Computer-Aided Diagnosis of Breast Cancer on Mammograms

Abstract

Computer-aided diagnosis (CAD) is a diagnosis made by a physician who takes into account the computer output of quantitative analysis of mammograms. CAD schemes in mammography have been developed to detect lesions such as clustered microcalcifications and masses, and also to distinguish between benign and malignant lesions. Computerized schemes are composed of three major steps which are image processing, quantitation of image features, and data classification. The current performance level of detecting clustered microcalcifications by computer is approximately 85% at a false positive rate of 0.5 per mammogram, whereas the detection accuracy of masses is approximately 90% at a false positive rate of 2 per mammogram. Observer performance studies indicated that computer output can improve the performance of radiologists in detecting clustered microcalcifications by increasing the dtection accuracy to 90% from 80% at a specificity of 90%. The automated classification of clustered microcalcifications is based on quantitative analysis of image features of individual microcalcifications and cluster, followed by artificial neural networks (ANNs) for data classification. With our database, the computer scheme correctly identified 82% of patients with benign lesions, all of whom had biopsies (ie, the radiologist thought the microcalcifications were suspicious for malignancy), and 100% of patients with malignant lesions. On the same set of images, the average of five radiologists was only 27% correct in classifying lesions as benign at 100% sensitivity. The automated classification of masses is made by the quantitation of image features of masses together with a rule-based and ANNs method for data classification. The computer scheme achieved, at 100% sensitivity, a positive predictive value of 83%, which was 12% higher than that of the experienced mammographer and 21% higher than that of the average of less experienced mammographers. The first prototype intelligent workstation for mammography was developed at the University of Chicago, and applied to approximately 12000 screening cases for the detection of early breast cancers. Promising initial results were obtained with the workstation.