Screening mammography is commonly the only imaging exam allowing early-stage detection of breast cancer. The early detection is, in fact, associated with a decreased breast cancer mortality rate amongst women. However, false positive recall is one of the main limitations of screening practices and it is often associated with unnecessary workups and biopsies. To tackle this issue and improve the medical image classification performance in order to carry out a screening/diagnosis task, we propose to use a multi-view deep convolutional neural network - the proposed network can extract discriminative features from Cranial Caudal (CC) and Medio-Lateral Oblique (MLO) views for each breast of a patient (a set of four images). We experiment it on an augmented-data based subset selected from the open Digital Database for Screening Mammography (DDSM) using 5400 images. We show how the proposed method can lead to a better performance than the state-of-the-art ones, especially in terms of prediction accuracy and false positive rate reduction. In fact, The results show statistically significant reduction in false findings without increasing false negative cases. Our method achieves a specificity rate of 98% and an accuracy rate of 98.88%. Index Terms–Mammography, Breast cancer diagnosis, False positive findings, Deep learning, Multi-view deep convolutional neural network.
展开▼
