Studying Sub-Dendrograms of Resting-State Functional Networks with Voxel-Wise Hierarchical Clustering.

Yanlu Wang,Tie-Qiang Li,Mussie Msghina

doi:10.3389/fnhum.2016.00075

Abstract

Hierarchical clustering is a useful data-driven approach to classify complex data and has been used to analyze resting-state functional magnetic resonance imaging (fMRI) data and derive functional networks of the human brain at very large scale, such as the entire visual or sensory-motor cortex. In this study, we developed a voxel-wise, whole-brain hierarchical clustering framework to perform multi-stage analysis of group-averaged resting-state fMRI data in different levels of detail. With the framework we analyzed particularly the somatosensory motor and visual systems in fine details and constructed the corresponding sub-dendrograms, which corroborate consistently with the known modular organizations from previous clinical and experimental studies. The framework provides a useful tool for data-driven analysis of resting-state fMRI data to gain insight into the hierarchical organization and degree of functional modulation among the sub-units.

Highlights

Different clustering techniques have been used for exploratory analysis of resting-state functional magnetic resonance imaging data aimed to group together functionally similar voxels or regions of interests (ROIs) and identify functionally connected brain networks
somatosensory motor (SSM) system in the human brain consists of S1, M1, and some pre/post- central gyrus areas divided into dorsal and ventral subgroups in addition to the parietal operculum and the auditory cortex (Power et al, 2011)
Comparing with previously determined probabilistic maps of visual topography, we found that the boundaries of cluster 6 with clusters 7–9 are consistent with the known boundaries between hV4/VO1 and V3v

Summary

Introduction

Different clustering techniques have been used for exploratory analysis of resting-state functional magnetic resonance imaging (fMRI) data aimed to group together functionally similar voxels or regions of interests (ROIs) and identify functionally connected brain networks These include, among others, fuzzy C-means (Baumgartner et al, 1997; Hilgetag et al, 2000), spectral clustering (Snyder et al, 1997; Mattingley et al, 1998), K-means clustering (Fogassi et al, 2005), hierarchical clustering (Cordes et al, 2002; Foxe et al, 2002; Menon and Uddin, 2010; Wang and Li, 2013), consensus clustering (Moretti and Muñoz, 2013), and constrained clustering (Snyder et al, 1997; Foxe et al, 2002). Extraction and characterization of such a hierarchical organization is an important issue in the study of brain function networks

Objectives

Methods

Results

Conclusion