MJNMF-GAT: Multi-Task Joint Non-Negative Matrix Factorization Graph Attention Network for Understanding Adolescent...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Understanding neurodevelopment requires integrating multi-task brain imaging data.
• Functional magnetic resonance imaging (fMRI) and graph neural networks (GNNs) are powerful tools for analyzing brain activity and network interactions.

Purpose of the Study:

• To introduce a novel framework, the Multi-Task Joint Non-Negative Matrix Factorization Graph Attention Network (MJNMF-GAT), for analyzing multi-task fMRI data.
• To investigate age-related differences in adolescent neurodevelopment using fMRI data.
• To enhance predictive performance and network interpretation in neuroimaging studies.

Main Methods:

• Integrated joint non-negative matrix factorization (J-NMF) for shared latent feature extraction across tasks.
• Employed graph attention networks (GATs) to model brain connectivity patterns.
• Utilized GNNExplainer for model interpretability and identification of key subnetworks.

Main Results:

• Achieved superior age prediction performance on the Philadelphia Neurodevelopmental Cohort (PNC) dataset with RMSE of 1.9212 ± 0.1742 and MAE of 1.5368 ± 0.1469.
• Demonstrated high correlation (0.8032 ± 0.0469) in age prediction.
• Identified critical functional connections and subnetworks that change with adolescent neurodevelopmental stages.

Conclusions:

• The MJNMF-GAT model offers improved predictive accuracy for age-related brain changes.
• The framework provides interpretable insights into informative brain connectivity patterns.
• This approach advances multi-task neuroimaging analysis for understanding adolescent neurodevelopmental changes.