Verhaert D, et al.Feasibility, accuracy, and reproducibility of real-

mailto:[email protected]

https://twitter.com/dineshpmcc1

http://refhub.elsevier.com/S1936-878X(19)30339-0/sref1





























































































































































1091

time full-volume 3D transthoracic echocardiogra-phy to measure LV volumes and systolic function:a fully automated endocardial contouring algo-rithm in sinus rhythm and atrial fibrillation. J AmColl Cardiol Img 2012;5:239–51.

31. Spitzer E, Ren B, Soliman OI, Zijlstra F, VanMieghem NM, Geleijnse ML. Accuracy of an auto-mated transthoracic echocardiographic tool for 3Dassessment of left heart chamber volumes. Echo-cardiography 2017;34:199–209.

32. Medvedofsky D, Kebed K, Laffin L, et al.Reproducibility and experience dependence ofechocardiographic indices of left ventricularfunction: side-by-side comparison of global lon-gitudinal strain and ejection fraction. Echocardi-ography 2017;34:365–70.

33. Medvedofsky D, Mor-Avi V, Amzulescu M,et al. Three-dimensional echocardiographic quan-tification of the left-heart chambers using anautomated adaptive analytics algorithm: multi-centre validation study. Eur Heart J CardiovascImaging 2018;19:47–58.

34. Sun L, Feng H, Ni L, Wang H, Gao D. Realiza-tion of fully automated quantification of leftventricular volumes and systolic function usingtransthoracic 3D echocardiography. CardiovascUltrasoun 2018;16:2.

35. Otani K, Nakazono A, Salgo IS, Lang RM,Takeuchi M. Three-dimensional echocardiographicassessment of left heart chamber size and functionwith fully automated quantification software inpatients with atrial fibrillation. J Am Soc Echo-cardiogr 2016;29:955–65.

36. Thavendiranathan P, Poulin F, Lim KD,Plana JC, Woo A, Marwick TH. Use of myocardialstrain imaging by echocardiography for the earlydetection of cardiotoxicity in patients during andafter cancer chemotherapy: a systematic review.J Am Coll Cardiol 2014;63:2751–68.

37. Dahl JS, Videbaek L, Poulsen MK, Rudbaek TR,Pellikka PA, Moller JE. Global strain in severeaortic valve stenosis: relation to clinical outcomeafter aortic valve replacement. Circ CardiovascImaging 2012;5:613–20.

38. Stanton T, Leano R, Marwick TH. Prediction ofall-cause mortality from global longitudinalspeckle strain: comparison with ejection fractionand wall motion scoring. Circ Cardiovasc Imaging2009;2:356–64.

39. Heimdal A, Stoylen A, Torp H, Skjaerpe T.Real-time strain rate imaging of the left ventricleby ultrasound. J Am Soc Echocardiogr 1998;11:1013–9.

40. Ingul CB, Torp H, Aase SA, Berg S, Stoylen A,Slordahl SA. Automated analysis of strain rate andstrain: feasibility and clinical implications. J AmSoc Echocardiogr 2005;18:411–8.

41. Amzulescu MS, Langet H, Saloux E, et al.Head-to-head comparison of global and regionaltwo-dimensional speckle tracking strain versuscardiac magnetic resonance tagging in a multi-center validation study. Circ Cardiovasc Imaging2017;10.

42. Brown J, Jenkins C, Marwick TH. Use ofmyocardial strain to assess global left ventricularfunction: a comparison with cardiac magnetic

resonance and 3-dimensional echocardiography.Am Heart J 2009;157:102.e1–5.

43. Delgado V, Mollema SA, Ypenburg C, et al.Relation between global left ventricular longitu-dinal strain assessed with novel automated func-tion imaging and biplane left ventricular ejectionfraction in patients with coronary artery disease.J Am Soc Echocardiogr 2008;21:1244–50.

