整合能谱CT定量参数的机器学习模型对症状性颈动脉斑块识别的应用研究

Application Research of A Machine Learning Model Integrating Quantitative Parameters of Energy Spectrum CT in the Identification of Symptomatic Carotid Atherosclerotic Plaque

目的评估整合能谱CT定量参数、传统CT特征及临床参数的机器学习(ML)模型对症状性颈动脉斑块的识别能力。方法回顾性分析171例行头颈部CTA检查发现颈动脉斑块的患者资料。由两位观察者独立观察、测量并计算3个能谱CT定量参数以及8个传统CT特征,并收集患者的临床资料。将研究人群分为症状组(n=104)和无症状组(n=67),并进行组间差异性分析以及传统的多因素Logistic回归分析。使用能谱CT定量参数、传统CT特征及临床参数,并基于XGboost算法构建症状性斑块的预测模型。绘制受试者工作特征(ROC)曲线,根据曲线下面积(AUC)计算模型的准确率、F1分数等指标评估模型的性能。结果包括前十个重要变量的XGboost整合模型[AUC 0.946(95%CI 0.890~1.000)]的AUC值显著高于能谱特征模型[AUC 0.778(95%CI 0.638~0.918),P=0.016]、传统CT特征模型[AUC 0.702(95%CI 0.548~0.856),P=0.001]及临床参数模型[AUC 0.754(95%CI 0.608~0.900),P=0.009]。此外,相较于传统的多因素Logistic回归分析[AUC 0.845(95%CI 0.785~0.904)],XGboost整合模型亦具有较高的症状性斑块识别能力。结论整合能谱CT定量参数、传统CT特征及临床参数的ML模型具有较好的症状性颈动脉斑块的识别能力。

Objective To evaluate the ability of machine learning(ML)models integrating quantitative parameters of energy spectrum CT,traditional CT features and clinical parameters to identify symptomatic carotid plaque.Methods The data of 171 patients with carotid artery plaque found by head and neck CTA were analyzed retrospectively.Three quantita-tive parameters of energy spectrum CT and eight traditional CT features were independently observed,measured and calcu-lated by two observers,and clinical data of patients were collected.The study population was divided into symptomatic group(n=104)and asymptomatic group(n=67),and the inter group difference analysis and traditional multivariate lo-gistic regression analysis were conducted.Using the quantitative parameters of energy spectrum CT,traditional CT features and clinical parameters,and based on XGboost algorithm,a prediction model of symptomatic plaque was constructed.Draw the ROC curve,and evaluate the performance of the model according to the area under the curve(AUC)and calculate the accuracy rate,F1 score and other indicators of the model.Results The AUC value of XGboost integrated model[AUC 0.946(95%CI 0.890-1.000)]including the first ten important variables was significantly higher than that of energy spectrum characteristic model[AUC 0.778(95%CI 0.638-0.918),P=0.016],traditional CT characteristic model[AUC 0.702(95%CI 0.548-0.856),P=0.001]and clinical parameter model[AUC 0.754(95%CI 0.608-0.900),P=0.009].In addition,compared with the traditional multivariate logistic regression analysis[AUC 0.845(95%CI 0.785-0.904)],the XGboost integrated model also has a higher ability to identify symptomatic plaque.Conclusion ML model integrating quantitative parameters of energy spectrum CT,traditional CT features and clinical parameters has bet-ter recognition ability of symptomatic carotid plaque.