Spatial Enhancement by Dehazing for Detection of Microcalcifications with Convolutional Nets

Microcalcifications are early indicators of breast cancer that appear on mammograms as small bright regions within the breast tissue. To assist screening radiologists in reading mammograms, supervised learning techniques have been found successful to detect microcalcifications automatically. Among them, Convolutional Neural Networks (CNNs) can automatically learn and extract low-level features that capture contrast and spatial information, and use these features to build robust classifiers. Therefore, spatial enhancement that enhances local contrast based on spatial context is expected to positively influence the learning task of the CNN and, as a result, its classification performance. In this work, we propose a novel spatial enhancement technique for microcalcifications based on the removal of haze, an apparently unrelated phenomenon that causes image degradation due to atmospheric absorption and scattering. We tested the influence of dehazing of digital mammograms on the microcalcification detection performance of two CNNs inspired by the popular AlexNet and VGGnet. Experiments were performed on 1, 066 mammograms acquired with GE Senographe systems. Statistically significantly better microcalcification detection performance was obtained when dehazing was used as preprocessing. Results of dehazing were superior also to those obtained with Contrast Limited Adaptive Histogram Equalization (CLAHE).