代价敏感的算法在超声诊断右心衰竭中的应用

The application of cost-sensitive algorithm in the diagnosis of right ventricular failure by echocardiography

目的:使用代价敏感的机器学习方法,依据超声参数对右心衰竭进行诊断,以增进超声在右心衰竭诊断中的价值。方法:选取右心衰竭患者82例为病例组,非右心衰竭者106例为对照组,使用的主要超声参数包括:三尖瓣环收缩期位移(TAPSE)、右室面积变化率(RVFAC)、三尖瓣环收缩期运动速度(Sm)及舒张早期运动速度(Em)、Tei指数、三尖瓣环等容收缩期加速度(IVA)及右室内径、右房内径、左室内径等。通过代价敏感的朴素贝叶斯分析,以上述参数为特征建立诊断模型,对该模型进行交叉验证,比较代价敏感的判别模型与代价均等的模型的诊断效果。结果:较单一参数诊断右心衰竭,机器学习模型能达到更好的正确率(91%),代价敏感的朴素贝叶斯方法可达到更小的总体代价。结论:通过机器学习建立的诊断模型较单一超声参数的效果更佳,较常规方案可降低诊断代价,该方法在右心衰竭的影像筛查中具有潜在实用价值。

Objective: To improve the efficacy of echocardiographic method in evaluating right ventricular function, and to explore the value of cost-sensitive machine learning method model based on ultrasound parameters in the diagnosis of right ventricular failure. Methods: Eighty-two subjects with right heart failure(RHF group) and 106 subjects without right heart failure(non-RHF group) were enrolled in this study. Multi-echocardiographic indices were selected as features, including the right ventricular fractional area change(RVFAC), tricuspid annular plane systolic excursion(TAPSE), tricuspid flow peak velocity at early diastole(E), tricuspid annular peak systolic velocity(Sm) and peak early diastolic velocity(Em), Tei index, isovolumic acceleration(IVA) at tricuspid annular, and the diameter of the left ventricle, right ventricle, right atrium were measured and calculated. Cost-sensitive naive bayes analysis was used to build diagnosis model by using those parameters. The model was tested by 10-fold cross-validation. The diagnostic results of the cost-sensitive model and the classical naive bayes model were compared. Results: Compared with any single parameter, machine learning model can achieve better accuracy in diagnosis of right heart failure(91%). The cost-sensitive naive bayes method achieved a smaller overall cost. Conclusion: The diagnostic model established by machine learning is better than any single ultrasound parameter, which can reduce the diagnostic cost compared with conventional method. This method has potential practical value in screening of right ventricular failure by imaging.