Using Convolutional Neural Networks with Direct Acyclic Graph Architecture in Segmentation of Breast Lesions in US Images

Abstract

Outlining lesion contours in Breast Ultrasound (BUS) images is an important step in breast cancer diagnosis. Malignant lesions infiltrate the surrounding tissue, generating irregular contours, with spiculations and angulated margins, while benign lesions produce contours with a sharp outline and elliptical shape. In breast imaging, the majority of the existing publications focus on using Convolutional Neural Networks (CNNs) for segmentation and classification of lesions in mammographic images. In this study we propose CNNs with direct acyclic graph (DAG) architecture for breast lesion segmentation in US images. We also compare the performance of these two proposed architectures with a series architecture. The best results were obtained with the DAG architecture. The following mean values were obtained for the metrics used to evaluate the segmented contours: Global accuracy = 9593%; IOU = 87.92%; BF score = 68.77%; and Dice coefficient = 89.11%.