Breast cancer screening: potential role of computer-aided detection (CAD)

Abstract

Four separate studies conducted as blinded retrospective radiologist reviews of screening mammograms where cancer subsequently developed found lesions requiring work-up or biopsy in 25% to 41% of cases. Three non-blinded retrospective reviews by radiologists identified 57% to 75% of cancers. Among breast cancers detected by radiologists on blinded retrospective review in one study, computer aided detection (CAD) would have found 77%. These studies suggest that improved cancer detection rates could result if radiologists who interpret screening mammography were aided by computers. One prospective study initiated on that premise found that CAD increased the screening detection rate by 19.5%. At the current state-of-the-art, CAD performs best for cancers presenting as calcifications, less well for cancers appearing as spiculated masses, and least well for cancers appearing as circumscribed masses. CAD does not seem to increase false positive call back rates for additional imaging work-up or to increase false positive biopsy rates.