Detecting Masses in Mammograms using Convolutional Neural Networks and Transfer Learning
Mor Yemini, Yaniv Zigel, Dror Lederman · 2018
This paper addresses the problem of mass detection in mammograms. It has long ago been shown that computer-aided diagnosis (CAD) schemes have the potential of improving breast cancer diagnosis performance. We propose a CAD scheme based on convolutional neural networks, using transfer representation learning and the Google Inception-V3 architecture. Artificially generated mammograms and data augmentation techniques are used to expand and balance the available database at train time. The performance of the proposed scheme is evaluated based on the receiver operating characteristics (ROC) curve. Areas under the ROC curve of 0.78 and 0.86 were obtained using artificially-generated mammograms and augmentation, respectively.