Evaluation of computer-assisted diagnosis system for carcinoma of the breast.

Purpose: Evaluation of a computer-assisted diagnosis (CAD) system (R2 linage Checker 1.2). Comparison of the results of three readers with and without knowledge of the computer results. Materials and methods: The mammograms of 52 patients, bilaterally in two planes each, with histologically proven carcinoma of the breast were included in the study. They were first scanned by the CAD machine and subsequently read by three readers with different degrees of experience in two sessions with and without knowledge of the computer results. Results: Of the 91 views of carcinomas, the readers detected 96 %, 89 %, and 85 %, respectively. With CAD, the values rose to 97 %, 93 %, and 96 %. The increase of the first observer was not significant. As for the 49 areas of malignant microcalcifications, the first reader showed a significant decrease of sensitivity, the other two readers showed no significant change. The sensitivity of CAD was 74 % for masses and 86 % for microcalcifications at a rate of 1.8 false positive markers per image. All but two tumors were correctly marked in at least one plane. Conclusions: Use of the CAD machine led to a significant increase of sensitivity in the detection of malignant masses by two of three observers. In the case of malignant microcalcifications, and for the most experienced observer, CAD did not improve the results. The most important problem is the high rate of false positive markers.