MEDICAL_IMAGE_PROCESSING_25sep06.pdf

8/10/2019 MEDICAL_IMAGE_PROCESSING_25sep06.pdf

1/7

MEDICAL IMAGE PROCESSINGK.M.M. Rao, V.D.P. Rao

Medical Imaging

Imaging technology in Medicine made thedoctors to see the interior portions of the bodyfor easy diagnosis. It also helped doctors tomake keyhole surgeries for reaching theinterior parts without really opening too muchof the body. CT Scanner, Ultrasound andMagnetic Resonance Imaging took over x-rayimaging by making the doctors to look at thebody's elusive third dimension. With the CTScanner, body's interior can be bared withease and the diseased areas can be identifiedwithout causing either discomfort or pain tothe patient. MRI picks up signals from thebody's magnetic particles spinning to its

magnetic tune and with the help of itspowerful computer, converts scanner data intorevealing pictures of internal organs. ImageProcessing techniques developed foranalyzing remote sensing data may bemodified to analyze the outputs of medicalimaging systems to get best advantage toanalyze symptoms of the patients with ease.

Advantages of Digital Processing forMedical Applications

Digital data will not change when it is

reproduced any number of times andretains the originality of the data.

Offers a powerful tool to physicians byeasing the search for re-presentativeimages;

Displaying images immediately afteracquiring;

Enhancement of images to make themeasier for the Physician to interpret;

Quantifying changes over time;

Providing a set of images for teaching todemonstrate examples of diseases orfeatures in any image;

Quick comparison of images.

Digital Image Processing Requirements forMedical Applications

Interfacing Analog outputs of sensorssuch as microscopes, endoscopes,ultrasound etc., to digitizers and in turn toDigital Image Processing systems.

Image enhancements.

Changing density dynamic range of B/W

images. Color correction in color images.

Manipulating of colors within an image.

Contour detection.

Area calculations of the cells of abiomedical image.

Display of image line profile.

Restoration of images.

Smoothing of images.

Registration of multiple images andmosaicing.

Construction of 3-D images from 2-Dimages.

Generation of negative images. Zooming of images.

Pseudo coloring.

Point to point measurements.

Getting relief effect.

Removal of artifacts from the image.

Advanced Digital Image ProcessingTechniques

Neural Network based Image Processing.

Statistical approach for texture analysis.

Segmentation in color and B/W images.

Expert system based Image Processing. Application of object oriented

programming techniques in ImageProcessing environments.

Shape in machine vision.

Multispectral classification techniques.

Auto focussing techniques for MRIimages.

Thresholding technique for findingcontours of objects.

Sequential segmentation technique to findout thin vessels in medical images andhair line cracks in NDT.

Fractal method for texture classification. Data compression techniques using

fractals and Descrete CosineTransformers.

Image restoration methods using PointSpread functions and Wiener filter etc.


2/7

Image Processing Systems for MedicalAppl ications

(a) Endoscopy

In each endoscope, there are two fiberbundles. One is used to illuminate the innerstructure of object. Other is used to collect thereflected light from that area. The endoscopeis a tubular optical instrument to inspect orview the body cavities, which are not visible tothe naked eye normally.

For a wider field of view and better imagequality, a telescope system is added in theinternal part of the endoscope. Gastro-intestinal fiberoscopes and laparoscopes areimportant endoscopes used in hospitals forexamination, treatment of diseases andsurgery. Technological advances in

computers and semiconductor chips havebrought about lot of changes in health careduring the last decade. For digestivediseases, this advancement is represented bythe incorporation of charge-coupled device(converts optical image to electronic image)into gastrointestinal endoscopy. These videoendoscopes use xenon arc lamps as lightsource. Color Imaging is achieved byincorporating RGB filters between XenonLamp Supply and the proximal end of theendoscope. The other approach to thegeneration of color image is to divide the

available cells on the CCD among the threeprimary colors by means of filters. Threeimages one for each color are then producedsimultaneously by the CCD.

Endoscopic pictures are converted to digitalimages by using CCD cameras andassociated image digitizer circuits into aPC/AT. The recorded images can be imageprocessed for better quality.

