A data-driven fMRI analysis method using temporal clustering technique and an adaptive voxel selection criterion

A data-driven fMRI (functional magnetic resonance imaging) analysis method is proposed for studying brain responses to stimulation when the information for predicting their onset or duration is un-available. The method is suitable for experiments involving a single event or non repetitive multiple events. The method is built upon the pre-existing temporal clustering analysis techniques with additional features that make use of the signal changes of the neighbouring voxels to ensure that the selected voxels for response detection are those most likely to have been activated by the stimuli. For method validation, eight sets of fMRI data from three different kinds of sensory experiments are applied. The results demonstrated that our method is able to detect the time bins during which the stimuli were administered and the identified voxels corresponding to the brain areas, which are typically activated in this kind of experiments. Moreover, in these eight sets of data, the accuracy of stimulation response detection is 75% compared to 58.33% without the selection criterion.