Resolution-based spectral clustering for brain parcellation using functional MRI.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Individual physiological variability necessitates data-driven brain parcellation.
• Recent research focuses on simultaneous network structure estimation and partitioning.
• Effective exploitation of neuroimaging data relies on accurate brain parcellation.

Purpose of the Study:

• To develop a brain parcellation method that generalizes to unseen data.
• To define brain parcels using an uncertainty quantification approach.
• To optimize preprocessing and parcellation strategies for functional imaging.

Main Methods:

• A predictive modeling framework for data preprocessing, parcellation, and validation.
• Utilizing uncertainty quantification to define parcels as groups of unresolvable variables.
• Cross-validation for model parameter selection and comparison based on repeatability and held-out data.

Main Results:

• The proposed method yields more accurate and repeatable brain parcellations compared to baseline clustering.
• Identified potential overfitting issues with baseline methods and bias in other approaches.
• Provided strategies for selecting evaluation metrics, model order, and tuning preprocessing.

Conclusions:

• Clustering of resolution offers a principled and robust brain parcellation approach.
• The proposed method significantly improves upon current baseline techniques.
• This approach enhances the analysis of functional neuroimaging data.