深度学习在首发精神分裂症患者诊断及脑电图异常分级中的应用

The application of deep learning in the diagnosis and grading of EEG abnormalities in patients with first-episode schizophrenia

投稿时间：2024-07-16 修订日期：2025-06-06

DOI：

中文关键词: 深度学习脑电图精神分裂症诊断异常程度分级长短期记忆模型

英文关键词:Deep learning eEG schizophrenia diagnosis abnormal degree classification long short-term memory model

基金项目:(2008085MH254)

作者	单位	地址
水利利^*	阜阳市第三人民医院	安徽省阜阳市颍州区文兴路2号
刘晨晨	阜阳市第三人民医院
李玉敏	阜阳市第三人民医院

摘要点击次数:

全文下载次数:

中文摘要:

目的探讨深度学习在首发精神分裂症患者诊断及脑电图（EEG）异常分级中的应用。方法选择2020年1月至2023年1月在阜阳市第三人民医院就诊的首发精神分裂症患者130例以及在本院进行体检且行EEG检查的健康人158例为研究对象。基于EEG检查结果建立长短期记忆（LSTM）深度学习网络模型。以准确度、召回率、精确度和F1 score、疾病诊断和EEG异常程度评估用时作为评价指标，并将深度学习模型表现与高年资医师评估结果进行比较。结果 LSTM模型用于精神分裂症诊断准确度为95.4%、召回率为93.3%、精确度为93.3%，F1 score为93.3，与高年资医师比较差异无统计学意义（P>0.05）；LSTM模型用于精神分裂症EEG异常程度评估准确度为83.3%、召回率为83.3%、精确度为84.4%，F1 score为83.9，与高年资医师比较差异无统计学意义（P>0.05）；LSTM模型诊断精神分裂症和疾病异常程度需时均低于高年资医师，差异具有统计学意义（P<0.05）。结论基于EEG的LSTM深度学习模型用于首发精神分裂症诊断和EEG异常程度分级其表现不劣于高年资医师，且耗时大大缩短，适宜在临床中用于首发精神分裂症患者的诊断和EEG异常程度分级辅助评估。

英文摘要:

OBJECTIVE To explore the application of deep learning in the diagnosis and electroencephalogram (EEG) abnormality classification of patients with first-episode schizophrenia.METHODS From January 2020 to January 2023,100 patients with first-episode schizophrenia who were treated in Fuyang Third People 's Hospital and 100 healthy people who underwent physical examination and EEG examination in our hospital were selected as subjects. A long short-term memory ( LSTM ) deep learning network model was established based on EEG examination results. Accuracy, recall, precision and F1 score, disease diagnosis and EEG abnormality evaluation time were used as evaluation indicators, and the performance of the deep learning model was compared with the evaluation results of senior physicians.RESULTS The accuracy of LSTM model in the diagnosis of schizophrenia was 93.5 %, the recall rate was 92.0 %, the accuracy was 94.8 %, and the F1 score was 93.4, which was not significantly different from that of senior physicians ( P > 0.05 ). The accuracy, recall rate, accuracy and F1 score of LSTM model in the evaluation of EEG abnormality in schizophrenia were 94.0 %, 75.7 %, 72.9 % and 74.2, respectively, which were not significantly different from those of senior physicians ( P > 0.05 ). The time required for LSTM model to diagnose schizophrenia and disease abnormalities was lower than that of senior physicians, and the difference was statistically significant ( P < 0.05 ).CONLCLUSIONS The EEG-based LSTM deep learning model is used for the diagnosis of first-episode schizophrenia and the grading of EEG abnormalities. Its performance is not inferior to that of senior doctors, and the time-consuming is greatly shortened. It is suitable for the diagnosis of first-episode schizophrenia patients and the grading of EEG abnormalities in clinical practice.

查看/发表评论下载PDF阅读器

关闭