Unsupervised Parcellation Of Brain Regions Based On Anatomical Connectivity And Application

The brain is a highly intelligent biological system.It is of great significance to construct a more detailed brain map for the development of brain science.Diffusion Tensor Image?DTI?is an imaging technique for tracking white fibers in vivo.With high spatial resolution,DTI is widely used in the study of brain anatomy.Hippocampus is mainly responsible for learning,memory,lateralization,and emotional response.Previous studies have shown that there is functional and structural heterogeneity within the hippocampus.In addition,hippocampus has been involved in pathophysiology of medial temporal lobe epilepsy.Based on DTI data,this paper investigated the anatomical subregions of the hippocampus.Then,anatomical changes in unilateral medial temporal lobe epilepsy were examined and made pattern classification.The main content of this paper is:Proposing a new voxel-level probability anatomical connectivity based brain regions parcellation method and achieving fine partition of hippocampus.In this paper,anatomical connectivity from each hippocampus voxel to the whole brain voxel was selected as feature in individual space.Unsupervised clustering algorithm was used and divided each hippocampus into 2 to 4 subregions.According to the results of individual parcellation,registration,template overlay,label transforms,clustering algorithm was used to get the group-level parcellation.Results showed that there was structural heterogeneity along the anterior to posterior axes of the hippocampus.The medial temporal lobe epilepsy was used as an example to examine the abnormal patterns of the patients' hippocampus anatomical connectivity which verify the rationality of parcellation.Changes of hippocampal anatomical connectivity of left and right mesial temporal lobe epilepsy and healthy subjects were examined.According to the study,patients with unilateral mesial temporal lobe epilepsy had weakened or enhanced anatomical connectivity which may be related to the declined episodic memory,depression and the different emotional expressions within left and right temporal lobe epilepsy.Multi-scale modularity analysis of hippocampus was adopted which achieved higher classification accuracy.The left and right mesial temporal lobe epilepsy patients and healthy subjects were classified,using support vector machine,based on anatomical connectivity.According to the result,within the left side patients and healthy subjects,left hippocampus had the highest classification accuracy of 78%;within the right side patients and healthy subjects,the right middle posterior subregion?R_4MP?had the highest classification accuracy of 79.7%,within the left and right side patients,the left middle posterior subregion?L_4MP?subregion had the highest classification accuracy of 74.4%.When feature fusion was used,the classification accuracy can reach at 81.6%,79.7% and 83.7% respectively.