Computer-Aided Diagnosis of Cross-Institutional Mammograms Using Support Vector Machines with Feature Elimination

In the analysis of digital or digitized mammographic images, a requirement is to learn to separate benign calcifications from malignant ones. Such an activity could form part of a computer-aided diagnosis (CAD) tool. We present a CAD study of calcification lesions to demonstrate that CAD of same-institutional mammograms provides significantly higher accuracy compared to that of cross-institutional mammograms. Moreover, using only a subset of the widely used six BI-RADS features together with patient age and subtlety value describing each calcification lesion is shown to increase the accuracy of CAD.