Mass and Calcification Detection from Digital Mammograms Using UNets
Anoop Sathyan, Dino Martis, Kelly Cohen
Abstract
Today Computer Aided Detection (CAD) systems are used to help radiologists in identifying and analyzing mammogram abnormalities. More recently, Convolutional Neural Networks have been proven effective in identifying abnormalities such as masse and calcification. Certain types of masse and clustered calcifications in mammograms are considered malignant growth. In this paper, we present an approach to detect the presence of these abnormalities, and segment both masses and calcifications in mammogram images. We use a fully convolutional architecture (UNet) trained to segment mass and calcification. The UNet for mass segmentation is trained on the CBIS-DDSM digitized image dataset. The InBreast dataset, which provides full field digital mammograms, has been used to train calcification segmentation.