DSMS-FCN: A Deeply Supervised Multi-scale Fully Convolutional Network for Automatic Segmentation of Intervertebral Disc in 3D MR Images

摘要

This paper addresses the challenging problem of segmentation of intervertebral discs (IVDs) in three-dimensional (3D) T2-weighted magnetic resonance (MR) images. We propose a deeply supervised multi-scale fully convolutional network for segmentation of IVDs in 3D MR images. After training, our network can directly map a whole volumetric data to its volume-wise labels. Multi-scale deep supervision is designed to alleviate the potential gradient vanishing problem during training. It is also used together with partial transfer learning to boost the training efficiency when only small set of labeled training data are available. The present method was validated on the MICCAI 2015 IVD segmentation challenge datasets. Our method achieved a mean Dice overlap coefficient of 92.0% and a mean average symmetric surface distance of 0.41 mm. The results achieved by our method are better than those achieved by the state-of-the-art methods.