Application of neural networks to computer-aided pathology detection in mammography
Richard G. Stafford, Jacob Beutel, Daniel J. Mickewich, S. L. Albers · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1993
Digitized mammograms are used as input to a system of feedforward neural networks for determining the presence of clusters of microcalcifications. Rather than presenting the locations of detected suspected pathologies on the digital image itself, the intent is to produce a transparent overlay marking these possible pathologies for later diagnosis by the radiologist. Current results indicate that this method will in general be more efficient than human observers in locating low contrast objects and clusters of small microcalcifications (down to 50 micron diameter).