Skip to content

Latest commit

 

History

History
46 lines (25 loc) · 2.07 KB

README.md

File metadata and controls

46 lines (25 loc) · 2.07 KB
Raw

Back to Projects List

Whole Heart Segmentation for Congenital Heart Disease

Key Investigators

  • Danielle Pace (MIT)

Project Description

We are working on segmenting the cardiac chambers and great vessels from 3D MRI, for patients with congenital heart disease. The aim is to enhance surgical planning via rendered or 3D-printed heart models.

We have previously developed a recurrent neural network (RNN) model that evolves a segmentation of each structure over time, and shown that it is more generalizable to severe CHD defects than conventional models that segment the image in one step.

Objective

  1. Objective A. Use data augmentation for inference.
  2. Objective B. Finish data augmentation (during training) that mimics stents and 'distractor' vessels.

Approach and Plan

  1. Implement and test various data augmentation schemes during inference for a growing segmentation.
  2. Debug NaN bug for data augmentation that mimics stents and 'distractor' vessels.

Progress and Next Steps

  1. Made good progress towards objective A: fixed bugs and implemented data augmentation during inference for a growing segmentation.

Illustrations

RNN-based segmentation of cardiac structures

Background and References

Iterative segmentation from limited training data: Applications to congenital heart disease

D.F. Pace, A.V. Dalca, T. Brosch, T. Geva, A.J. Powell, J. Weese, M.H. Moghari, P. Golland MICCAI Workshop on Deep Learning in Medical Image Analysis, 2018