Dementia classification using a graph neural network on imaging of effective brain connectivity

Area of Science:

• Neuroscience and Computational Psychiatry
• Medical Imaging and Machine Learning

Background:

• Alzheimer's disease (AD) and Parkinson's disease (PD) are leading neurodegenerative conditions.
• Effective brain connectivity (EBC) shows potential for differentiating AD, PD, and healthy controls (HC).
• Current methods for using EBC in disease diagnosis are limited, especially with complex brain networks.

Purpose of the Study:

• To develop and validate a novel graph neural network (GNN) approach for diagnosing neurodegenerative diseases using EBC.
• To explore the effectiveness of combining EBC estimations with power spectrum density (PSD) features for classification.
• To create a framework for simultaneously applying univariate and multivariate features in disease diagnosis.

Main Methods:

• Development of a directed structure learning GNN (DSL-GNN) tailored for processing directional information in EBC.
• Application of DSL-GNN to EBC estimations and PSD features from neuroimaging data.
• Validation of the DSL-GNN framework in four discrimination tasks: AD vs. HC, PD vs. HC, AD vs. PD, and AD vs. PD vs. HC.

Main Results:

• The proposed DSL-GNN achieved superior performance compared to existing methods across all discrimination tasks.
• Highest accuracies were recorded at 94.0% (AD vs. HC), 94.2% (PD vs. HC), 97.4% (AD vs. PD), and 93.0% (AD vs. PD vs. HC).
• The framework successfully integrated univariate and multivariate features for enhanced classification.

Conclusions:

• The developed DSL-GNN provides a robust analytical framework for analyzing complex brain networks with causal directional information.
• This approach demonstrates significant potential for improving the diagnostic accuracy of common neurodegenerative diseases like AD and PD.
• The study highlights the effectiveness of combining GNNs with EBC for advancing neurodegenerative disease research.