Computer-aided detection (CAD) of breast-cancer diagnostic performance of a further improved system

The diagnostic value of the use of a computer-aided detection-system (CAD) in the interpretation of mammograms as an alternative to double -reading by a second radiologist is not yet clear. On the base of 148 histologically proven breast cancer in 144 patients depicted in 519 mammograms we compared the performance of a CAD-system in two different versions. The mammograms were secondarily digitized. Algorithms were applied for detecting suspicious findings i.e. clusters of microcalcifications as well as spiculated masses. These suspicious regions of interest were marked and displayed on monitors. Two investigators compared these markings with the findings on the original mammograms and decided the position of markers to be correct or not in consensus form. The total number of markings on mammograms decreased at -25.7% (1.46 markers per picture, version 1.2, vs. 1.96, version 1.0). Correct identification of cancers happened in 128/144 (88.9%), version 1.2, instead of 107/144 (74.3%), version 1.0. Positive predictive value of marking spiculated masses rose to 0.45 (version 1.2) instead of 0.27 (version 1.0) of marking microcalcifications 0.66 (version 1.2) vs. 0.27 (version 1.0). In conclusion the actual version 1.2 of this CAD-system allows for a reliable identification of breast cancer. In comparison to the former version 1.0 the diagnostic performance increased clearly. For further statistical assessment inclusion of a true negative control group is necessary.