Normalized cut group clustering of resting-state FMRI data
Martijn van den Heuvel1, Rene Mandl, Hilleke Hulshoff Pol
1Rudolf Magnus Institute of Neuroscience, Department of Psychiatry, University Medical Center Utrecht, Utrecht, The Netherlands. M.P.vandenheuvel@umcutrecht.nl
Plos One
|April 24, 2008
Summary
Researchers developed a novel clustering method to identify resting-state networks (RSNs) in the brain. This approach automatically determines the optimal number of RSNs, revealing 7 distinct networks including motor, auditory, and default mode networks.
Area of Science:
- Neuroscience
- Brain Imaging
- Network Analysis
Background:
- Distinct brain regions exhibit coordinated spontaneous activity during rest, forming resting-state networks (RSNs).
- Previous RSN analysis methods, including seed-based and model-free approaches, often require manual selection of networks for group studies.
- Group analysis of RSNs remains complex, often necessitating human intervention.
Purpose of the Study:
- To develop and validate a voxel-based, model-free normalized cut graph clustering approach for group analysis of resting-state functional magnetic resonance imaging (fMRI) data.
- To automate the identification and characterization of resting-state networks (RSNs) across a group of subjects without requiring manual input.
- To determine the optimal number of RSNs through a data-driven clustering fit.
Main Methods:
- Utilized a voxel-based, model-free normalized cut graph clustering algorithm for whole-brain resting-state fMRI data.
- Inter-voxel time-series correlations were grouped at the individual level.
- Group-level clustering of individual networks identified consistent group resting-state networks (RSNs).
- Scanned 26 subjects at rest using a fast BOLD-sensitive fMRI protocol on a 3 Tesla scanner.
Main Results:
- An optimal group clustering fit identified 7 distinct resting-state networks (RSNs).
- The identified RSNs encompassed motor/visual, auditory, attention, and the default mode network.
- The discovered RSNs demonstrated significant overlap with previously reported resting-state findings.
- The results support the existence of spatially discrete RSNs during human brain rest.
Conclusions:
- The novel clustering approach effectively identifies group-level resting-state networks (RSNs) in a data-driven manner.
- The study confirms the presence of 7 consistent RSNs, including known networks like the default mode network.
- This method offers an automated and robust alternative for analyzing resting-state fMRI data in group studies.
