Bayesian Hierarchical Microstructure Modelling in the Brain

Project info

Title: Sharing information across voxels with Bayesian hierarchical modelling to improve brain microstructure mapping

Screenshot 2021-01-22 at 17 53 12

Project lead: Paddy Slator (email: p.slator@ucl.ac.uk, mattermost: paddyslator)

Project collaborators: Chris Parker Lizzie Powell Matteo Battocchio

Registered Brainhack Global 2020 Event: Brainhack Atlantis The Atlantic Ocean - Micro2Macro

Project Description:

AIM: Implement a hierarchical Bayesian fitting procedure for a range of brain microstructural models.

Typically microstructural models are fitted “voxel-by-voxel” to diffusion MRI (dMRI) data, with the implicit assumption that each image voxel is an independent measurement. Some recent techniques break this assumption, exploiting data redundancy to improve model fits and subsequent mappings.

One such method is the Bayesian hierarchical intravoxel incoherent motion model (IVIM) introduced by Orton et al. (https://doi.org/10.1002/mrm.24649). Here the posterior distribution encodes voxelwise microstructural parameter estimates, and the prior distribution encodes parameter means and covariance across a larger ROI. By applying Bayes’ Rule and inferring the model with a Markov chain Monte Carlo (MCMC) algorithm, they improve IVIM parameter mappings of liver dMRI compared to standard methods.

This project will adapt this approach to brain microstructure modelling.

Data to use:

We will test the method on a (to be chosen later) Human connectome project (HCP) dMRI scan (https://www.humanconnectome.org/study/hcp-young-adult/data-releases).

Link to project repository/sources:

https://github.com/PaddySlator/dmipy This is a fork of the dmipy (Diffusion Microstructure Imaging in Python) repository. The project will utilise and adapt this code, with the ultimate goal of integrating the developed tools with dmipy.

Goals for Brainhack Global 2020:

Deliverable 1: Implement MCMC algorithm for Bayesian hierarchical brain microstructure modelling Deliverable 2: Test MCMC algorithm on an HCP dMRI scan

Good first issues:

  1. Discuss and choose which brain microstructure models to focus on (e.g. ball-stick, NODDI, SMT,…)
  2. Simulate simple test datasets for the microstructural models (dmipy)
  3. Implement MCMC algorithm for inference of the Bayesian hierarchical microstructure model on simulations (adapt dmipy)
  4. Download a suitable preprocessed HCP dMRI scan.
  5. Segment HCP dMRI scan into WM/GM/CSF ROIs (SPM or FSL)
  6. Apply MCMC algorithm to HCP dMRI scan (adapt dmipy)
  7. Baseline least squares fitting for microstructure models (dmipy)
  8. Baseline MCMC fitting for microstructure models (MDT or cuDIMOT)
  9. Write the MCMC algorithm as pseudocode

Skills:

Tools/Software/Methods to Use:

Not required (don’t worry about installing beforehand) but could be useful if you already have them installed:

Communication channels:

https://mattermost.brainhack.org/brainhack/channels/micro2macro-bayesian-fitting

Project labels

Project Submission

Submission checklist

Once the issue is submitted, please check items in this list as you add under ‘Additional project info’

Optionally, you can also include information about:

We would like to think about how you will credit and onboard new members to your project. If you’d like to share your thoughts with future project participants, you can include information about:


Date
Jan 1, 0001 12:00 AM