Thomas Bazeille

"Template inference after arbitrary alignments" under the supervision of B. Thirion (PARIETAL). There exist a natural anatomical and functional brain variability across individuals, that hampers the generalization of effects observed in individual studies to groups. Functional alignment are promising techniques that learn transforms capturing this variability from one subject to another. They can be used to transfer signal across subjects to better tackle better prediction tasks such as: stimuli to brain activation(encoding) or brain to stimuli(decoding). These techniques are also a way promising way of finding more consistent group effects in small sample cognitive experiments. However, they are primarly designed to capture variability between pairs of subject which is highly impractical and costly to transfer many subjects signals to a new problem. The goal of my phd is to assess their potential contributions and design a way to synthetize all subjects informations in a meaningful template. This can be used as a functional atlas deformable to transfer it's information to a new subjects own functional space. To do this, we're using mathematical tools such as Optimal transport, Wasserstein barycenters, ensembling, and developing evaluation frameworks to measure similarities of brain states complex observational data.

During this internship, I assessed the performance of various methods proposed in the litterature to realign functional MRI images across subject, i.e. find transform that can capture the intrinsic variability that exist across individuals.