基于机器学习分类判断算法构建心力衰竭疾病分期模型

背景：早期发现心力衰竭及心力衰竭分期的正确诊断是获得良好治疗效果的基础，但由于缺乏简单有效的心力衰竭分期诊断模型，使临床诊断心力衰竭较为困难，导致心力衰竭的确诊率和控制率都比较低。目的：采用基于机器学习的分类判断算法，建立心力衰竭分期模型，提高心力衰竭诊断和分期准确度。方法：选择心力衰竭患者和健康体检者共194例，以美国心脏病协会和美国心脏协会分期为依据，采集与心力衰竭分期密切相关的特异性临床特征参数指标，对参数进行筛选，参考专家的临床确诊结果，采用Adaboost模型和SVM模型训练心力衰竭诊断和分期模型，获得诊断模型。结果与结论：采用每搏输出量、心输出量、左室射血分数、左房内径、左室收缩末内径、N末端B型利钠肽原、心率变异性等指标，Adaboost模型的灵敏度很高，达到了100%，而特异性为94.4%；同时SVM模型也具有良好的灵敏度和特异性，分别为86.5%和89.4%。Adaboost分类模型可准确诊断心力衰竭症状，准确率达到89.36%。而在进一步确诊为心力衰竭的基础上，SVM 分类模型能对心力衰竭的严重程度进行有效分期，B分期和C分期的准确率分别达到了86.49%和81.48%。说明结合Adaboost和SVM两种机器学习模型，能为心力衰竭的诊断及分期提供较准确的模型。

A heart failure staging model based on machine learning classification algorithms

BACKGROUND：Early detection and accurate staging diagnosis of heart failure are the basis of good clinical therapy efficacy. Due to lack of simple and effective staging model for the diagnosis of heart failure, it is difficult to diagnose heart failure in clinics, leading to poor control of heart failure. OBJECTIVE：To establish the disease staging model based on Adaboost and SVM for heart failure, and improve the accuracy of diagnosis and staging of heart failure. METHODS：A total of 194 cases were roled into this study, including heart failure patients and healthy physical examination persons. According to the stage standards formulated by American Colege of Cardiology and American Heart Association, specific clinical feature parameters closely related to heart failure were colected and selected. Based on clinical diagnosis results and using Adaboost model and SVM model, we trained the models for heart failure diagnosis and staging, thus obtaining diagnosis model. RESULTS AND CONCLUSION： The parameters included stroke volume, cardiac output, left ventricular ejection fraction, left atrial diameter, left ventricular internal diameter at end-systole, N-terminal pro-brain natriuretic peptide and heart rate variability. As for the Adaboost model, its sensitivity and specificity was 100% and 94.4%, respectively. At the same time the SVM model had good sensitivity and specificity, 86.5% and 89.4% respectively. Adaboost classification model can be accurate in the diagnosis of heart failure symptoms, the accuracy reached 89.36%. On the basis of the diagnosis of heart failure, the SVM classification model is effective in staging the severity of heart failure, staging accuracy for staging B and C was 86.49% and 81.48%, respectively. The findings indicate that, combining Adaboost and SVM machine learning models could provide an accurate diagnosis and staging model for heart failure.