Diffusion MRI analysis


Principal Investigator: Ghassan Hamarneh
Students: Brian G. Booth
  Colin J. Brown
  Shawn Andrews

Diffusion MRI (dMRI) has received considerable interest from the medical image analysis community due to their complicated, high-dimensional nature. At each voxel in a dMRI, we have a function describing the diffusion at that point. These diffusion functions are manifold-valued quantities requiring unique image processing and analysis techniques the ensure the manifold of diffusion functions is properly respected. We work on developing diffusion representations, distance metrics, differential operators, and filtering algorithms to appropriately process dMRI scans. We also work on applying these techniques to dMRI segmentation through the use of appropriate image models and edge detectors.

Diffusion MRI is also unique in another respect: the diffusion functions at each voxel contain orientation-dependent information. This orientation information is directly related to the direction of neuronal pathways in the brain and provides us with the new computational challenge of tractography: tracing out these neuronal pathways from dMRI scans. We work on graph-based approaches and intelligent seeding techniques to perform tractography in efficient and effective ways.

The high dimensionality of dMRI makes visualization difficult and puts greater emphasis on the accuracy of computational methods. We work on improving visualization techniques of dMRI data while generating statistical techniques for examining uncertainty in both dMRI scans and tractography results. We also work on generating techniques for validation of dRMI segmentation algorithms.

Our work in diffusion MRI can be found in these publications. We have also published the following software related to diffusion MRI analysis:

  • PerceptVis: perception-based visualization of manifold-valued medical images.
  • Bilateral Filtering of Diffusion Tensor MRI
  • View3D: MATLAB viewer for 3D scalar, vector, and tensor-valued medical images
  • DeformIt 2.0: Simulate novel images with ground truth segmentations from a single image-segmentation pair, now with support for scalar, vector and tensor-valued images.

Alumni Members:

  • Yonas Weldeselassie
  • Judith Hradsky
  • Krishna Nand