Development of Methods for Computer-Assisted Interpretations of Digital Mammograms for Early Breast Cancer Detection.

摘要

The goal of the research is to develop a computer-vision module as an aid to radiologists. Specifically, we have: (1) Further developed our computerized schemes for the detection and classification of masses and microcalcifications in digitized mammograms. Using new lesion features and genetic algorithms, we have improved mass detection by reducing the number of false positives per image from 2.6 to 1.5. Shift-invariant artificial neural networks have been used to reduce the number of false-positive detections in microcalcification detection. Computerized classification methods for both masses and clustered microcalcifications have demonstrated performance levels that were higher than that of 'average radiologists' in the task of classifying lesions as malignant or benign. We have also developed a dedicated intelligent workstation for screening mammography, which has been used in our clinical mammography reading area for over a year. We evaluated the dedicated 'clinical' workstation using the first 1149 mammographic screening cases. The computer found 6 of the 7 biopsy-confirmed cancers with false-positives rates of 0.9 per image for clustered microcalcifications and 1.4 per image for mass lesions. The significance of this research is that it the detectability of cancers can be increased by employing a computer to aid the radiologist's diagnosis, then the treatment of patients with cancer can be initiated earlier and thus chance of survival improved.