Horsch K et al. (2006) Classification of breast lesions with multimodality computer-aided diagnosis: observer study results on an independent clinical data set. Radiology 240: 357–368

A recent study has shown that the use of a computer-aided multimodality intelligent workstation improves interpretation of mammograms and breast sonograms, and the differentiation between benign and malignant lesions.

Five breast radiologists and five breast imaging fellows assessed images from a total of 97 biopsy-confirmed lesions, and gave confidence ratings (rating their confidence that the lesion was malignant on a continuous scale of 1–100%) and patient management decisions (follow-up or biopsy) without and with computer aid. Analysis showed that the average performance of the 10 observers was improved with use of the computer aid. There was a significant increase in area under the receiver operating characteristic (ROC) curve (Az) value (0.87–0.92; P <0.001), partial Az value (0.47–0.68; P <0.001), and sensitivity (0.88–0.93; P = 0.005), although the difference in specificity with and without computer aid did not reach significance (0.66–0.69; P = 0.20). Overall, more observers changed their treatment decision correctly (i.e. from biopsy to follow-up for benign lesions, or follow-up to biopsy for malignant lesions) following computer-aided analysis than changed it incorrectly. Only one malignant lesion was incorrectly diagnosed as benign by the computer program, while four lesions were incorrectly diagnosed by at least one observer without computer aid. The authors conclude that further validation of this promising system in the clinical situation is warranted.