Population Graph-Based Multi-Model Ensemble Method for Diagnosing Autism Spectrum Disorder.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Machine learning and brain imaging advancements have improved Autism Spectrum Disorder (ASD) diagnosis.
• Graph neural networks (GNNs) show promise for analyzing population structures in brain imaging data.
• Defining appropriate population graphs for GNNs in clinical diagnosis remains a challenge.

Purpose of the Study:

• To develop a robust population graph-based multi-model ensemble for enhanced ASD diagnosis.
• To address the challenge of selecting the optimal population graph structure for GNNs.
• To improve the accuracy of differentiating between healthy subjects and individuals with ASD.

Main Methods:

• Constructed multiple population graphs using diverse imaging and phenotypic features.
• Evaluated graph properties using Graph Signal Processing tools.
• Developed a neural network architecture to ensemble multiple graph-based models.

Main Results:

• The proposed ensemble model demonstrated superior performance compared to state-of-the-art methods.
• The model achieved significant improvements in predicting ASD diagnosis on the ABIDE dataset.
• The ensemble approach proved effective regardless of the specific underlying graph structure used.

Conclusions:

• The population graph-based multi-model ensemble offers a powerful and flexible approach for ASD diagnosis.
• This method enhances diagnostic accuracy by leveraging diverse data features and ensemble learning.
• The findings highlight the potential of advanced graph-based machine learning in clinical neuroscience.