机器学习对预测颈内动脉非急性闭塞患者血管内再通术成功的潜在价值

Predicting successful endovascular recanalization for non-acute occlusion of internal carotid artery:potential value of machine learning

目的使用机器学习算法预测模型探究影响颈内动脉非急性闭塞(NAOICA)血管内再通治疗成功的因素,并比较其与传统预测模型的预测效能。方法收集2016年1月至2021年12月因NAOICA在北京医院神经外科接受血管内再通术的患者的临床数据。采用包括正则化的Logistic回归(RLR)、支持向量机(SVM)、决策树(DT)、随机森林(RF)和极限梯度提升(XGBoost)在内的机器学习算法构建预测血管内再通技术成功的模型,评估其受试者工作特征曲线(ROC)的曲线下面积(AUC),并与基于变量筛选的传统Logistic回归模型比较其预测价值。结果共纳入NAOICA闭塞再通术患者69例(73例次),其中男性62例,年龄为(64.8±8.8)岁,技术成功率为67.1%。在预测血管内再通成功与否方面,表现最差的机器学习模型(DT)仍取得了与标准Logistic回归模型近似的效能(平均AUC分别为0.66和0.65),其他机器学习模型的表现均明显优于标准模型(平均AUC为0.74~0.84)。在大部分机器学习模型中,前交通动脉的代偿、亚急性闭塞、自发再通征象和闭塞段远端显影管腔部位是对模型贡献较大的重要预测变量。结论在NAOICA患者中,机器学习模型对血管内再通技术成功的预测效能要优于标准Logistic回归模型。

Objective To explore the factors affecting the success of endovascular revascularization treatment for non-acute occlusion of the internal carotid artery(NAOICA)by machine learning predictive models,and to compare their predictive value with that of classic predictive models.MethodsPatients who underwent endovascular revascularization for NAOICA at Department of Neurosurgery of Beijing Hospital from Jan.2016 to Dec.2021 were included.Machine learning algorithms including regularized Logistic regression,support vector machine,decision tree,random forest,and extreme gradient boosting were used to build models to predict the success of endovascular recanalization technology.Area under the ROC curves(AUC)and difference in mean AUC between the models were assessed.ResultsA total of 73 consecutive NAOICA recanalization attempts were performed in 69 patients[62 men;mean age of(64.8±8.8)years] with an overall technical success(primary outcome)rate of 67.1%.In terms of predicting endovascular recanalization success,even the worst prediction model based on machine learning had equivalent predictive ability to the standard Logistic regression model(mean AUC of 0.66 and 0.65,respectively),and all other machine learning models significantly outperformed the standard Logistic regression model(mean AUC of 0.74-0.84).Collateral blood flow from the anterior communicating artery,subacute occlusion,spontaneous recanalization,and segment of distal carotid visibility appeared as the crucial variables in most machine learning models.Conclusion In patients with NAOICA,machine learning algorithms can outperform classic Logistic regression model in predicting the technical success of endovascularrevascularization.