A heterogeneous graph convolutional attention network method for classification of autism spectrum disorder

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition.
• Functional magnetic resonance imaging (fMRI) and deep learning show promise for ASD classification.
• Existing graph neural networks struggle to fully integrate subject phenotype data.

Purpose of the Study:

• To develop a novel deep learning model for improved ASD classification.
• To leverage both fMRI and phenotypic data for enhanced diagnostic accuracy.
• To address limitations of homogeneous graph structures in current methods.

Main Methods:

• Proposed a heterogeneous graph convolutional attention network (HCAN) model.
• Constructed a heterogeneous population graph integrating fMRI and phenotypic data.
• Employed a multilayer HCAN for feature extraction followed by an MLP classifier.

Main Results:

• Achieved a classification accuracy of 82.9% on the ABIDE I dataset (871 subjects).
• Demonstrated superior performance compared to existing methods on the same dataset.
• Validated the model's effectiveness in classifying ASD using fMRI data.

Conclusions:

• The HCAN model effectively integrates heterogeneous graph convolutional networks and attention mechanisms.
• Subject phenotype features are fully utilized, improving classification performance.
• The proposed method shows significant potential for diagnosing brain functional disorders using fMRI.