基于机器学习的疾病诊断模型研究  

Research on Disease Diagnosis Model Based on Machine Learning

作　　者：张千[1] 方丽华 王庆玮 孙晓 梁鸿[1] 张万义[2] ZHANG Qian;FANG Lihua;WANG Qingwei;SUN Xiao;LIANG Hong;ZHANG Wanyi(Department of Computer Science and Communication Engineering,China University of Petroleum,Qingdao 266580;Shengli Petroleum Administration Shengli Hospital,Dongying 257091)

机构地区：[1]中国石油大学(华东)计算机与通信工程学院,青岛266580 [2]中石化胜利石油管理局胜利医院老年病医院,东营257091

出　　处：《计算机与数字工程》2020年第7期1705-1709,1714,共6页Computer & Digital Engineering

基　　金：中央高校基本科研业务专项基金项目“面向糖尿病大数据的自动化机器学习”(编号:18CX02019A)资助。

摘　　要：糖尿病视网膜病变作为糖尿病的严重并发症之一,患者日趋年轻化。因此糖尿病的筛查预防工作尤为重要。医生诊断主要依赖其临床经验,传统的诊断方式在一定程度上会被医生自身积累的实际经验所束缚,从而导致最终诊断的结果出现较大的误差。并且由于医院医疗数据的复杂化、高维化以及特征差异的缩小化,传统的医疗疾病诊断模型所获得的诊断准确率已远远低于当今社会对医疗疾病诊断的高准确率要求。论文将机器学习算法应用到糖尿病并发症的检测和诊断中,以2型糖尿病视网膜病症为例,通过对大量样本建模、训练和预测,找出对疾病影响最为关键的指标。实验采用的数据集来自于中国人民解放军总医院的电子医疗记录。通过对301医院数据集的特征变量进行逐步回归分析,建立了基于逻辑回归的疾病特征分析模型,实验结果输出了糖尿病视网膜病变的相关重要特征排序,并取得了较高的训练准确率和测试准确率。实验分析显示,糖化血红蛋白浓度与慢性肾病是2型糖尿病视网膜病变的主要影响因素,为医生提供了可信的诊断参考。Diabetic retinopathy is one of the serious complications of diabetes,and patients are getting younger.Therefore,screening and prevention of diabetes are particularly important.Doctor’s diagnosis mainly depends on his clinical experience.To a certain extent,the traditional diagnosis method will be bound by the accumulated practical experience of the doctor himself,which will lead to large errors in the final diagnosis results.Because of the complexity,high dimensionality and narrowing of the differences in characteristics of hospital medical data,the diagnostic accuracy of traditional medical disease diagnosis model is far lower than the high accuracy requirement of medical disease diagnosis in today’s society.In this paper,machine learning algorithm is applied to the detection and diagnosis of diabetic complications.Taking type 2 diabetic retinopathy as an example,the most critical indicators of disease impact are found by modeling,training and prediction of a large number of samples.The data set used in the experiment is from the electronic medical records of General Hospital of PLA.Through stepwise regression analysis of characteristic variables of 301 Hospital Data set,a disease characteristic analysis model based on logistic regression is established.The experimental results output the ranking of relevant important features of diabetic retinopathy,and achieve higher training accuracy and test accuracy.The experimental analysis showes that glycosylated hemoglobin is the main factor affecting diabetic retinopathy.Concentration and chronic nephropathy are the main influencing factors of type 2 diabetic retinopathy,providing reliable diagnostic reference for doctors.

关 键 词：糖尿病视网膜病变 机器学习 逻辑回归 特征提取 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]