Fuzzy approach to incorporate hemodynamic variability and contextual information for detection of brain activation

We propose to use fuzzy c-means clustering with contextual modeling on features extracted from fMRI data for detection of brain activation. Five discriminating features are extracted from fMRI data by using a sequence of temporal-sliding-windows. Fuzzy membership maps of individual subjects obtained through clustering with spatial regularization is capable of taking into account both hemodynamic variability and contextual information of brain activation. The present method outperforms statistical parametric mapping (SPM) approach on experiments with synthetic fMRI data contaminated by both independent and correlated noise. Performance on real fMRI data are comparable to those obtained with SPM.