COMPUTERIZED DETECTION OF CLUSTERED MICROCALCIFICATIONS - EVALUATION OF PERFORMANCE ON MAMMOGRAMS FROM MULTIPLE CENTERS

To investigate the performance of a computerized method for the automa ted detection of clustered microcalcifications in digitized mammograms from a variety of screening centers, the authors invited 118 radiolog ists to bring up to five mammograms to their scientific exhibit at the 1993 meeting of the Radiological Society of North America (RSNA). For ty-three mammograms from 14 sites were brought to the exhibit, where t hey were digitized and analyzed. Results of the analysis on the RSNA c ases were compared with those obtained on a standard database of 39 ma mmograms collected from two centers. The performance of the detection algorithm on the RSNA images was lower than that achieved on the stand ard database, This lower performance was due in part to the higher fra ction of very subtle clustered microcalcifications in the RSNA cases, as well as the apparent dependence of the algorithm on image character istics (eg, contrast and noise), which varied from center to center, T he authors conclude that the algorithm is robust and accurate enough t o undergo clinical testing, When it is implemented clinically, the com puterized scheme must be customized to the image characteristics at ea ch specific screening center to obtain optimal performance.