An End-to-end Approach to Semantic Segmentation with 3D CNN and Posterior-CRF in Medical Images

  • 2018-11-08 17:00:05
  • Shuai Chen, Marleen de Bruijne
  • 10

Abstract

Fully-connected Conditional Random Field (CRF) is often used aspost-processing to refine voxel classification results by encouraging spatialcoherence. In this paper, we propose a new end-to-end training method calledPosterior-CRF. In contrast with previous approaches which use the originalimage intensity in the CRF, our approach applies 3D, fully connected CRF to theposterior probabilities from a CNN and optimizes both CNN and CRF together. Theexperiments on white matter hyperintensities segmentation demonstrate that ourmethod outperforms CNN, post-processing CRF and different end-to-end trainingCRF approaches.

 

Introduction (beta)

None

 

Conclusion (beta)

None