Classification of Autism Spectrum Disorder Using rs-fMRI data and Graph Convolutional Networks

Area of Science:

• Neuroscience
• Data Science
• Medical Imaging

Background:

• Autism spectrum disorder (ASD) impacts many individuals globally, necessitating early and accurate diagnosis.
• Resting-state functional magnetic resonance imaging (rs-fMRI) reveals distinct neural patterns in individuals with ASD.
• High-dimensional MRI data requires advanced analytical methods for effective interpretation.

Purpose of the Study:

• To develop and evaluate a novel Graph Convolutional Network (GCN) model for classifying individuals with ASD from healthy controls (HC) using rs-fMRI data.
• To integrate graphlet topological counting as a feature within the GCN framework to enhance diagnostic capabilities.
• To assess the model's accuracy and interpretability in differentiating ASD from HC.

Main Methods:

• Utilized resting-state fMRI (rs-fMRI) data from the ABIDE-I dataset, comprising 1035 subjects.
• Developed a novel Graph Convolutional Network (GCN) model incorporating traditional correlation matrices and graphlet topological features.
• Trained and tested the GCN model to classify subjects into ASD and healthy control (HC) groups.

Main Results:

• The proposed GCN model achieved an average accuracy of 64.27% across the entire ABIDE-I dataset.
• The model demonstrated a highest site-specific accuracy of 75.9%.
• Graphlet features were shown to preserve crucial topological information, aiding in the differentiation between ASD and HC groups.

Conclusions:

• The novel GCN model effectively utilizes rs-fMRI data and graphlet topological features for ASD classification.
• The model's performance is comparable to existing state-of-the-art methods.
• The approach offers potential for improved interpretability in ASD diagnosis using neuroimaging data.