Brain MR Perfusion Image Segmentation Using Independent Component Analysis and Hierarchical Clustering

Extraction of various perfusion components from dynamic-susceptibility-contrast (DSC) MR brain images is critical for the analysis of brain perfusion. According to the variation of temporal signal on different brain tissues, one can segment whole brain area into distinct blood supply patterns which are vital for the profound analysis of cerebral hemodynamics. In this study, independent component analysis (ICA) is used to project the perfusion image data into independent components from which each elucidated tissue cluster can be automatically segment out by using the hierarchical clustering (HC). Five normal subjects and a case of internal carotid artery stenosis subjects were analyzed. The results demonstrated that ICA-HC is effective in multi-tissue hemodynamic classification which improves differentiation of pathological and physiological hemodynamics.