2014 Project Week:CardiacCongenitalSegmentation

• 

  Projects List

• 

  Double outlet right ventricle

• 

  Example input to Carrera on axial slice

• 

  Carrera output after manual interaction

Key Investigators

Danielle Pace, MIT

Polina Golland, MIT

Project Description

Objective

• Develop a semi-automatic segmentation algorithm for cardiac MR images of patients with congenital heart defects.
• Goal is to build a surface model showing the endocardial and epicardial boundaries, for surgical planning.
• Challenges:
  • Very large inter-subject variability due to heart defects
  • Intensity inhomogeneities within myocardium and blood pool
  • Similar intensity distributions within adjacent tissues (e.g. liver, chest muscle)

Approach, Plan

• Try existing open-source tools for segmentation and registration on the five datasets that we have so far.
• See where these methods fail, to focus our efforts for developing algorithms for segmenting hearts with congenital defects.

Progress

• Focused on blood pool segmentation in one test case. Tried:
  • CARMA tools: isolated connected / connected threshold operators
  • Editor level tracing effect
  • Editor Fast Marching effect
  • Editor Grow Cuts effect
  • Carrera interactive segmentation
  • Robust statistic active contour segmentation
• Best tool = Carrera interactive segmentation
  • Main difficulty = small chamber/vessel walls assigned as blood pool, but these can be fixed within Carrera somewhat easily
• Also tried affine registrations across my 5 subjects, using BRAINSFIT
  • Works very roughly, as expected
• Next steps:
  • Try Carrera on additional datasets for blood pool segmentation
  • Myocardium segmentation is still a challenge
• Thanks to: Josh Cates, Salma Bengali, Yi Gao, Ivan Kolesov for your help and suggestions!