Cluster analysis for defining regions-of interest using PET, CT, and MR images following human traumatic brain injury

One of the major factors that have hindered the reproducible results of traumatic injured brains in the literature has been the inconsistency of region definitions. The aim of this study was to develop a method that minimizes the operator dependency and standardizes the procedure in the definition of regions-of-interest (ROI). Eight traumatic brain injury (TBI) patients with intracerebral confusions were studied. Each patient had an acute CT, MR and quantitative PET (cerebral blood flow (CBF) and oxygen extraction fraction (OEF)) studies (59±31 hrs post-injury). We generated cluster images by applying cluster analysis on the voxel values of CT/MR images. The voxels that were traditionally characterized as contusion or edematous tissue by neurosurgeons were grouped into an individual cluster in the cluster images. By using these voxels to define an ROI, we obtained consistent physiological results among the 8 patients. This study demonstrated that definition of ROI using cluster analysis is robust and useful in study of the physiology of the injured brains.