Project Week 25/CNN for multi-plane prostate segmentation

From NAMIC Wiki
Revision as of 11:42, 30 June 2017 by Marianna (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search
Home < Project Week 25 < CNN for multi-plane prostate segmentation


Back to Projects List

Key Investigators

Project Description

Objective Approach and Plan Progress and Next Steps

For prostate lesion assessment, MRI volumes in sagittal, coronal and axial orientation are acquired. The goal is to segment the prostate in these volumes and fuse the segmentations in order to obtain better accuracy, especially in the apex and base regions (compared to segmentations obtained only on axial volumes).

  • Work done:
    • created training data for different orientations
    • Automatic segmentation of each plane with Encoder-Decoder Network
  • Work ahead:
    • Explore, discuss and investigate different approaches for fusion of the segmentations, e.g. simple majority voting, suface reconstruction from point cloud, ...


  • built framework for predicting and fusing the segmentations
  • tested simple majority voting in isotropic voxel volumes
  • Future Work:
    • weighted voting depending on location and plane
    • Surface extraction from point clouds


Illustrations

MRIs of the prostate (axial, sagittal and coronal volume)
Example segmentation (apex, mid-gland and base slices).

Background and References