44. Villanueva-Fernandez E, Ruiz-Ortiz M, Mesa-Rubio D, et al. Feasibility of bidimensionalspeckle-tracking echocardiography for strainanalysis in consecutive patients in daily clinicalpractice. Echocardiography 2012;29:923–6.

45. Belghitia H, Brette S, Lafitte S, et al. Auto-mated function imaging: a new operator-independent strain method for assessing leftventricular function. Arch Cardiovasc Dis 2008;101:163–9.

46. Negishi T, Negishi K, Thavendiranathan P,et al. Effect of experience and training on theconcordance and precision of strain measure-ments. J Am Coll Cardiol Img 2017;10:518–22.

47. Kleijn SA, Aly MF, Terwee CB, van Rossum AC,Kamp O. Three-dimensional speckle trackingechocardiography for automatic assessment ofglobal and regional left ventricular function basedon area strain. J Am Soc Echocardiogr 2011;24:314–21.

48. Zoghbi WA, Adams D, Bonow RO, et al. Rec-ommendations for noninvasive evaluation ofnative valvular regurgitation: a report from theAmerican Society of Echocardiography developedin collaboration with the Society for Cardiovascu-lar Magnetic Resonance. J Am Soc Echocardiogr2017;30:303–71.

49. Thavendiranathan P, Liu S, Datta S, et al.Automated quantification of mitral inflow andaortic outflow stroke volumes by three-dimensional real-time volume color-flow Dopplertransthoracic echocardiography: comparison withpulsed-wave Doppler and cardiac magnetic reso-nance imaging. J Am Soc Echocardiogr 2012;25:56–65.

50. Shimada E, Zhu M, Kimura S, et al. Quantita-tive assessment of mitral inflow and aortic outflowstroke volumes by 3-dimensional real-time full-volume color flow Doppler transthoracic echo-cardiography: an in vivo study. J Ultrasound Med2015;34:95–103.

51. Kato A, Sandoval JP, Mroczek D, et al. Auto-mated 3-dimensional single-beat real-time vol-ume colour flow Doppler echocardiography inchildren: a validation study of right and left heartflows. Can J Cardiol 2018;34:726–35.

52. Matthews F, Largiader T, Rhomberg P, van derLoo B, Schmid ER, Jenni R. A novel operator-independent algorithm for cardiac outputmeasurements based on three-dimensional trans-oesophageal colour Doppler echocardiography.Eur J Echocardiogr 2010;11:432–7.

53. Son JW, Chang HJ, Lee JK, et al. Automatedquantification of mitral regurgitation by three-dimensional real time full volume color Dopplertransthoracic echocardiography: a validation withcardiac magnetic resonance imaging and compar-ison with two-dimensional quantitative methods.J Cardiovasc Ultrasound 2013;21:81–9.

54. Thavendiranathan P, Liu S, Datta S, et al.Quantification of chronic functional mitral regur-gitation by automated 3-dimensional peak andintegrated proximal isovelocity surface area andstroke volume techniques using real-time3-dimensional volume color Doppler echocardi-ography: in vitro and clinical validation. Circ Car-diovasc Imaging 2013;6:125–33.

55. Choi J, Hong GR, Kim M, et al. Automaticquantification of aortic regurgitation using 3D fullvolume color Doppler echocardiography: a vali-dation study with cardiac magnetic resonanceimaging. Int J Cardiovasc Imaging 2015;31:1379–89.

56. Heo R, Son JW, Ó Hartaigh B, et al. Clinicalimplications of three-dimensional real-time colorDoppler transthoracic echocardiography in quan-tifying mitral regurgitation: a comparison withconventional two-dimensional methods. J Am SocEchocardiogr 2017;30:393–403.e7.

57. Gruner C, Herzog B, Bettex D, et al. Quantifi-cation of mitral regurgitation by real time three-dimensional color Doppler flow echocardiographypre- and post-percutaneous mitral valve repair.Echocardiography 2015;32:1140–6.

