基于机器学习的左室舒张功能不全评估模型研究

The Research of Left Ventricular Diastolic Dysfunction Evaluation Model Based on Machine Learning Algorithm

目的:拟利用机器学习算法建立左室舒张功能不全的评估模型,为临床早期筛查左室舒张功能不全提供更加廉价、便捷的解决方案。方法:收集2017年1月至2018年12月就诊于河南省人民医院门诊或住院部处于心力衰竭Stag A或Stag B阶段的病人,共2 347例,分为左室舒张功能不全组(1 493例)、左室舒张功能正常组(854例),记录患者的心电图各项参数、年龄、性别、合并症等特征,对特征归一化预处理后,分别使用4种方法来构建左室舒张功能不全的评估模型,分别是:K最近邻、随机森林、前向神经网络、支持向量机,然后通过3折交叉验证的方式,测试和比较模型的性能。结果:基于支持向量机算法的评估模型显示出最好的性能,ROC曲线下面积为0.92(95%置信区间:90%~93%),用于评估左室舒张功能不全的敏感性和特异性分别为91%和78%。结论:基于机器学习算法的预测模型可作为评估左室舒张功能不全的早期筛查工具,进而能够早期做出干预,改善患者的预后。

Objective: This study intends to use the characteristics of electrocardiogram, age, and cardiovascular comorbidity to establish an evaluation model of left ventricular diastolic dysfunction based on machine learning algorithm, in order to provide a cheaper and more convenient solution for early screening of left ventricular diastolic dysfunction. Methods: Patients who were admitted to the He’nan Provincial People’s Hospital outpatient or inpatient department at the Stag A or Stag B of heart failure from January 2017 to December 2018 were collected, and finally 2 347 patients were included and they were divided into left ventricular diastolic dysfunction group(1 493 cases), normal diastolic function group(854 cases), they all have detailed records such as electrocardiogram, age, gender, comorbidities and other characteristics, after the feature normalization pretreatment stage, we used 4 methods to construct left ventricular diastolic dysfunction evaluation models, the 4 methods are K-Nearest Neighbor(KNN), Random Forest(RF), MultiLayer Perceptron(MLP) and Support Vector Machine(SVM). The performance of the model is then tested and compared by a 3-fold cross-validation. Results: Support vector machine algorithm evaluation model showed the best performance with an area under the ROC curve(AUC) of 0.92(95% confidence interval: 90% to 93%), and the sensitivity and specificity for assessing left ventricular diastolic dysfunction were 91% and 78% respectively. Conclusion: The predictive model based on machine learning algorithm can be used as a powerful tool for assessing left ventricular diastolic dysfunction. It can be used as an early screening method to early intervention and improve patient prognosis.