Cluster validation indices for fMRI data: Fuzzy C-Means with feature partitions versus cluster merging strategies

Fuzzy C-Means (FCM) is a standard technique for exploratory analysis and is readily adaptable to integrate unique data characteristics and auxiliary feature relations. Distinguishing between the spatial and temporal features of functional magnetic resonance imaging (fMRI) time courses (TC) has proved effective in reducing the presence of false positives for stimulation studies. The fuzzy partitions generated by this FCM variant (FCMP) are compared to several cluster merging techniques using cluster validation indices. These indices quantify the degree to which a dataset justifies a particular membership partition. A basic cluster merging strategies is examined where closest samples in a distance matrix are merged. A novelty is the use of alternate centroid definitions. Finally, the dynamic modeling employed by the CHAMELEON clustering algorithm is examined. All algorithms are evaluated on a Tourette´s fMRI dataset.