A Hierarchical Feature Extraction and Multimodal Deep Feature Integration-Based Model for Autism Spectrum Disorder...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Autism Spectrum Disorder (ASD) diagnosis is challenging due to subject heterogeneity and limitations in current predictive models.
• Existing methods often fail to optimally integrate resting-state functional magnetic resonance imaging (rs-fMRI) and non-imaging data for accurate ASD prediction.

Purpose of the Study:

• To develop a novel framework, HE-MF, for improved Autism Spectrum Disorder (ASD) prediction.
• To enhance classification accuracy by effectively utilizing both rs-fMRI and non-imaging information.

Main Methods:

• The HE-MF framework features a Hierarchical Feature Extraction Module for multi-level feature extraction and a Multimodal Deep Feature Integration Module for fusing rs-fMRI and non-imaging data.
• An attention mechanism is employed for dynamic weight allocation during deep feature fusion.
• The model was evaluated on the ABIDE and ADNI datasets.

Main Results:

• The HE-MF model achieved 95.17% accuracy in ASD identification on the ABIDE dataset.
• Demonstrated superior performance compared to existing state-of-the-art methods.
• Validated generalization capabilities on the ADNI dataset.

Conclusions:

• The HE-MF framework offers a highly effective and superior approach for ASD prediction.
• The model's ability to integrate multimodal data and handle heterogeneity significantly improves predictive performance.
• HE-MF shows strong potential for clinical application in neurodevelopmental disorder identification.