基于前馈神经网络电子病历辅助诊断系统的研究  

Research on the assistant diagnosis system of EMR based on feedforward neural network

作　　者：汤学民 吴晓云 TANG Xuemin;WU Xiaoyun(IT Department,Shenzhen People's Hospital,Shenzhen 518020,Guangdong Province,China;Statistics Office,Shenzhen People's Hospital,Shenzhen 518020,Guangdong Province,China)

机构地区：[1]深圳市人民医院信息技术部,广东深圳518020 [2]深圳市人民医院统计室,广东深圳518020

出　　处：《中国数字医学》2023年第3期42-48,共7页China Digital Medicine

基　　金：深圳市科技研发资金自由探索项目-基于人工智能的全电子病历自动识别系统(UCYU20180228164327786)。

摘　　要：目的:利用前馈神经网络实现电子病历的辅助诊断。方法:通过定义数据元,按照规则从入院记录、首次病程记录中提取临床信息;然后基于前馈神经网络训练学习模型进行辅助诊断。系统从首次病程记录的诊断依据中提取临床信息作为神经网络的输入,以病历的出院诊断(ICD-10疾病编码)作为输出,利用科室的归档病历训练模型。主管医生录入入院记录后,系统自动解析病历,推荐诊断。结果:基于该模型对心血管内科1000份、神经内科914份电子病历进行测试,诊断符合率分别达到医生诊断水平的86%、76%。结论:基于前馈神经网络的电子病历辅助诊断系统性能整体表现良好,是人工智能辅助诊断的有益探索。Objective To realize diagnostic assistant of electronic medical record(EMR)based on the feedforward neural network(FNN).Methods Clinical information was extracted from inpatient admission and initial progress records by predefining data elements according to rules,and then the learning model was trained for assistant diagnosis based on FNN.The system extracts the clinical information from the diagnosis basis of the initial progress in disease record was the input of the neural network,takes the discharge diagnosis of the medical record(ICD-10 disease code)as the output,and trains the model by utilizing the archived medical record.After the admission record was updated by the attending physician,the system can automatically analyze the medical record and recommend diagnosis.Results The model was used to test 1,000 EMRs of Cardiovascular Department and 914 EMRs of Neurology Department,and the diagnostic compliance rate reached 86%and 76%of the physician's diagnostic level.Conclusion The overall performance of the assistant diagnosis system of EMR based on FNN is good,which serves as a beneficial exploration for AI application in assisted diagnosis.

关 键 词：电子病历 辅助诊断 神经网络 深度学习 数据元 

分 类 号：R319[医药卫生—基础医学]