基于机器学习的肝动脉化疗栓塞术后栓塞综合征预测模型构建和比较

Construction of the model based on machine learning algorithm technique used for predicting postembolization syndrome after hepatic artery chemoembolization

目的探讨不同的机器学习技术在预测经肝动脉化疗栓塞术后发生栓塞综合征的效能。方法收集2020年1月至2021年12月在广西医科大学附属肿瘤医院接受肝动脉化疗栓塞术的453例患者临床资料。分别采取逻辑回归、支持向量机、随机森林、梯度提升决策树、极端梯度森林和Lightgbm等6种机器学习技术,构建术后发生栓塞综合征的预测模型。采取五折交叉验证的方式计算不同机器学习算法的准确率、精确率、召回率、F1值和曲线下面积(AUC)。结果经动脉化疗栓塞术后栓塞综合征的发生率为62.47%,基于随机森林算法的平均评估指标值为0.768,优于其他机器学习模型。结论基于随机森林方法建立经动脉化疗栓塞术后栓塞综合征发生的预测模型效能最佳。

Objective To assess the efficacy of different machine learning algorithm techniques in predicting the occurrence of post-embolization syndrome after hepatic artery chemoembolization.Methods The clinical data of 453 patients with primary hepatocellular carcinoma(HCC),who underwent transcatheter hepatic arterial chemoembolization between January 2020 and December 2021 at the Affiliated Tumor Hospital of Guangxi Medical University of China,were retrospectively analyzed.Six machine learning algorithm techniques,including logistic regression(LR),support vector mac(SVM),random forest(RF),gradient boosting decision tree(GBDT),eXtreme Gradient Boosting(XGBoost),and Lightgbm,were separately adopted to construct the predictive model of post-embolization syndrome.The accuracy,precision,recall,F1 value,and area under AUC curve of dfferent machine learning algorithm techniques were calculated by five cross-validation way.Results The incidence of post-embolization syndrome after hepatic artery chemoembolization was 62.47%.The average evaluation index value of the algorithm based on random forest was 0.768,which was better than that of all the other machine learning algorithm techniques.Conclusion The prediction model established on basis of the random forest machine learning algorithm technique carries the optimal predictive efficacy for the occurrence of post-embolization syndrome after hepatic artery chemoembolization.