Classification of Autism Spectrum Disorder Based on Transformer and Graph Convolutional Neural Networks
Autism Spectrum Disorder(ASD)is a neurodevelopmental disorder characterized by severe deficits in social in-teraction,communication,and behavior.Early accurate diagnosis and intervention are crucial for improving patient outcomes,yet early diagnosis is often challenging due to the diversity and complexity of symptoms.In recent years,the combination of neuroimaging techniques such as functional magnetic resonance imaging(fMRI)and deep learning algorithms has provided more accurate means for diagnosing ASD.Functional connectivity network data exhibits complex patterns of both local and global connections,with traditional Graph Convolutional Networks(GCNs)being better at capturing local connectivity patterns,while Transformer models are more suitable for capturing global dependencies.Therefore,effectively integrating these two models to leverage both local and global information is essential for improving the performance of ASD classifica-tion.By establishing brain networks using techniques like functional MRI,combined with GCN and Transformer models,and utilizing data augmentation techniques,we have enhanced the modeling effectiveness for ASD classification.Using the joint model and data augmentation techniques,we achieved an AUC of 87.10 on the test set,significantly higher than the performance of using GCN and Transformer models alone.This study not only provides an efficient approach for the diagnosis of ASD but also holds promise for offering important insights into its underlying neural mechanisms.
autism spectrum disorderfunctional magnetic resonance imagingbrain functional connectivity networksdeep learning
NETL
NSTL
万方数据
维普
