A hybrid graph network model for ASD diagnosis based on resting-state EEG signals

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Autism spectrum disorder (ASD) is a prevalent neurodevelopmental disorder requiring early diagnosis for effective intervention.
• Reliable biomarkers are crucial for understanding ASD's etiology and enhancing diagnostic precision.
• Electroencephalography (EEG) signals offer a promising avenue for identifying stable biomarkers in ASD diagnosis.

Purpose of the Study:

• To introduce Rest-HGCN, a novel hybrid graph convolutional network framework for ASD diagnosis.
• To leverage resting-state EEG signals for capturing differential brain connectivity patterns between typically developing children and those with ASD.
• To develop a robust and accurate diagnostic system for ASD using graph learning strategies.

Main Methods:

• The study proposes a hybrid graph convolutional network (Rest-HGCN) integrating brain network analysis and data-driven approaches.
• Resting-state EEG signals were utilized to extract discriminative graph features representing brain connectivity.
• Graph learning strategies were employed to identify differential patterns indicative of ASD.

Main Results:

• The Rest-HGCN model demonstrated high accuracy in ASD diagnosis on the ABC-CT resting EEG dataset.
• Achieved 87.12% accuracy in single-subject analysis and 85.32% in cross-experiment analysis.
• The framework effectively captures discriminant graph patterns from resting EEG signals for robust diagnosis.

Conclusions:

• Rest-HGCN provides an effective and convenient tool for clinical ASD diagnosis.
• The model's ability to capture differential graph patterns highlights the potential of EEG-based biomarkers.
• This approach offers a promising direction for improving the accuracy and efficiency of ASD diagnostic tools.