A multistep unsupervised fuzzy clustering analysis of fMRI time series

A paradigm independent multistage strategy based on the Unsupervised Fuzzy Clustering Analysis (UFCA) and its potential for fMRI data analysis are pre sented. The influence of the fuzziness index is studied using Receiver Oper ating Characteristics (ROC) methodology and an interval of choice, around t he widely used value 2, is shown to yield the best performance. The ill-bal anced data problem is also overcome using a pre-processing step to reduce t he number of voxels presented to the method. Statistical and anatomical cri teria are proposed to exclude some voxels and enhance the UFCA sensitivity. An Original postprocessing step aiming at statistically characterizing the obtained clusters is also developed. Two similarity criteria are used: the correlation coefficient on temporal profiles and a novel fuzzy overlap coe fficient on membership degree maps. This final step provides a useful analy sis tool to study intra-individual reproducibility of the classes across se ries (stimulation vs. stimulation, noise vs, noise or stimulation vs. noise ). Finally, a comparison between this technique and some existing or locall y developed postprocessing algorithms is presented using ROC methods. Its s ensitivity and robustness is compared to the classical FCA or other techniq ues as a function of several parameters such as Contrast-to-Noise Ratio (CN R) and noise amplitude. Even without knowledge about the paradigm, the hemo dynamic response function and the number of clusters, the performances of t he proposed strategy are comparable to those of the classical approaches wh ere extensive prior knowledge has to be added. (C) 2000 Wiley-Liss, Inc.