Computer-Aided Detection in Full-Field Digital Mammography: Detection in Dependence of the BI-RADS Categories

Abstract

The object of this study was to determine the performance of a computer-aided detection system in full-field digital mammography (Senographe 2000D, General Electric, Buc, France) in detecting carcinomas in breasts in dependence of the initial Breast Imaging Reporting and Data System (BI-RADS) categories. A total of 226 mediolateral oblique (MLO) and 186 craniocaudal (CC) view mammograms of histologically proven cancers were retrospectively evaluated with a primary digital computer-aided detection system (Image Checker V2.3; R2 Technology, Los Altos, CA). According to BI-RADS of the American College of Radiology (ACR), the lesions were classified in MLO view as BI-RADS 1 in 2 cases, BI-RADS 2 in 11 cases, BI-RADS 3 in 37 cases, BI-RADS 4 in 56 cases, and BI-RADS 5 in 120 cases, and in CC view as BI-RADS 1 in 2 cases, BI-RADS 2 in 8 cases, BI-RADS 3 in 26 cases, BI-RADS 4 in 46 cases, and BI-RADS 5 in 104 cases. The computer-aided detection system shows markers also in mammograms classified as BI-RADS categories 1-3 by the radiologist. Furthermore, BI-RADS categories 4 and 5 in most cases demonstrate masses in mammography. With increasing BI-RADS category, the computer-aided detection system shows decreasing numbers of overlooked carcinomas. In MLO view, no markers were found in 100% (2/2), 81.8% (9/11), 59.5% (22/37), 46.4% (26/56), and 15% (18/120) for BI-RADS categories 1-5, respectively. False-positive markers, however, were seen in 0.5 per image (205/412). In conclusion, the high rate of false-positive markers shows that the specificity of the computer-aided detection system is limited and that improvements are necessary.