Identification of Young High-Functioning Autism Individuals Based on Functional Connectome Using Graph Isomorphism...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Autism spectrum disorder (ASD) is characterized by altered functional connectivity in the brain.
• Machine learning approaches are increasingly used for ASD classification based on neuroimaging data.
• Graph Neural Networks (GNNs) offer a novel framework for analyzing brain network features in neurological disorders.

Purpose of the Study:

• To develop an improved Graph Isomorphism Network (GIN) model for enhanced ASD classification.
• To increase the interpretability of machine learning models by considering node importance.
• To identify potential neuroimaging biomarkers for ASD using functional connectome data.

Main Methods:

• Proposed an enhanced Graph Isomorphism Network (GIN) model incorporating the Weisfeiler-Lehman (WL) graph isomorphism test.
• Applied the GIN model to multisite resting-state functional connectome data from the Autism Brain Imaging Data Exchange (ABIDE) dataset.
• Evaluated the model's performance against established classification methods using five metrics and analyzed salient brain regions.

Main Results:

• The improved GIN model demonstrated superior performance compared to other classification methods across five evaluation metrics.
• The study successfully identified significant Regions of Interest (ROIs) within the visual and frontoparietal control networks.
• The model's enhanced interpretability allowed for the identification of key network features contributing to ASD classification.

Conclusions:

• The developed GIN model offers a promising and interpretable approach for classifying autism spectrum disorder using functional connectivity.
• Identified brain regions may serve as potential neuroimaging biomarkers for ASD diagnosis and further research.
• This work highlights the potential of advanced GNNs in understanding the neural underpinnings of ASD.