卷积神经网络在精神分裂症诊断中的应用研究  

Application of convolutional neural networks in the diagnosis of schizophrenia

作　　者：刘津[1] 贺勇 王久菊[1] 权文香[1] 田菊[1] 冯超南 于淏岿 南彩 纪俊 Liu Jin;He Yong;Wang Jiuju;Quan Wenxiang;Tian Ju;Feng Chaonan;Yu Haokui;Nan Cai;Ji Jun(Diagnosis;Information Department,Peking University Six Hospital,Beijing 100191,China;Beijing Wanling Pangu Technology Co.,Ltd.,Beijing 100089,China;College of Computer Science&Technology,Qingdao University,Qingdao 266000,China;Beijing Wanling Pangu Technology Co.,Ltd.,Beijing 100089,China)

机构地区：[1]北京大学第六医院转化医学科,100191 [2]北京大学第六医院信息科 [3]北京万灵盘古科技有限公司,100089 [4]青岛大学计算机科学技术学院,266000 [5]武汉大学人民医院精神科,430060

出　　处：《中华行为医学与脑科学杂志》2019年第7期622-626,共5页Chinese Journal of Behavioral Medicine and Brain Science

基　　金：国家自然科学基金项目(61503208);北京市科委重大项目三期课题(D171100007017002);山东省自然科学基金项目(ZR2015PF002).

摘　　要：目的探讨卷积神经网络用于精神分裂症辅助诊断的方案,并评价其效果。方法利用卷积神经网络从138个正常人和183个精神分裂症患者的导联数据中训练得到模型,采用20-折交叉验证的方式对模型进行验证。结果使用卷积神经网络训练模型得到的精神分裂症预测的真阳性率为0.749,假阳性率为0.275,准确率为0.738。结论该模式能实现对精神分裂症患者较强的诊断能力。因此,卷积神经网络用于精神分裂症的诊断,将成为以后一个重要的研究方向。Objective To explore the program of convolutional neural networks for the diagnosis of schizophrenia and evaluate its effects. Methods Using the convolutional neural network, the training model was trained in the lead data of 138 normal people and 183 schizophrenic patients, and the model was validated by 20-fold cross-validation. Results The true positive rate of schizophrenia prediction using the convolutional neural network training model was 0.749, the false positive rate was 0.275, and the accuracy was 0.738. Conclusion This model can achieve a strong diagnostic ability for patients with schizophrenia.Therefore, convolutional neural network for the diagnosis of schizophrenia will become an important research direction in the future.

关 键 词：精神分裂症 卷积神经网络 脑电波 诊断 

分 类 号：R749.3[医药卫生—神经病学与精神病学]