An Optimized Clustering Technique for Functional Parcellation of Hippocampus

Introduction Functional sub-division of important anatomic regions in the human brain is normally based on anomaly in structural connectivity patterns [1] or functional connectivity maps, after subdividing the region of interest on trial basis [2]. Quantification of functional heterogeneity, and determining number of sub-regions on the basis of that, has rarely been a focus of study. This work is centered around implementation of self organized maps (SOM) to classify the functionally different regions in the hippocampus as it exhibits functional and anatomical differences in patients with disorders such as schizophrenia, bipolar disorder, and depression [3]-[5]. We rigorously tested the performance of SOM with conventional k-mean clustering techniques to optimize how many heterogeneous compartments the left hippocampus possesses on the basis of connectivity maps associated with each voxel in the ROI. We designed and used an in house software with SPM5 to parcellate functionally heterogeneous ROIs using SOM.