Channel Graph Neural Network Revealing Multimodal Brain Connectivity Abnormalities in Schizophrenia.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Schizophrenia research benefits from understanding abnormal brain network characteristics.
• Graph learning excels at integrating multimodal data for complex network analysis.
• Existing methods may struggle with precise localization of network abnormalities.

Purpose of the Study:

• To propose a channel-based graph neural network (C-GNN) for improved multimodal data integration and accurate localization of brain network abnormalities in schizophrenia.
• To enhance the understanding of disease mechanisms and identify potential intervention targets.

Main Methods:

• Node embedding was used to capture structural connectivity patterns in brain regions.
• A branched attention module with channel attention adaptively identified critical brain regions.
• A graph feature-constraint module extracted salient features by analyzing differences across feature channels.

Main Results:

• The C-GNN model achieved 84.37% accuracy in classifying individuals with schizophrenia.
• Interpretability analysis identified specific abnormal brain regions (e.g., orbital cortex, temporal fusiform cortex, lingual gyrus).
• Key multimodal metrics, including cortical thickness and ReHo, were found to be significant contributors to classification.

Conclusions:

• The C-GNN model effectively integrates multimodal data to reveal schizophrenia-related brain network alterations.
• Findings provide insights into neural alterations in schizophrenia and support the development of targeted interventions.