Retrospective Cluster Size Thresholding for MRF-based Detection of Activated Regions in fMRI

Detection of activation regions in functional Magnetic Resonance Imaging (fMRI) experiments can be improved by cluster size thresholding, where small regions are removed in the post-processing step. It is believed that such small regions are spurious and hence can be removed without affecting the truly activated ones. We show that in the context of Markov Random Field (MRF) based segmentation, simple removal of small regions after the segmentation is not optimal. The developed retrospective correction approach finds the best modification (removal or growth) and achieves superior performance to that of the standard MRF segmentation.