Computer-aided diagnosis: analysis of mammographic parenchymal patterns and classification of masses on digitized mammograms

Identification and enhanced surveillance of women at high risk may lead to earlier detection of breast cancer. A computerized method was developed to identify mammographic parenchymal patterns that are associated with breast cancer risk. In this method, various features were first extracted to characterize mammographic parenchymal patterns. These features were then related to breast cancer risk using two different approaches, one based on BRCAI/BRCA2-mutation carriers and one based on clinical risk assessment models. Useful features were identified using stepwise linear discriminant analysis or stepwise linear regression analysis. Results show that increased mammographic density and coarse mammographic patterns correlate with increased breast cancer risk. To potentially help radiologists improve their diagnostic accuracy in classifying benign and malignant breast masses on mammograms, thus reducing the number of biopsies of benign lesions, a computerized classification scheme was developed. The scheme includes three components: (1) automated segmentation of a mass from its parenchymal background, (2) automated feature extraction, and (3) automated classification yielding an estimated likelihood of malignancy. Results show that the scheme can perform similarly to an expert radiologist and significantly better than average general radiologists in distinguishing between benign and malignant masses