基于深度学习的院内感染诊疗决策支持系统

Hospital infection diagnosis and treatment decision support system based on deep learning

目的为了提高院内感染诊疗的自动化程度和正确性,降低人工检测判定的代价。方法利用LSTM对电子病历文本中的院感症状短语进行识别获取,结合院感其它结构化字段数据组成62个院感特征作为深度神经网络模型的输入。采用2382个匿名院内感染患者的特征数据进行模型训练、优化及保存。最后以医生确诊结果为标准,对随机抽取430个检验出病原菌的患者进行院感测试分类验证。结果院感相关症状短语识别F值高于92%,AUC高于85%。院内感染诊疗决策支持系统正确率达98.4%。结论院内感染诊疗决策支持系统可以智能辅助医生更快更准确地对院内感染患者进行诊断。

Objective To improve the performance and automation of hospital infection diagnosis and treatment,and reduce the cost of manual detection to determine hospital infection.Methods LSTM was used to recognize and acquire the phrases of infection symptoms in the electronic medical record text,and combined with other structured data to construct 62 hospital sense features.Then,the deep neural network was used to train the model data of 2382 anonymous nosocomial infection patients,and 430 test patients were tested based on the doctor's diagnosis results.Results The F value of phrase recognition related to hospital sense was higher than 92%,and the AUC was higher than 85%.The accuracy rate of decision support system for nosocomial infection diagnosis and treatment reached 98.4%.Conclusion Nosocomial infection diagnosis and treatment decision support system can intelligently assist doctors to diagnose nosocomial infection patients more quickly and accurately.