Some PCA for fMRI experiments

  • Looking at a one subject analysis using PCA leads to analyse a brain x time data matrix. Here we see first of all the results are different from subject to subject (obviously).

    summary of the decomposition Sujet 1

    summary of the decomposition Sujet 2

    summary of the decomposition Sujet 8

    We show the best correlated to paradigm. Note that the old second order statistics still performs well to detect something very well structured (the paradigm). Nonetheless variation from subject to subject encline to use some group analysis method such as a PTA3-modes to able to give a "population" describtion or result.

  • Now the same thing using a non-identity metric a -the inverse of a robust estimate of the within "group"(without knowing the brain group structure). When the data is very large this can be prohibitive directly but supposing the group struture sparse can be used to split the computation.

    summary of the decomposition Sujet 1

    see some group analysis using PTAk


    Didier Leibovici
    Last modified: Mon Jul 16 14:30:04 BST 2001