Start date: 1st of December 2013
Learning functional brain atlases modeling inter-subject variability
Recent studies have shown that resting-state spontaneous brain activity unveils intrinsic cerebral functioning and complete information brought by prototype task study. From these signals, we will set up a functional atlas of the brain, along with an across-subject variability model. The novelty of our approach lies in the integration of neuroscientific priors and inter-individual variability in a probabilistic description of the rest activity. These models will be applied to large datasets. This variability, ignored until now, may lead to learning of fuzzy atlases, thus limited in term of resolution. This program yields both numerical and algorithmic challenges because of the data volume but also because of the complexity of modelisation.
Studying functional MRI, applied to prototype tasks, has lead to the estasblishment of brain functional maps, where each area is related to a cognitive function. However, recent developments in neurosciences have shown that, when no specific task is asked, fMRI exhibits the intrisic functioning of the brain. Systematic study of this activity, e.g. using Independent Component Analysis, could lead to the identification of new functional cerebral areas. This study raises the question of data processing, for which we have decided to use source separation and clustering techniques, but also modelisation because our atlas will have to integrate inter-individual variability. This analysis could provide therapeutic informations that would be helpful for rehabilitatients of stroke patients.
The objective of this project is to furnish an atlas of brain functional regions, along with a model of inter-individual variability. This atlas is useful as it provides a brain parcellation that is independant of any cognitive protocole. In clinical applications, biomarker extraction could be restrained to useful regions picked in our atlas. To enhance the neuroscientific relevance of our results, we will reuse penalities that have proven useful in the context of cognitive tasks. In order to run our algorithm on thousands of subjects, we may implement our algorithms on computer grids and use stochastic approximation algorithms like on-line learning methods.
Dimitris is the director of Image Analysis lab at Stony Brook Univeristy. He has a strong background in computer vision and computer science and thus has its own approach of neurosciences. I hope that it will lead me on tracks that I would not have think of.
http://www.gael-varoquaux.info is formerly a physicist that fell into the machine learning pot. He also has his own original approach of the problems: he particularly forces us to develop an intuition about we are doing. You can find his name in almost all Python libraries at some point.
Bertrand is the head of the Parietal team. You can find more information about him (and his publications) on his website.