基于图注意力网络的脑电信号分类研究  

A Study of EEG Signal Classification Based on Graph Attention Network

作　　者：朱耿 林萍 徐信毅[1,2] 施浩洋 郝丽俊 李晓欧 李斌 ZHU Geng;LIN Ping;XU Xinyi;SHI Haoyang;HAO Lijun;LI Xiaoou;LI Bin(College of Medical Instruments,Shanghai University of Medicine&Health Sciences,Shanghai 201318,China;College of Health Science and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Shanghai Yangpu Mental Health Center,Shanghai 200093,China)

机构地区：[1]上海健康医学院医疗器械学院,上海201318 [2]上海理工大学健康与科学工程学院,上海200093 [3]上海市杨浦区精神卫生中心,上海200093

出　　处：《电子器件》2024年第5期1403-1407,共5页Chinese Journal of Electron Devices

基　　金：上海市科委地方院校能力建设项目(22010502400);上海健康医学院精神卫生临床研究中心项目(20MC2020005);上海市杨浦区科学技术委员会、卫生健康委员会科研项目(YPM202114)。

摘　　要：脑电信号已被广泛应用于精神疾病诊疗领域,能够反映出大脑的活动异常。利用人工智能模型在精神疾病的预测、诊断、干预等方面有优于传统模式的潜力。结合图论分析方法和图注意力网络建立了一种人工智能模型,实现了精神分裂症患者脑电信号的自动识别。以强化学习任务时和工作记忆任务时获得的脑电信号为样本,利用相位滞后指数构建脑功能连接图,通过图注意网络模型,实现对精神分裂症患者和健康人的脑电信号的自动分类。处理强化学习任务数据,该算法的识别准确率为87.5%,特异性为72.7%;处理工作记忆任务数据,该算法的准确率为70.0%,特异性为85.7%。相较于对比方法,所提算法有显著改善。本文所提的脑电信号处理方法不依赖于人工特征的提取,并提高了分类准确率,有望为精神疾病的诊断提供可靠依据,具有良好的临床应用前景。EEG signals have become crucial in the field of psychiatric diagnosis and treatment as they can reflect abnormal brain activity.Artificial intelligence models hold the potential to outperform traditional methods in predicting,diagnosing,and intervening in mental disorders.Graph theory analysis method and graph attention network are combined to build an artificial intelligence model to achieve automatic recognition of EEG signals of schizophrenia patients.By utilizing event-related potential datasets obtained during a reinforcement learning task and a working memory task,a brain functional connectivity matrix using the phase lag index is constructed.A graph attention network model is employed to achieve automatic classification of EEG signals from both schizophrenic patients and healthy individuals.The algorithm achieves a recognition accuracy value of 87.5%and a specificity value of 72.7%when dealing with data from a reinforcement learning task,and an accuracy value of 70.0%and a specificity value of 85.7%when dealing with data from a working memory task.Compared with other similar methods,the algorithm proposed shows significant improvement.The EEG signal processing method proposed does not rely on manual feature extraction and improves the classification accuracy,which is expected to provide a reliable basis for the diagnosis of mental diseases and has a good prospect for clinical application.

关 键 词：图注意力网络 脑电信号 脑功能连接 精神分裂症 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]