Robust brain parcellation using sparse representation on resting-state fMRI.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Mapping

Background:

• Resting-state fMRI (rs-fMRI) is crucial for brain parcellation, identifying distinct functional modules via connectivity patterns.
• Existing rs-fMRI parcellation methods show inconsistencies, potentially due to noise, hindering reliable brain segmentation.

Purpose of the Study:

• To develop a robust brain parcellation method for rs-fMRI data.
• To enhance the accuracy and reproducibility of brain module identification using advanced signal processing techniques.

Main Methods:

• Constructed a sparse similarity graph using sparse representation coefficients of seed voxels.
• Applied spectral clustering to identify distinct brain modules based on the graph.
• Utilized both local time-varying BOLD signals and whole-brain connectivity as features.

Main Results:

• The sparse representation method demonstrated high accuracy and robustness to noise on simulated rs-fMRI data.
• Achieved stable and reproducible parcellation of the medial frontal cortex (MFC) and parietal operculum (OP) across multiple real rs-fMRI datasets.
• Parcellation of MFC was minimally affected by spatial smoothing, and OP parcellation correlated well with cytoarchitectonic and somatotopic data.

Conclusions:

• The proposed sparse representation-based method offers a promising approach for robust rs-fMRI brain parcellation.
• This technique enhances the reliability and consistency of brain module identification, overcoming limitations of previous methods.
• The findings support the use of sparse representation for precise and reproducible brain mapping using resting-state functional connectivity.