Unsupervised fuzzy clustering analysis of fMRI series

The potential of an fMRI data analysis strategy using a paradigm independent unsupervised fuzzy clustering is presented. The power of the method is demonstrated to discriminate different types of responses without prior knowledge relative to the paradigm in contrast with classic statistical methods which require a priori knowledge about the stimulus. Different performance measure functions are proposed to solve the cluster validity problem and to detect the number of substructures present in the data. The results are presented on both simulated data (with Contrast to Noise Ratios similar to those observed in fMRT), and in vivo EPI data for a motor paradigm. This new data analysis procedure may be very useful for optimization of data analysis and quality in fMRI.