Adaptive Hypergraph Contrastive Learning for ASD Classification Using fMRI Connectome.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Autism Spectrum Disorder (ASD) diagnosis is challenging due to complex symptoms and the need for reliable biomarkers.
• Resting-state functional magnetic resonance imaging (rs-fMRI) and deep learning show promise, but existing methods often overlook higher-order brain network interactions.

Purpose of the Study:

• To propose an Adaptive Hypergraph Contrastive Learning (AHCL) framework to improve Autism Spectrum Disorder (ASD) classification.
• To enhance the capture of higher-order relationships in brain networks for more accurate diagnostic models.
• To improve model interpretability by identifying disease-related brain connections and regions.

Main Methods:

• Developed an Adaptive Hypergraph Contrastive Learning (AHCL) framework utilizing a trainable masking mechanism to create adaptive hyperedges and generate distinct hypergraph views.
• Incorporated low-rank loss to improve the compactness of intra-class samples, addressing limitations in distinguishing negative samples in traditional contrastive learning.
• Jointly optimized view similarity and contrastive loss to ensure semantic consistency while enhancing topological differences for robust feature representation.

Main Results:

• The AHCL framework demonstrated superior performance in ASD classification compared to existing methods.
• The study successfully identified disease-related connections and brain regions, offering valuable insights into ASD.
• The proposed method achieved robust and noise-resistant feature representations with minimal information redundancy.

Conclusions:

• AHCL offers a novel and effective approach for ASD classification by leveraging higher-order brain network interactions.
• The framework provides a more interpretable method for ASD diagnosis, potentially leading to more precise diagnostic strategies.
• This research advances the application of deep learning in neuroimaging for understanding and diagnosing complex neurodevelopmental conditions like ASD.