Detecting low-conspicuity mammographic findings - The real added value of CAD

Abstract

This study investigates the effectiveness of CAD for low-conspicuity malignant lesions that are subtle and sometimes missed in conventional analysis. 280 malignantcases were retrospectively reviewed by a non-blinded radiologist, who identified 676 findings. A conspicuity score was assigned to each finding on each view, and 11 findings were of low conspicuity. CAD sensitivity of a prototype CAD algorithm (Siemens), for the high-conspicuity findings was 91.5%. The sensitivity for the 67 cases with low-conspicuity findings in both views (65.7%) was considerably higher than that reported for similar cases in conventional interpretation (40.2%). For the 2688 normal cases, CAD generated 1.24 false marks per case. CAD sensitivity for low-conspicuity findings did not significantly depend on breast density, and was significantly better for non-invasive lesions and for masses in younger women. Thus, CAD should be most beneficial for avoiding oversight of low-conspicuity breast cancers, particularly non-invasive lesions and masses in younger women. © 2012 Springer-Verlag Berlin Heidelberg.