Brain-Cognition Fingerprinting via Graph-GCCA with Contrastive Learning.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Longitudinal neuroimaging studies are crucial for understanding brain aging and diseases.
• Accurate encoding of the multidimensional relationship between brain function and cognition is needed.
• Individual variability over time must be accounted for in these analyses.

Purpose of the Study:

• To propose an unsupervised learning model, CoGraCa, for encoding the dynamic relationship between brain function and cognition.
• To create individualized and multimodal brain-cognition fingerprints using contrastive learning.
• To capture unique neural and cognitive phenotypes for each individual.

Main Methods:

• Developed Contrastive Learning-based Graph Generalized Canonical Correlation Analysis (CoGraCa).
• Employed Graph Attention Networks and generalized Canonical Correlational Analysis for relationship encoding.
• Utilized individualized and multimodal contrastive learning for fingerprint creation.

Main Results:

• Applied CoGraCa to longitudinal resting-state fMRI and cognitive data from healthy individuals.
• Generated fingerprints effectively captured significant individual differences.
• CoGraCa outperformed existing single-modal and CCA-based multimodal models in identifying sex and age.

Conclusions:

• CoGraCa provides an effective method for creating interpretable brain-cognition fingerprints.
• The model enhances the understanding of individual variability in brain aging and disease.
• CoGraCa offers interpretable insights into the interactions between brain function and cognition.