Segmentation of Mammary Lesions in Ultrasound Images Applying Mask R-CNN

Abstract

Breast cancer is the most frequent malignant tumor in women and one of the most common in the world. One of the most important issues in this condition is early detection. Computer-aided diagnostic (CAD) systems are objects of research, aiming to provide a second opinion to the health professional. A fundamental aspect within the CAD system is the segmentation of the lesion, allowing an adequate extraction of the lesion characteristics. The use of a computerized segmentation method helps eliminate human variability and, consequently, improve the performance of the lesion classifier. Convolutional Neural Networks (CNNs) are being used in segmentation problems, such as various types of medical imaging, people and road signs detection, for example. So inspired by these promissing results, the present work has as main objective to analyze and implement the Mask R-CNN as a tool of segmentation of mammary lesions in images obtained by ultrasound, to propose an efficient method of segmentation, aiding in the classification process of the CAD systems.