Stereo Endoscope

Two cameras are mounted on a singlelaproscope. Images from the cameras aretransmitted alternately to a video monitor.Few types of display techniques are used torealise stereo images from two-dimensionalimages recorded from the above cameras. Asthe cameras transmits images at 60-120cycles per second a three-dimensional, realtime image is perceived. As the images aretransmitted at a high frequency, the effect is

that of seeing different imagessimultaneously.

(b) ComputerTomography (Ct)

Computerised Axial Tomographyor computertransmission tomography or computertomography is a method of forming imagesfrom X-rays. Measurements are taken fromX-rays transmitted through the body. Thesecontain information on the constituents of thebody in the path of the X-ray beam. By usingmultidirectional scanning of the object,multiple data is collected.

An image of a cross-section of the body isproduced by measuring the total attenuationalong rows and columns of a matrix and thencomputing the attenuation of the matrixelements at the intersections of the rows and

columns. The number of mathematicaloperations necessary to yield clinicallyapplicable and accurate images is so largethat a computer is essential to do them. Theinformation obtained from these computationscan be presented in a conventional rasterform resulting in a two dimensional picture.

The timing, anode voltage and beam currentare controlled by a computer through a controlbus. The high voltage d.c. power supplydrives an X-ray tube that can be mechanicallyrotated along the circumference of a gantry.

The patient lies in a tube through the center ofthe gantry. The X-rays pass through thepatient and are partially absorbed. Theremaining X-ray photons impinge uponseveral radiation detectors fixed around thecircumference of the gantry.

The detector response is directly related tothe number of photons impinging on it andhence to the tissue density. When they strikethe detector, the X-ray photons are convertedto scintillations. The computer senses theposition of the X-ray tube and samples the

output of the detector along a diameter lineopposite to the X-ray tube. A calculationbased on data obtained from a complete scanis made by the computer. The output unit thenproduces a visual image of a transverse planecross-section of the patient on the cathode raytube. These images are also stored intocomputer for image processing.


3/7

(c) UltrasonicImaging System

Ultrasonography is a technique by whichultrasonic energy is used to detect the state ofthe internal body organs. Bursts of ultrasonicenergy are transmitted from a piezo-electric ormagnetostrictive transducer through the skinand into the internal anatomy. When thisenergy strikes an interface between twotissues of different acoustical impedance,reflections (echoes) are returned to thetransducer. The transducer converts thesereflections to an electric signal proportional tothe depth of the interface, which is amplifiedand displayed on an oscilloscope. An imageof the interior structure is constructed basedon the total wave travelling time, the averagesound speed and the energy intensity of thereflected waves. The echoes from the patientbody surface are collected by the receiver

circuit. Proper Depth Gain Compensation(DGC) is given by DGC circuit. The receivedsignals are converted into digital signals andstored in memory. The scan converter controlreceives signals of transducer position and TVsynchronous pulses. It generates X & Yaddress information and feeds to the digitalmemory. The stored digital image signals areprocessed and given to digital-to-analogconverter. Then they are fed to the TVmonitor. These signals are converted to digitalform using frame grabber and can be storedonto PC/AT disk. Wherever the images lack

in contrast and brightness, Image Processingtechniques may be used to get full detailsfrom Ultrasound images. Figure 1 showsUltrasound Imaging System.

Fig 1. Ultrasound Imaging System

Magnetic Resonance Imaging (MRI)

Superconducting magnets are used in MRIsystems to provide strong uniform, steadymagnetic fields. The superconductingmagnetic coils are cooled to liquid heliumtemperature and can produce very high

magnetic fields. Hence the signal to noiseratio of the received signals and image qualityare better than the conventional magnetsused in the MRI systems.

