Classification of ductal carcinoma in situ by image analysis of calcifications from digital mammograms

Abstract

There is evidence to suggest that non-comedo cases of ductal carcinoma in situ (DCIS) may have a lower risk of progressing to invasive disease than comedo DCIS and may be managed with less radical treatment. Most cases of DCIS present as calcifications on the mammogram and, due to differences in the tumour architecture, comedo calcifications often have a characteristic linear or branching appearance. In this study, computer methods were developed to identify the comedo cases using the imaging features of individual calcifications and of calcification clusters. Classifier performance was measured using the area under the receiver operating characteristic (ROC) curve and was optimized by systematic testing of many different combinations of these features. On a test-set of 42 cases the computer achieved a ROC curve area of 0.91 using six combined cluster features. These results are encouraging given the observed overlap in appearance between the comedo and non-comedo cases.