1. Introduction
The rapid development of Deep Neural Networks (DNNs) has boosted their wide applications in many domains, including medical image processing. For example, they can perform ventricle segmentation on Magnetic Resonance Imaging (MRI) images [1] or 3D whole-heart segmentation on Computed Tomography (CT) images [2]. High performance on these tasks would benefit clinical diagnoses while simultaneously reducing the workload of medical workers.