Large-Scale Brain Functional Network Integration for Discrimination of Autism Using a 3-D Deep Learning Model.

Area of Science:

• Neuroimaging
• Machine Learning
• Biomarkers

Background:

• Resting-state functional magnetic resonance imaging (fMRI) is used to construct brain functional networks (BFNs).
• Aberrant functional connectivity in BFNs is observed in autism spectrum disorder (ASD).
• Utilizing BFNs as biomarkers for ASD discrimination remains challenging.

Purpose of the Study:

• To classify autism spectrum disorder (ASD) patients and normal controls (NCs) using BFNs derived from rs-fMRI.
• To develop a deep learning framework for ASD diagnosis based on BFNs.

Main Methods:

• A deep learning framework integrating convolutional neural network (CNN) and channel-wise attention was proposed.
• The model simultaneously modeled intra- and inter-BFN associations.
• Performance was evaluated by investigating individual BFN effects, inter-network connectivity, and comparison with state-of-the-art algorithms.

Main Results:

• The study utilized the ABIDE-I dataset, including 79 ASD patients and 105 NCs.
• The proposed CNN model achieved a mean accuracy of 77.74% in classifying ASD versus NCs.
• The model demonstrated the ability to integrate information from multiple BFNs.

Conclusions:

• The proposed deep learning model effectively integrates information from multiple BFNs to enhance ASD detection accuracy.
• Large-scale BFNs show potential as reliable biomarkers for the diagnosis of ASD.