New version page

Computer-aided Diagnosis

Upgrade to remove ads

This preview shows page 1-2-3-4-5 out of 14 pages.

Save
View Full Document
Premium Document
Do you want full access? Go Premium and unlock all 14 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 14 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 14 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 14 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 14 pages.
Access to all documents
Download any document
Ad free experience

Upgrade to remove ads
Unformatted text preview:

205Computer-aided Diagnosis inDiagnostic Mammographyand MultimodalityBreast Imaging1Maryellen L. Giger, PhDRSNA Categorical Course in Diagnostic Radiology Physics: Advances in Breast Imaging—Physics, Technology,and Clinical Applications 2004; pp 205–217.1From the Department of Radiology, University of Chicago, MC 2026, 5841 S Maryland Ave, Chicago, IL 60637 (e-mail:[email protected]).Supported in part by U.S. Public Health Service grants CA89452 and T32 CA09649 and by U.S. Army Medical Researchand Materiel Command grant DAMD 97-2445.M.L.G. is a shareholder in R2 Technology, Inc, Sunnyvale, Calif. It is the policy of the University of Chicago that investigatorsdisclose publicly actual or potential significant financial interests that may appear to be affected by the research activities.Useful interpretation in mammography depends on the quality of the mammographicimages and the ability of the radiologists who interpret them. Improvements in radio-graphic technique, as well as mandatory accreditation programs, have made the earlysigns of breast cancer more apparent on mammograms. However, radiologists stillsometimes miss cancer on a mammogram. One approach to improving performance isreplicated interpretations, in which more than one observer reviews the images. For ex-ample, investigators have shown that detection of early breast cancer can be increasedwith double reading by two radiologists (1). In addition, results of studies have shownthat interpretation performance varies greatly among radiologists (2–4).An alternate approach is to use a computer as the second reader. Use of output froma computerized analysis of an image by radiologists may help them in the tasks of de-tection or diagnosis and potentially improve the overall interpretation of breast imagesand the subsequent patient care. Many factors motivate the attempts to aid or auto-mate radiologic diagnosis. Inadequacies in interpretation performance may be due tothe presence of image noise or normal anatomic structures, as well as to known limita-tions in the human search and perception process. Ultimately, computer-aided diagno-sis (CAD) may become an integrated tool in the diagnostic work-up of suspect breastlesions by using multimodality images.This chapter reviews various CAD methods in breast imaging (mammography, ultra-sound [US], and magnetic resonance [MR] imaging), which are focused on the charac-terization of lesions and the estimation of the probability of malignancy for use in thediagnostic work-up of suspect lesions. CAD systems in diagnostic work-up usually in-volve having the computer extract the margin of the lesion from the surrounding pa-renchyma, extract characteristics (features) of the lesions, merge these computer-ex-tracted features into an estimate of the probability of malignancy, and, as an option,automatically retrieve similar lesions from an online reference library. The aim of CADin diagnostic work-up is to increase classification sensitivity and specificity, as well asto reduce intra- and interobserver variability. Various reviews have been written aboutCAD in breast imaging (5–11).Giger206Computer-aided detection has already been incor-porated into clinical screening mammography, and itsstatus is reviewed elsewhere in this RSNA syllabus (seechapter by Chan et al). There are currently computer-aided detection systems approved by the Food andDrug Administration, with many insurance carriers,including Medicare, providing coverage for such tech-nology. Development of promising computer-aideddiagnosis prototypes is also underway, and aspects ofthese future systems are reviewed in this chapter.The general techniques employed in the computeranalysis of images can be broadly categorized as com-puter vision and artificial intelligence (12,13). Com-puter vision involves having a computer extract from adigital image features that may or may not be other-wise perceived by a human reader. The developmentof computer vision schemes requires a priori informa-tion about the medical image (eg, the mammogram)and knowledge of various computer processing tech-niques and decision analysis methods. The requireda priori knowledge includes the physical imagingproperties of the digital image acquisition system andmorphologic information concerning the abnormality(eg, mass lesion or cluster of microcalcifications),along with its associated anatomic background. Thatis, a sufficient database is needed to cover the entirerange of abnormal and normal findings. Computer vi-sion techniques include, in general, image processing,image segmentation, and feature extraction (12,13).Computer vision algorithms can be initially used(a) to isolate, or segment, the breast from the remain-der of the image and so limit the computer search re-gion or (b) to enhance the peripheral breast border re-gion to compensate for reduced breast thickness at theedge. For example, investigators have described meth-ods that use computer-defined unexposed and direct-exposure image regions to generate a border aroundthe breast region (14,15). Next, segmentation tech-niques can be employed to separate the image into re-gions with similar attributes (eg, regions exhibitinghigh contrast that might reflect the presence of calcifi-cations) or to isolate a lesion from its surrounding pa-renchymal background, as demonstrated in Figure 1(16). Once segmented, each region is then furtheranalyzed with feature extraction and feature analysistechniques, which yield mathematical descriptors ofthe radiographic features.Radiologists seem to extract and interpret simulta-neously many radiographic image features correspond-ing to signs of malignancy. For example, a high degreeof spiculation exhibited by a mass is a strong sign ofmalignancy, and many computerized methods havebeen developed to quantitate spiculation (17–21).Thus, computer vision methods involve determiningthe mathematical descriptors of image features, alongwith the selection of which individual computer-ex-tracted features are clinically important (7,22).Much as the radiologist weighs different aspects ofa mammographic finding, artificial intelligence tech-niques can be used to merge and/or select image fea-tures obtained with computer vision into a diagnosticdecision (22–32). Various classifiers have been ex-plored as means to merge the computer-extracted fea-tures, including rule-based methods, discriminantFigure 1. (a) Segmentation contours resulting from region growing by using


Download Computer-aided Diagnosis
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Computer-aided Diagnosis and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Computer-aided Diagnosis 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?