Analysis of mammography: a blind interpretation of BCDDP radiographs

An ROC curve analysis of mammography is presented. This study is based on a blind interpretation of radiographs of patients who underwent screening during the Oklahoma Breast Cancer Detection Demonstration Project (BCDDP). In particular, we evaluated the accuracy of mammography based on 38 incidence cancer cases. Incidence cases are those that develop in a patient after an initial visit and at least one additional annual visit before a biopsy is performed. Using 40 normal subjects, a sensitivity of 72% was attained at a 28% false positive rate. We also evaluated the contribution of viewing radiographs of both breasts (right and left) simultaneously as compared with viewing radiographs of the individual breasts singly. In cases on the threshold of detectability, the breast that did not result in the cancer diagnosis was occasionally identified as the most suspicious for cancer. When mammograms of the two breasts were compared, the false positive rate was lower. These results can be used to assess various screening strategies and to yield a more realistic accuracy estimate of mammography on a rescreened population than is currently available.