Diagnostic performance of a computer-aided image analysis system for breast ultrasound

We developed a sophisticated computer-aided diagnostic system optimized to the specific application of standardizing interpretation of diagnostic breast ultrasound. The implemented Computer-Aided Diagnostic tool was evaluated on a database of 332 cases with findings known ("truth") via biopsy or two-year benign follow up. These same cases were interpreted by a group of four experienced radiologists. Performance of the developed CAD was measured as Sensitivity, Specificity, PPV, NPV and ROC Area (A(Z)). For the CAIS A(Z)=0.96, which exceeded the performance of four expert radiologists (average A(Z)=0.86). Az for the radiologists was not significantly different (0.86, 0.85, 0.87, 0.86 0.03). Statistical Power for the study was 93%. The excellent performance of CAIS appears to justify further evaluation in a clinical setting to determine its impact on decision making.