Deep Canonical Correlation Fusion Algorithm Based on Denoising Autoencoder for ASD Diagnosis and Pathogenic Brain...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by atypical neural activity.
• Early intervention is critical for managing ASD progression.
• Current diagnostic approaches often utilize structural (sMRI) and resting-state functional MRI (rs-fMRI), but autoencoder applications remain underexplored.

Purpose of the Study:

• To introduce and evaluate a novel framework, Deep Canonical Correlation Fusion algorithm based on Denoising Autoencoder (DCCF-DAE), for Autism Spectrum Disorder classification.
• To explore the efficacy of autoencoders in disease classification using high-dimensional neuroimaging data.
• To integrate functional and structural MRI data for enhanced ASD diagnosis and identification of critical brain regions.

Main Methods:

• Development of the DCCF-DAE framework utilizing advanced autoencoders for efficient feature extraction from multimodal neuroimaging data.
• Application of the Deep Canonical Correlation Fusion (DCCF) model to integrate extracted features.
• Utilizing fused features for Autism Spectrum Disorder classification and identification of Regions of Interest (ROIs).

Main Results:

• The DCCF-DAE framework demonstrated effective handling of high-dimensional data.
• Integration of functional and structural data through DCCF improved ASD diagnostic accuracy.
• The proposed framework achieved outstanding performance in ASD diagnosis when compared against other methods using the ABIDE database.

Conclusions:

• The DCCF-DAE framework shows significant potential as a tool for accurate early diagnosis of Autism Spectrum Disorder.
• The method effectively integrates multimodal neuroimaging data, offering insights into ASD mechanisms.
• This approach highlights the value of advanced autoencoders and fusion techniques in neurodevelopmental disorder research.