58. Biner S, Rafique A, Rafii F, et al. Reproduc-ibility of proximal isovelocity surface area, venacontracta, and regurgitant jet area for assessmentof mitral regurgitation severity. J Am Coll CardiolImg 2010;3:235–43.

59. Plicht B, Kahlert P, Goldwasser R, et al. Directquantification of mitral regurgitant flow volumeby real-time three-dimensional echocardiographyusing dealiasing of color Doppler flow at the venacontracta. J Am Soc Echocardiogr 2008;21:1337–46.

60. Brugger N, Wustmann K, Hurzeler M, et al.Comparison of three-dimensional proximal iso-velocity surface area to cardiac magnetic reso-nance imaging for quantifying mitralregurgitation. Am J Cardiol 2015;115:1130–6.

61. Sitges M, Jones M, Shiota T, et al. Real-timethree-dimensional color Doppler evaluation of theflow convergence zone for quantification of mitralregurgitation: validation experimental animalstudy and initial clinical experience. J Am SocEchocardiogr 2003;16:38–45.

62. Schmidt FP, Gniewosz T, Jabs A, et al. Use-fulness of 3D-PISA as compared to guidelineendorsed parameters for mitral regurgitationquantification. Int J Cardiovasc Imaging 2014;30:1501–8.

63. Choi J, Heo R, Hong GR, et al. Differentialeffect of 3-dimensional color Doppler echocardi-ography for the quantification of mitral regurgi-tation according to the severity andcharacteristics. Circ Cardiovasc Imaging 2014;7:535–44.

64. de Agustin JA, Viliani D, Vieira C, et al. Prox-imal isovelocity surface area by single-beat three-dimensional color Doppler echocardiographyapplied for tricuspid regurgitation quantification.J Am Soc Echocardiogr 2013;26:1063–72.

65. Mediratta A, Addetia K, Medvedofsky D, et al.3D echocardiographic analysis of aortic annulusfor transcatheter aortic valve replacement usingnovel aortic valve quantification software:











































































































































































































1092

comparison with computed tomography. Echo-cardiography 2017;34:690–9.

66. Calleja A, Thavendiranathan P, Ionasec RI,et al. Automated quantitative 3-dimensionalmodeling of the aortic valve and root by 3-dimensional transesophageal echocardiography innormals, aortic regurgitation, and aortic stenosis:comparison to computed tomography in normalsand clinical implications. Circ Cardiovasc Imaging2013;6:99–108.

67. Prihadi EA, van Rosendael PJ, Vollema EM,Bax JJ, Delgado V, Ajmone Marsan N. Feasibility,accuracy, and reproducibility of aortic annular androot sizing for transcatheter aortic valve replace-ment using novel automated three-dimensionalechocardiographic software: comparison withmulti-detector row computed tomography. J AmSoc Echocardiogr 2018;31:505–14.e3.

68. Garcia-Martin A, Lazaro-Rivera C, Fernandez-Golfin C, et al. Accuracy and reproducibility ofnovel echocardiographic three-dimensional auto-mated software for the assessment of the aorticroot in candidates for transcatheter aortic valvereplacement. Eur Heart J Cardiovasc Imaging2016;17:772–8.

69. Bersvendsen J, Beitnes JO, Urheim S,Aakhus S, Samset E. Automatic measurement ofaortic annulus diameter in 3-dimensional trans-oesophageal echocardiography. BMC Med Imaging2014;14:31.

70. Khalique OK, Kodali SK, Paradis JM, et al.Aortic annular sizing using a novel 3-dimensionalechocardiographic method: use and comparisonwith cardiac computed tomography. Circ Car-diovasc Imaging 2014;7:155–63.

71. Tsang W, Weinert L, Sugeng L, et al. The valueof three-dimensional echocardiography derivedmitral valve parametric maps and the role ofexperience in the diagnosis of pathology. J Am SocEchocardiogr 2011;24:860–7.