Different gradient coil systems produce a timevarying, controlled spatial non-uniformmagnetic fields in different directions. Thepatient is kept in this gradient field space.There are also transmitter and receiving RFcoils surrounding the site on which the imageis to be constructed. There is a superpositionof a linear magnetic field gradient on to theuniform magnetic field applied to the patient.When this superposition takes place, theresonance frequencies of the processingnucleii will depend on the positions along thedirection of the magnetic field gradient. Thisproduces a one- dimensional projection of thestructure of the three dimensional object. By

taking a series of these projections at differentgradient orientations using X, Y and Zgradient coils a two or three-dimensionalimage can be obtained. The slice of the imagedepends upon the gradient magnetic field.The gradient magnetic field is controlled bycomputer and that field can be positioned inthree time invarient planes (X, Y and Z). Thetransmitter provides the RF signal pulses. Thereceived nuclear magnetic resonance signal ispicked up by the receiver coil and is fed intothe receiver for signal processing. By two-dimensional Fourier Transformation, the

images are constructed by the computer andanalysed using image processing techniques.

Multispectral tissue classification ofMagnetic Resonance Imaging (MRI)

MRI data consists of multiple channels ofindependent but geometrically registeredmedically significant data, it analogous tomultispectral remote sensing data.

Multispectral analysis of proton MRimages may provide tissue characteristic

information encoded therein.

Using well-established methods forcomputer processing of multispectralimages, tissue characterization signaturesare sought; using supervised orunsupervised classification methods.

The principal advantages of Multispectralanalysis of MRI include:


4/7

It is a quantitative means of analyzing

multidimensional image data, such as MRI.

In other applications, multispectral methodshave been useful in identifying subtletiesthat would otherwise be overlooked.

MR images are intrinsically multispectral.The data in a set of MR images is highlyredundant, both geometrically andradiometrically.

Multispectral methods are well developed,have been implemented on computers, forwhich software is readily available that canprocess MR image data efficiently, and canadapt to existing MR scanners.

Associated Problems With X-Ray Imaging

X-ray films have large dynamic range toaccommodate maximum possible details of X-ray image (as shown in Fig.2). Details in thedarker portion and brighter portion getsaturated and are not possible to interpreteasily. X-ray images can be converted intodigital form using X-ray flouroscopy techniqueor by digitising X-ray film using scanners. Byapplying Image Processing techniques, thedigital images can be manipulated for easyinterpretation. Using these techniques,additional X-ray exposures can be reduced to

the patient.

Fig2. Sensitometric curve of X-ray image

Medical Imaging System (MIS)

During the next years, profound changes areexpected in computer and communicationtechnologies that will offer the medicalimaging systems industry a challenge todevelop advanced telemedicine applications

of high performance. Medical industry,vendors, and specialists need to agree on auniversal MIS structure that will provide astack of functions, protocols and interfacessuitable for coordination and management ofhigh-level image consults, reports and reviewactivities.

Most hospital imaging departments have tocomputerise information systems in whichpatient images and reports are to be stored.The stored information can be handled by twomajor types of medical applications, theintegrated Report and Review applications.The former is performed by experts (eg.radiologists) in four steps :

1. Retrieving and viewing images;2. Processing, interpretation and annotation

of the diagnosis;

3. Composition of final diagnostic multimediareport; and

4. Permanent storing in the database.

The latter allows many simultaneous users(authorized patient-care personnel) to view,read, and listen to the diagnostic report; theseusers do not alter the handled data.

However, recent advances in high-speednetworking technology have opened tohospitals the possibility to obtain applicationsenhanced with conferencing services, referred

to as Consultapplications, to geographysicallydistributed users. For example, physicianslocated in rural health care centres couldconsult one or more experts located in theregional hospital of an urban area and gainaccess to patient images and otherinformation through parallel interfacing withthe database of the hospital's imagingdepartment.

The implementation of such applicationsrequires an appropriate infrastructure withhigh-speed networking and image

manipulation facilities. The Consultapplication is achieved by accessing picturearchiving and communication systems(PACS).

Picture Archiving And CommunicationSystems (PACS)

In application domains such as remotesensing, astronomy, cartography,


5/7

meteorology and medical imaging,images comprise the vast majority ofacquired, processed and archived data.

The medical imaging field in particular,has grown substantially in recent years,

and has generated additional interest inmethods and tools for the management,analysis and communication of medicalimage data.

PACS are currently used in many medicalcentres to manage the image data.

