Edit this page

NA-MIC Project Weeks

Back to Projects List

Brain Mask Prediction Based on MRI Skin Data

Key Investigators

Project Description

We postulate that there is a relationship between the shape of ones head and the shape of ones brain. This project aims test that by developing an AI solution for predicting a brain mask given surface data for a head. The eventual goal and application is to map the predicted brain mask to a scanned patient. This project is part of the TMS module project.


  1. Objective A. Build and test a CNN model
  2. Objective B. Migrate TMS model and implement on Slicer
  3. Objective C. Build and test a geometric CNN model*

Approach and Plan

  1. We have some MRI from the HCP Human Connectome Project
  2. Skin masks and Brain Masks were obtained from these MRIs using HDBET and FieldTrip toolbox
  3. Using these as ground truths, train a CNN model to see the feasibility
  4. Implement TMS model on Slicer as a module
  5. Convert ground truth data into surface meshes
  6. Using the new mesh data, train a geometric CNN model and compare results

Progress and Next Steps

Not a lot of progress was made.

  1. Some issues with the MRI Masks, data misaligned.
    • Has been resolved, will start training next week
  2. Started a TMS Prediction Module, Source below
    • Prediction is working
    • Need to create post-processing script to return niftii


TMS Prediction UI TMS Prediction Output

Background and References