Classification of microcalcifications in radiographs of pathologic specimens for the diagnosis of breast cancer

Abstract

Early detection of breast cancer depends on accurate classification of microcalcifications. We have developed a computer vision system that has the potential to classify microcalcifications objectively and consistently to aid radiologists in diagnosing breast cancer. A convolution neural network (CNN) was used to classify benign and malignant microcalcifications in radiographs of pathologic specimens. Digital images were acquired by digitizing radiographs at a high resolution of 21 microns x 21 microns. Eighty regions of interest selected from digitized radiographs of pathologic specimens were used for training and testing of the neural network system. The CNN achieved an Az value (area under the receiver operating characteristic curve) of 0.90 in classifying clusters of microcalcifications associated with benign and malignant processes. Classification of microcalcifications in pathologic specimens for diagnosis of breast cancer was achieved at a high level in our computer vision system, which consists of high-resolution digitization of mammograms and a CNN.