Utilizing Deep Learning Technology to Segment Pectoral Muscle in Mediolateral Oblique View Mammograms

Abstract

Breast cancer is the most commonly diagnosed cancer and the leading cause of cancer death among women all over the world. Accurate pectoral muscle segmentation in mediolateral oblique(MLO) view mammograms is an important pre-processing step for subsequent analysis in computer aided diagnosis (CAD) systems. In this paper, a deep learning based approach for pectoral muscle segmentation is proposed. The method includes the following steps. First, the artifacts in the mammogram is eliminated by applying connected component labeling. Second, a pre-trained U-Net model is utilized to get the initial pectoral muscle segmentation result. Finally, an improved scanline filling method is employed to solve the under-segmentation problem existing in initial segmentation result. Preliminary investigation on a test dataset demonstrated that the proposed method yielded average values of 93.5 ± 3.0% and 88.0 ± 5.0% for the Dice and Jaccard similarity metrics, respectively.