From spatial models of fMRI datasets to templates of brain function
Speaker: Bertrand Thirion , Neurospin/INRIA, France
Date: May 8 2009
Time: 2:00PM to 3:00PM
Location: 32-D507
Host: Polina Golland, CSAIL
Contact: Polina Golland, x38005, polina@csail.mit.edu
Relevant URL: In this presentation, we will discuss the problem of inferring which
brain regions are active in a certain functional protocol, given
observations form a limited cohort of functional MRI datasets (10 to
20 typically). After a brief review of the standard analysis
approaches and a discussion of their limitations, we will describe a
novel strategy that consists in treating the inference problem as a
template inference problem where the template can be described as a
set of discrete objects (brain modules). This yields typically a
hierarchical approach, in which individual candidate patterns are
first identified and possibly endowed with a structural or topological
description; then false positives have to be discriminated from true
positives, while explicit correspondences between individual patterns
are inferred: we will successively discuss the main alternatives for
spatial, structural and probabilistic models of these fMRI patterns of
interest. We will particularly emphasize the statistical validation
issues, which are important to neuroscientists. We will finally
describe various experiments that show that the proposed approaches
can be much more sensitive and reliable than standard parametric
mapping procedures, and conclude with ongoing and future related
projects.
See other events that are part of Biomedical Imaging and Analysis 2008/2009
See other events happening in May 2009
