I develop algorithms and models for extracting salient and reproducible spatial features from the correlation structure of functional MRI images without using a paradigm, such as in resting-state studies. I focus on group models, opening the door to between-subject comparisons.
We have adapted spatial Independent Component Analysis (ICA) to extract reproducible, sparse, brain maps. Spatial ICA is a widely-used technique in neuroimaging to extract brain networks.
|We have introduced a multivariate random effects group model to conduct multi-subject analysis with good reproducibility.|
|We formulate ICA as a sparse-recovery problem to give statistical control on the extracted brain maps base on a probabilistic model of the noise based on sole assumption that the interesting latent factors are sparsely-activated.|
How to compare correlation structures to extract differences in the underlying graphical models that can explain the observations?
Finding from large correlation structures which edges can explain differences is an ill-posed problem. We are developing statistical models and machine learning tools to tackle this challenge.
We apply machine-learning techniques for multivariate brain decoding: starting from brain activity data we find predictive features of behavior.
Papers in preparation: Total Variation regularization enhances regression-based brain activity prediction