72. Grewal J, Suri R, Mankad S, et al. Mitralannular dynamics in myxomatous valve disease:new insights with real-time 3-dimensional echo-cardiography. Circulation 2010;121:1423–31.

73. Biaggi P, Jedrzkiewicz S, Gruner C, et al.Quantification of mitral valve anatomy by three-dimensional transesophageal echocardiography inmitral valve prolapse predicts surgical anatomyand the complexity of mitral valve repair. J Am SocEchocardiogr 2012;25:758–65.

74. Chandra S, Salgo IS, Sugeng L, et al. Charac-terization of degenerative mitral valve diseaseusing morphologic analysis of real-time three-dimensional echocardiographic images: objectiveinsight into complexity and planning of mitralvalve repair. Circ Cardiovasc Imaging 2011;4:24–32.

75. Ender J, Eibel S, Mukherjee C, et al. Predictionof the annuloplasty ring size in patients under-going mitral valve repair using real-time three-dimensional transoesophageal echocardiography.Eur J Echocardiogr 2011;12:445–53.

76. Calleja A, Poulin F, Woo A, et al. Quantitativemodeling of the mitral valve by three-dimensionaltransesophageal echocardiography in patientsundergoing mitral valve repair: correlation withintraoperative surgical technique. J Am SocEchocardiogr 2015;28:1083–92.

77. Farsalinos KE, Daraban AM, Unlu S,Thomas JD, Badano LP, Voigt JU. Head-to-headcomparison of global longitudinal strain mea-surements among nine different vendors: theEACVI/ASE Inter-Vendor Comparison Study. J AmSoc Echocardiogr 2015;28:1171–81.e2.

78. de Agustin JA, Marcos-Alberca P, Fernandez-Golfin C, et al. Direct measurement of proximalisovelocity surface area by single-beat three-dimensional color Doppler echocardiography inmitral regurgitation: a validation study. J Am SocEchocardiogr 2012;25:815–23.

79. Queiros S, Papachristidis A, Morais P, et al.Fully automatic 3-D-TEE segmentation for theplanning of transcatheter aortic valve implanta-tion. IEEE Trans Biomed Eng 2017;64:1711–20.

80. Podlesnikar T, Prihadi EA, van Rosendael PJ,et al. Influence of the quantity of aortic valve cal-cium on the agreement between automated 3-dimensional transesophageal echocardiographyand multidetector row computed tomography foraortic annulus sizing. Am J Cardiol 2018;121:86–93.

81. Pouch AM, Wang H, Takabe M, et al. Fullyautomatic segmentation of the mitral leaflets in3D transesophageal echocardiographic imagesusing multi-atlas joint label fusion and deformablemedial modeling. Med Image Anal 2014;18:118–29.

82. Kagiyama N, Toki M, Hara M, et al. Efficacy andaccuracy of novel automated mitral valve quanti-fication: three-dimensional transesophagealechocardiographic study. Echocardiography 2016;33:756–63.

83. Jin CN, Salgo IS, Schneider RJ, et al. Auto-mated quantification of mitral valve anatomy us-ing anatomical intelligence in three-dimensionalechocardiography. Int Jf Cardiol 2015;199:232–8.

84. Kato N, Shibayama K, Noguchi M, et al. Su-periority of novel automated assessment of aorticannulus by intraoperative three-dimensionaltransesophageal echocardiography in patientswith severe aortic stenosis: Comparison withconventional cross-sectional assessment. J Cardiol2018;72:321–7.

85. Aquila I, Gonzalez A, Fernandez-Golfin C, et al.Reproducibility of a novel echocardiographic 3Dautomated software for the assessment of mitralvalve anatomy. Cardiovasc Ultrasoun 2016;14:17.

KEY WORDS automation,echocardiography





















































































































Automated Quantification in Echocardiography · 2019-05-25 · cardiac magnetic resonance imaging....

Documents

Transcript of Automated Quantification in Echocardiography · 2019-05-25 · cardiac magnetic resonance imaging....