DeepASD: a deep adversarial-regularized graph learning method for ASD diagnosis with multimodal data.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Informatics

Background:

• Autism Spectrum Disorder (ASD) presents complex challenges in diagnosis and understanding its underlying mechanisms.
• Comorbidities in ASD significantly impact mental health, necessitating precise diagnostic tools.
• Single-modality data often fails to capture the full complexity of ASD.

Purpose of the Study:

• To develop an advanced method for Autism Spectrum Disorder (ASD) prediction using multimodal data.
• To enhance the understanding of ASD pathogenesis by incorporating inter-patient relationships.
• To improve the accuracy and comprehensiveness of ASD diagnosis.

Main Methods:

• Proposed DeepASD, an end-to-end trainable regularized graph learning framework.
• Integrated heterogeneous multimodal data and latent inter-patient relationships.
• Employed a multimodal adversarial-regularized encoder for feature representation and graph neural networks for classification.

Main Results:

• DeepASD achieved superior performance compared to eight state-of-the-art methods on the ABIDE dataset.
• Demonstrated significant improvements in accuracy (13.25%), AUC-ROC (7.69%), and specificity (17.10%).
• Effectively leveraged multimodal data and patient relationships for enhanced ASD prediction.

Conclusions:

• DeepASD offers a promising approach for more comprehensive insight into ASD mechanisms.
• The method holds potential for significantly improving ASD diagnosis performance.
• Integrating multimodal data and patient relationships is key to advancing ASD research.