Innovative biomarker exploration in ASD: Combining Graph Neural Networks and permutation testing on fMRI data

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Autism Spectrum Disorder (ASD) diagnosis relies on behavioral observation, lacking objective biomarkers.
• Understanding the neural basis of ASD is crucial for developing diagnostic tools.
• Functional Magnetic Resonance Imaging (fMRI) provides insights into brain network activity.

Purpose of the Study:

• To identify potential brain-based biomarkers for Autism Spectrum Disorder (ASD).
• To leverage Graph Neural Networks (GNNs) for analyzing brain network differences.
• To explore novel brain regions associated with ASD using unsupervised GNNs.

Main Methods:

• Employed unsupervised Graph Neural Networks (GNNs) to extract node embeddings from brain regions.
• Utilized functional Magnetic Resonance Imaging (fMRI) data from ASD and control groups.
• Applied permutation tests to identify significant differences in node embeddings between groups.

Main Results:

• Identified significant differences in brain region embeddings between ASD and control groups.
• Confirmed known affected areas such as the cerebellum, temporal lobe, and occipital lobe.
• Discovered novel regions including Vermis_3, Vermis_4_5, Fusiform areas, Parietal, and Cuneus.

Conclusions:

• Graph Neural Networks show promise for Autism Spectrum Disorder (ASD) biomarker discovery.
• The identified brain regions require further investigation for validation and clinical relevance.
• This approach advances the potential for objective ASD diagnosis and understanding its neural underpinnings.