In longitudinal fMRI studies the challenge is to localize in the brain the effects of a treatment interleaving two recordings. The issue is to assess how the treatment affects the BOLD response, independently of the underlying inherent variance of the measured signal, caused by the subject variability or by the scanner sensitivity to environmental conditions. In this work we propose a model-free method able to compute a brain map capturing the effects of the treatment. The approach, performs a pairwise similarity-based analysis of two fMRI sessions using a state-of-the-art multivariate approach. We illustrate the empirical results on a dataset concerned with a study on aphasia rehabilitation. The pairwise method allows to reproduce the same brain areas obtained with a reference approach based on GLM analysis. In addition the proposed method highlights new brain regions that are compliant with a neuroscientific interpretation.
|Number of pages||8|
|Journal||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Publication status||Published - 2012|