EVIDENT: A Two-Stage Strategy for the Exploratory Analysis of Functional MRI Data by Fuzzy Clustering

We describe EVIDENT, a two-stage strategy, based on fuzzy cluster analysis (FCA), as a realization of Exploratory Data Analysis (EDA), specifically designed for analyzing functional MR neuroimaging data. The first stage involves time-course (TC) normalization (scaling), and preselection, the latter based on " trend " and " noise " tests. In the second stage, FCA is applied to the selected TCs, followed by allocation of the initially exc luded TCs to the clusters found by the FCA. We show that employing the two-stage EVIDENT process not only speeds up execution significantly, it also improves results relative to clustering that did not use preselection. We compared the computational and other consequences of this strategy on the quality of the clustering results when using different distance measures, normalization, and preselection options. Several phantom data sets and three real fMRI data sets were analyzed in detail. Full timing studies were carried out on sixteen fMRI data sets, using all combinations of six normalizing, and two distance options. We propose speed and robustness (noise resistance, reproducibility) as the two major requirements that any practically effective EDA method ought to satisfy. We show that EVIDENT fulfills these requirements.