Spectral Clustering of fMRI Data within Regions of Interest: Clarification of L-dopa effects in Parkinson's Disease

Identifying active regions of the brain that are task-related is important in fMRI study. Current methods of determining functional regions of interest (ROIs) are unsatisfactory because they either reduce the effect size or bias the statistical results. We propose a spectral clustering method for assessing those voxels within an ROI that are suitable for further task-activation analysis. Different similarity functions are studied and the correlation index is chosen based on the simulation study. In real fMRI study, further group analysis employing regression is investigated to identify different brain activation patterns between groups in order to reveal the effects of disease and medicine. A real fMRI case study in Parkinson's disease suggests that the technique is promising, warranting further study.