Automated individual cortical parcellation via consensus graph representation learning.

Area of Science:

• Neuroscience
• Brain Imaging
• Computational Biology

Background:

• Cortical parcellation is crucial for understanding brain organization.
• Existing functional magnetic resonance imaging (fMRI) methods struggle to balance individual brain specificity with group consistency.
• Generating high-quality, subject-consistent cortical parcellations remains a significant challenge.

Purpose of the Study:

• To propose a fully automated method for individual cortical parcellation.
• To achieve an adaptive balance between intra-individual specificity and inter-individual consistency.
• To improve the quality and reliability of brain network representations.

Main Methods:

• Developed a novel method based on consensus graph representation learning.
• Integrated spectral embedding with low-rank tensor learning into a unified optimization model.
• Utilized group-common connectivity patterns to optimize individual functional networks and eliminate spurious connections.

Main Results:

• The proposed method demonstrated superior performance on a Human Connectome Project (HCP) test-retest dataset.
• Outperformed existing methods in reproducibility, functional homogeneity, and alignment with task activation.
• Functional networks derived from this method showed enhanced capabilities in gender identification and behavior prediction on the HCP S900 dataset.

Conclusions:

• The consensus graph representation learning method offers a robust solution for automated cortical parcellation.
• This approach effectively balances individual specificity and inter-individual consistency in brain network analysis.
• The improved functional networks hold promise for advancing neuroimaging research and applications.