A New Computerized Method for Missed Cancer Detection in Screening Mammography

The development of computer-aided diagnosis (CAD) methods to improve a radiologist's decision-making in reading mammography is facing a bottle-neck problem. One critical reason is that almost all current CAD methods are actually developed to perform as an independent "reader" rather than a complementary assistant "second reader" to the radiologists. This paper proposes a new computer aided detection (CAD) method to improve early detection of breast cancer in screening mammograms by focusing on computerized analysis and detection of cancers missed by radiologists. The strategics taken in this study include (a) multi-mode detection by breast density classification, (b) breast area partition and region based adaptive detection, (c) weighted classification using the distinguishing features identified in missed cancer analysis. The results demonstrated that, due to the effective modification strategics taken in the new system, detection performance was improved significantly for mammograms at both detected and missed stages. With the focus on missed cancer analysis and detection, a bigger improvement was obtained in detecting missed cases even though the general detection performance is still lower than that at detected stage. It is also observed that the cancer could be detected earlier with less false positive signals by using the new CAD strategy.