October, Wednesday 4th

14:30 (room 2014, building 660) (see location):

Hugo Richard

(Parietal/TAU teams)

Title: Data based alignment of brain fmri images


Abstract:

There is a very important biological variability between human brains both in an
anatomical and functional point of view. It is influenced by genetics but also by lifestyle or
social environment. It is also linked to behavioral differences and diseases.
This variability between human brains results in differences in the spatial arrangement
of functional responses that can be observed through functional imaging. This constitutes a
fundamental challenge specific to human brain understanding.
Standard registration methods only make it possible to match brain organization across
individuals at a coarse scale, hence they typically lose fine-grain details. However, the ongoing
accumulation of brain data at the individual level opens a new perspective: that of a feature-based alignment of brain activity.

Different feature alignment algorithms have been studied such as the hyperalignment and the shared response model : two linear models using Procrustes transform to build more efficient registration methods. We jointly call them feature-based alignment approaches. Improvements were made to these two methods, using different constrains and penalization schemes. All models were benchmarked in terms
of their ability to deal with unseen data.



Contact: guillaume.charpiat at inria.fr