基于FT-LSTM模型的临床事件诊断序列预测研究  

Research on intelligent diagnosis sequence prediction based on FT-LSTM model

作　　者：王露潼 王红[1,2] 宋永强 王倩 Wang Lutong;Wang Hong;Song Yongqiang;Wang Qian(School of Information Science&Engineering,Shandong Normal University,Jinan 250358,China;Shandong Provincial Key Laboratory of Distributed Computing Software,Jinan 250358,China)

机构地区：[1]山东师范大学信息科学与工程学院,济南250358 [2]山东省分布式计算软件新技术重点实验室,济南250358

出　　处：《计算机应用研究》2020年第10期2961-2965,2970,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(61672329,61373149,61472233,61572300,81273704);山东省科技计划资助项目(2014GGX101026);山东省教育科学规划资助项目(ZK1437B010);山东省泰山学者基金资助项目(TSHW201502038,20110819);山东省精品课程资助项目(2012BK294,2013BK399,2013BK402)。

摘　　要：针对不同患者的临床数据信息难以得到合理表示,且事件之间时间间隔不同,从而导致预测困难等问题,提出一种基于词向量表示,并添加时控单元的时控长短期记忆神经网络(FT-LSTM)预测模型。首先通过FastText方法对医学事件进行可解释性的向量表示,用以更有效地捕获富含医学信息的概念关系;然后针对临床数据对时间戳有着强依赖性的现象,在原有LSTM模型的基础上设计时控门,以更好地捕获长短期信息,对事件信息进行建模,从而改善预测表现。在MIMIC-Ⅲ数据集上的实验结果表明,使用FT-LSTM模型预测的召回率、准确率皆高于多种对比模型,证明了该方法的有效性。Aiming at the problems of time series data extraction and medical concept expression difficulty in electronic medical record(EMR)field,this paper proposed a time-controlled long-term and short-term memory neural network(FT-LSTM)prediction model.The model efficiently captured multidimensional features,and further enabled accurate prediction of future clinical events.First,it used the FastText method to interpret vector representations of medical events,more effectively capture rich conceptual relationships in medical information.Then,it designed a time gate based on the original LSTM model to better capture the long-term and short-term information,model the event information to improve the prediction performance.The experimental results on the MIMIC-Ⅲdataset show that,using the FT-LSTM model can predict event results with higher precision,which is of great significance.

关 键 词：电子病历 长短期记忆神经网络 词向量表示 医学概念 可变时间间隔 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]