ERG-Graph: Graph Signal Processing of the Electroretinogram for Classification of Neurodevelopmental Disorders.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Objective biomarkers for neurodevelopmental disorders are needed.
• Electroretinogram (ERG) shows promise but current methods have limited accuracy.
• Existing classification approaches struggle with multi-group scenarios.

Purpose of the Study:

• Introduce ERG-Graph, a novel graph signal processing (GSP) framework for ERG analysis.
• Extract topological and spectral features from ERG waveforms.
• Improve classification accuracy for autism spectrum disorder (ASD) and attention deficit/hyperactivity disorder (ADHD).

Main Methods:

• Transformed ERG waveforms into weighted, undirected graphs using amplitude quantization and temporal-adjacency.
• Extracted nine graph-theoretic features (topological and spectral).
• Evaluated features using machine learning classifiers on 278 participants (ASD, ADHD, ASD+ADHD, Control).

Main Results:

• ERG-Graph features achieved high balanced accuracies (e.g., 0.91 for ASD vs. control).
• Fusion of ERG-Graph and time-domain features improved three-group classification accuracy to 0.81 (vs. 0.70 benchmark).
• Graph-theoretic features were dominant predictors, showing significant topological differences between groups.

Conclusions:

• Graph-based topological features capture unique discriminative information in ERG signals.
• ERG-Graph advances objective biomarker development for neurodevelopmental disorders.
• This approach offers a promising tool for clinical screening of ASD and ADHD.