Automatic wrist cartilage segmentation in MR images by a patchbased convolutional neural network

摘要

Introduction: Measuring of knee cartilage volume by segmentation from 3D MR images is a routine procedure. The most promising techniques are based on convolutional neural networks (CNNs) with the Dice similarity coefficient (DSC) achieving 0.881 with respect to the manual labels. At the same time, only a few methods have been developed for wrists2. The automatic segmentation of wrist cartilage could facilitate research of novel imaging biomarkers of wrist osteoarthritis. In this work, we propose a method based on dedicated CNN and investigate its performance in comparison with manual and state-of-the-art CNN-based1 methods.