Classification of ASD based on fMRI data with deep learning.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Autism spectrum disorder (ASD) is a neurodevelopmental condition impacting social interaction.
• Abnormal functional connections (FCs) in the brain are observed in individuals with ASD.
• Identifying these FCs offers a potential biological basis for ASD diagnosis.

Purpose of the Study:

• To develop and validate a novel method for classifying autism spectrum disorder (ASD).
• To enhance the accuracy of ASD diagnosis using neuroimaging data and machine learning.

Main Methods:

• A combined deep feature selection (DFS) and graph convolutional network (GCN) approach was proposed.
• DFS was employed to select critical functional connection (FC) features from brain imaging data.
• A GCN model utilized selected FCs and phenotypic information for ASD classification.

Main Results:

• The proposed DFS method effectively selected relevant FC features, significantly improving GCN classifier performance.
• The method achieved state-of-the-art accuracy of 79.5% and an AUC of 0.85 on the ABIDE dataset.
• Analysis revealed widespread FCs and a higher prevalence of weak connections in the ASD group.

Conclusions:

• The combined DFS and GCN method demonstrates superior performance in classifying ASD.
• The identified critical FCs provide insights into the neurobiological underpinnings of ASD.
• This approach holds promise for improving the objective diagnosis of autism spectrum disorder.