Deep Neural Networks for Surface Segmentation Meet Conditional Random Fields

TL;DR

This paper introduces a novel 3-D CNN combined with CRFs for direct surface segmentation in medical images, providing an end-to-end trainable model that guarantees topology and improves segmentation accuracy.

Contribution

It is the first to integrate 3-D CNNs with CRFs for direct surface segmentation, addressing topology issues without post-processing.

Findings

Achieved promising results on NCI-ISBI 2013 MR prostate dataset.

Demonstrated effectiveness on Medical Segmentation Decathlon Spleen dataset.

First application of 3-D CNN with CRFs for surface segmentation.

Abstract

Automated surface segmentation is important and challenging in many medical image analysis applications. Recent deep learning based methods have been developed for various object segmentation tasks. Most of them are a classification based approach (e.g., U-net), which predicts the probability of being target object or background for each voxel. One problem of those methods is lacking of topology guarantee for segmented objects, and usually post processing is needed to infer the boundary surface of the object. In this paper, a novel model based on 3-D convolutional neural networks (CNNs) and Conditional Random Fields (CRFs) is proposed to tackle the surface segmentation problem with end-to-end training. To the best of our knowledge, this is the first study to apply a 3-D neural network with a CRFs model for direct surface segmentation. Experiments carried out on NCI-ISBI 2013 MR prostate dataset and Medical Segmentation Decathlon Spleen dataset demonstrated promising segmentation results.