Next Statistics Seminar, University of Edinburgh
(note a different day)
Date: 1 December 2008 (Monday) at 3pm
Location: JCMB 5215
Alexey Koloydenko (Royal Holloway, London)
Statistical Analysis of Diffusion Weighted MRI data
Abstract
Diffusion weighted MRI (DMRI) has relatively recently become available
for probing microstructure and studying structural integrity and
organization of various types of matter. In particular, white matter
tractography based on DMRI attempts to integrate local charts of white
matter fiber bundles in order to assess brain connectivity. At the core
of these efforts is the ability to estimate reliably dominant fiber
directions from a set of DMRI probes registered to a particular location
(voxel) in the brain.
In this talk, we will consider two issues associated with the popular
diffusion tensor approach to DMRI data analysis. First, we will discuss
how to do basic statistics with diffusion tensors taking into account
that they form a non-euclidian manifold. This theme turns out to be more
general since diffusion tensors are essentially semidefinite symmetric
matrices arising in many other fields. Second, we will discuss
suitability of a hidden Markov random field model (HMRFM) for DMRI data.
Preliminary results will be shown suggesting that this kind of models
are quite different from the commonly used model with an additive mean
structure.
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|