Computer Aided Detection of Abnormalities in Mammograms

A system that automatically identifies suspicious regions in mammograms could be useful to radiologists by drawing attention to abnormalities that may otherwise have been overlooked. Two detection algorithms are described; one based on the combination of evidence from multiple cue generators and the other based on fuzzy pyramid linking. The latter algorithm proved to be the more effective for locating mammographic abnormalities and was used to generate attention cues, or prompts, for our system. We have performed an experiment in which 100 mammograms were presented to eight radiologists in a manner similar to routine screening practice. These films were presented both with and without prompts and our results demonstrate that the detection performance of the radiologists was significantly improved by prompting.