Important considerations in the designand implementation of Image Database(IDB) systems are:

Image feature extraction; Image content representation and

organization of stored information; Search and retrieval strategies; User interface design

Recent proposals regarding the design ofIDB systems and the management ofimage data are influenced by the 'object-oriented' approach.

This approach offers a framework withinwhich different types of entities (eg.different kinds of image data) andoperations (eg. image processing

functions, image access mechanisms,etc.) may be uniformly represented as'objects'.

Objects are grouped into classes, whichcan also be objects. Object classes areorganised into hierarchies, thus takingadvantage of the property of 'inheritance'.

Image retrieval using color, shape andtexture.

Content based image retrieval for image

databases.

Region detection in medical images. Eg.locate endocardium boundaries of the leftand right ventricles from gradient - echoMR images.

Current research activities are interested inthe implementation of smaller and more

flexible systems than PACS, calledminiPACS.

Digital Image Processing ForOphthalmology

To analyse retina, optic nerve, pigmentepithelium and choroid in the ocularfundus.

Color slides have a resolution of 4000 x3000 pixels.

Fluorescein Angiograms have a resolutionof 1800 x 1350 pixels.

Common standard digital cameras haveresolution of 512 x 480, which may besufficient for obtaining relevantinformation of blood vessels etc. (Presentday technology: 2048 x 2048 elementresolution cameras).

8-bit resolution (indicative of contrast) issufficient for most of the Opthalmologyimages.

Indo Cyanine Green (Icg) Imaging

ICG Angiography is basically similar tothat of fluorescein angiography.

Differences are :

Spectral characteristics and, Permeability from choroidal

capillaries.

Sodium fluorescein dye used influorescein angiography has a maximumabsorption at 485nm and peak emissionat 520nm.

The largest portion of excitationand emission energy of thisvisible light is absorbed by the

retinal pigment epithelium andmacular xanthopyll.

As a result, it is difficult to obtainsufficient fluorescence from thedeeper layers of choroidalvessels.

In blood, about 20% to 40% of injectedsodium fluorescein remains unbound to


6/7

serum albumin. This unbound fluoresceinleaks rapidly from the highly fenestratedchoriocapillaries into the choroidalanatomy. Because of this, details of thechoroidal vascular pattern are obscured.For these two reasons, fluoresceinangiography cannot provide usefulinformation on choroidal circulation.

In ICG angiography, the maximumabsorption and peak fluorescence ofIndocyanine dye is in the spectrum at805nm and 835nm respectively. Thisnear infrared light can penetrate theretinal pigment epithelium much moreeffectively than visible light, allowinguninterrupted examination of the choroidalvascular network. In addition, sinceapproximately 98% of ICG dye in blood isbound to serum proteins, it leaks very

slowly from the choroidal capillaries.

Corneal Image Analyzer

Corneal Image Analyzer (CIA) is softwaredeveloped at NRSA for analyzing Endothelialimages of cornea. The corneal image consistsof a set of hexagonal shaped cells of differentsizes. The aim of this software is to computethe statistics of the cornea endothelial such asthe cell density, minimum, maximum & meancell sizes, their standard deviation, co-variance etc. This data is useful for various

studies such as dystrophy and degeneration,intraocular lens implantation, cornealtransplantation, drug toxicity, glaucoma, etc.

The cornea of the patient is scanned byspecial devices and these images can beanalyzed using computers at greater speedand precision. Requirements such asendothelial cell analysis and cell areameasurement are currently done manually.

The images of endothelium can be filmed witha 35mm still camera attached to the specular

microscope.

The images can be fed to the computers asinput by scanning the films/prints or bycapturing the images directly from thepatient's eye through high resolution and highsensitive CCD camera whose video output isconnected to a Frame grabber board on a PC-

AT. The analysis is done on a high resolutionmonitor. The quantitative measurements can

be done on computers using imageprocessing techniques. Prints of theseimages are used for diagnosis.

The software gives a printout of the celldensity, minimum, maximum and mean cellsizes, standard deviation and coefficient ofvariance of the cell sizes, histogram of cellsizes and distribution of cell areas. Severalreports of a patient at various dates arecombined to give a combined report.

