ASD-SAENet: A Sparse Autoencoder, and Deep-Neural Network Model for Detecting Autism Spectrum Disorder (ASD) Using...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Imaging

Background:

• Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder with current diagnostic limitations.
• Existing clinical methods rely on behavioral observation, lacking insight into underlying neurological mechanisms.
• Need for novel biomarkers for early detection and understanding of ASD brain development.

Purpose of the Study:

• To develop a deep learning model, ASD-SAENet, for classifying individuals with ASD from typical controls using fMRI data.
• To optimize feature extraction for improved classification of ASD-related brain patterns.
• To enhance the accuracy and specificity of ASD detection through advanced computational methods.

Main Methods:

• Developed ASD-SAENet, a deep learning model integrating a sparse autoencoder (SAE) and deep neural network (DNN).
• Utilized SAE for optimized feature extraction from fMRI data.
• Trained the model to minimize both data reconstruction error and classification error.

Main Results:

• ASD-SAENet achieved 70.8% accuracy and 79.1% specificity on the Autism Brain Imaging Data Exchange (ABIDE) dataset.
• Demonstrated superior performance compared to state-of-the-art methods on 12 out of 17 imaging centers.
• Showcased superior generalizability across diverse data acquisition sites and protocols.

Conclusions:

• ASD-SAENet offers a promising computational approach for identifying ASD using fMRI data.
• The model's generalizability suggests potential for widespread clinical application.
• This deep learning framework aids in understanding neurological underpinnings and facilitates early ASD detection.