DeYOLOv3: An Optimal Mass Detector for Advanced Breast Cancer Diagnostics

Abstract

Mass detection is an essential phase in any breast cancer diagnosis procedure. This paper proposes an optimal mass detector based on a novel YOLOv3 framework that can handle both detection and classification of mass in full-field digital mammograms just in one single pass through a deep convolution neural network (DCNN). To tackle the conflict between classification and regression tasks that remained in the original model, we replace the coupled detection head with decoupled one. It can be observed in the result that our proposed model yields better performance than other advanced detection models.