Fundus Image Analyzer

Many diseases of the human visual systemand of the whole body can have a dramaticimpact on the 3-dimensional geometry of theocular fundus. Glaucoma is probably the mostimportant disease in this category. Itincreases the cupping of the optic nerve head

at an early stage of the disease, in manycases before a reliable diagnosis can bemade and visual field losses occur. The earlydiagnosis of glaucoma is a major issue ingeneral public health care. Quantitativeassessment of fundus geometry is expectedto be helpful for this purpose.

The ocular fundus consists of several layersof highly transparent tissue, each havingindividual physical properties, reflectivity,absorption and scatter. 2-dimensionalgeometry normally specifies sub-structures

such as the vessel pattern or the area ofpallor delineated by contrast or colorvariations. It is less important how deep theyare located within the fundus. Depth iscommonly associated with the topography ofthe interior limiting surface of compact retinaand optic disc tissue.

A system for ocular fundus analysis consistsof two parts, the image acquisition and theanalysis software. The image is normallyobtained using a telecentric fundus camera(Ex. Ziess-30 degree fundus camera). The

image is captured onto a slide film. The film isscanned using a slide film scanner andtranscribed to a Personal Computer.

Alternatively the image can be directlyacquired from the camera by interfacing it tothe Personal Computer using a frame grabbercard. Fundus Image Analyzer (FIA) is theoptic disc analysis software developed atNRSA. The present version operates on 2-Dimages only and does not support


7/7

depth/volume calculation. Figure 3 shows theFundus Image Analysis System.

Fig 3. Image Acquisition and processing ofendothelial cells of cornea

The following is the list of parameterscalculated using this software:

Disc diameter: Specifies the horizontal/verticaldiameter of the selected disc edge.Cup diameter:Specifies the horizontal/verticaldiameter of the selected cup edge.Disc Area: Specifies the area within theselected disc edge.

Cup Area: Specifies the area within theselected cup edgeRim Area: Specifies the area between theselected disc edge and the calculated cupedge.Cup-to-Disc Ratio: Specifies the ratio of thecup size to disc size.

Quantitative measurements of fundus imageare shown in Figure 4.

This package is useful for quantitativemeasurements of the optic nerve. If the depthinformation is derived from the stereo pair,then volume calculations & profile generationscan also be done.

Automatic Classi ficat ion of CancerousCells from a Digitized Picture of aMicroscope Section

Mathematical morphology is used to removethe background noise from the image and thenuclei of the cells are segmented. Thesenuclei are analysed for the shape and size.The texture of the nuclei is evaluated by usingKOHONEN NEURAL NETWORK; automaticclassification of the image is done.Structure-Sensitive Adaptive ContrastEnhancement Methods

For medical images the overall goal ofdisplay is the detection, localization andqualitative characterization of anatomicalobjects represented by the intensityvariations in the recorded data.

Global histogram equalization is justified

by the argument that for noise-freeimages, it maximally transmits informationas to scene intensity values.

Adoptive Histogram Equalization hasdemonstrated its effectiveness in thedisplay of images from a wide range ofimaging modalities, including CT, MRIand Radiotherapy portal films.

Gordon's techniqueis another method forcontrast-based enhancement for thedetection of edges within the contextual

region.

References1. Proceedings of 2

nd Singapore

International Conference on ImageProcessing 1992.

2. Fundamentals Of Digital Image Processing- Anil K. Jain, Prentice-Hall, 1989.

3. Digital Image Processing - Chellappa, 2nd

Edition, IEEE Computer Society Press, 19924. Digital Image Processing - R.C. Gonzalez

Woods, Addison Wesley, 1992.5. Digital Image Wrapping, George Wolberg,

IEEE Computer Society Press 1999.

6. Biomedical Instrumentation M. Arumugam,Anuradha Agencies, PublishersKumbakonam 1992.

7. Proceedings of the 14thworld conference on

NDT (14thWCNDT) 1996.

START

AcquireStereoFundusIma e

GetpatientInfor-mation

Mark Cup/Discdimensions &

Area

Actual AreaCalculation

Report

MEDICAL_IMAGE_PROCESSING_25sep06.pdf

Documents

Transcript of MEDICAL_IMAGE_PROCESSING_25sep06.pdf