深度电子病历分析研究综述
2018-07-28蒋友好
蒋友好
摘要:伴随医院信息化建设,大量的电子病历数据得以保存,但如何分析和利用这些数据成为医疗健康领域一个重要的研究课题。深度电子病历分析以深度学习技术为基础,通过特征自学习,避免了在数据预处理和特征工程上耗费大量时间,而且还能有效捕获数据间的未知关系,提高算法性能。本文首先概述了5类常用的深度学习模型及其变体,其次详细分析了这5类模型在电子病历分析上的应用情况,最后从数据异质性、公开数据集和模型可解释性三个方面对这一领域当前的机遇和挑战做了总结。
关键词:电子病历;深度学习;卷积神经网络;循环神经网络
中图分类号:TP311 文献标识码:A 文章编号:1009-3044(2018)15-0301-04
An Overview of Research on Deep Electronic Health Record Analysis
JIANG You-hao1,2
(1.Department of Control Science and Engineering School of Electronics and Information Engineering Tongji University,Shanghai 201804,China;2.Shanghai Putuo District Central Hospital, Shanghai 200062, China)
Abstract:With the development of hospital informatization, the vast amounts of raw electronic health records have been saved. But how to analyze and utilize these data becomes an important research topic in the field of healthcare. Based on deep learning technologies, deep electronic health record analysis models not only can learn features directly from the data itself, avoiding the cost of time on data preprocessing and feature engineering, but also can gain high performance by effectively capturing latent relationships between data. In this paper, five commonly used deep learning models and their variants are firstly discussed, and then analyzes some electronic health record analysis applications in detail. Finally, we summarize the current opportunities and challenges from three aspects: data heterogeneity, public datasets and model interpretability.
Key words: Electronic Health Record (HER); Deep Learning; Convolutional Neural Networks (CNN); Recurrent Neural Network (RNN)
引言
隨着医院信息化建设不断深入,电子病历(Electronic Health Record,EHR)系统在临床诊疗过程中得到了广泛应用,也因此产生了大量的医疗数据。这些数据记录了患者所有的诊治历史,包括人口统计学信息、诊断、实验室检验结果、放射影像、处方、临床记录等[1]。之前,电子病历数据主要被用来提高临床诊疗效率,并方便医院管理。但随着大数据和人工智能技术的飞速发展,许多研究者认为电子病历数据对提高医护质量、保障患者安全、降低治疗费用等方面大有裨益[2-4]。
尽管电子病历数据越来越容易获取,但其异质的特性给分析带来了巨大的挑战。从表示形式上来看,电子病历数据有以下5种类型:1)数值型,如年龄、体重;2)时间日期型,如入院时间、处方开立日期;……
