多模态影像联合血清同型半胱氨酸的机器学习模型对颈动脉斑块患者发生缺血性脑卒中的预测价值-东南大学学报医学版

多模态影像联合血清同型半胱氨酸的机器学习模型对颈动脉斑块患者发生缺血性脑卒中的预测价值

单位：南京市江宁医院, 江苏南京 211100

分类号：R743.3

出版年·卷·期（页码）：2026·45·第四期（581-591）

摘要：

目的:探索多模态影像联合血清标志物的预测模型对颈动脉斑块患者发生缺血性脑卒中的早期预测的可行性及效能。方法:连续纳入于2021年6月至2023年5月本院收治的458例颈动脉斑块患者,所有患者于入组基线(随访起始前1周内)完成颈动脉超声、颈动脉计算机断层扫描血管造影(CTA)检查及血清同型半胱氨酸(homocysteine, Hcy)检测,提取上述基线时点的典型影像学征象与血清 Hcy 数据;对所有入组患者进行为期 24 个月的随访,以随访期间发生缺血性脑卒中为主要研究终点。采用单因素与多因素Logistic回归模型筛选独立预测因子。按8:2随机分为训练集和测试集,分别构建逻辑回归、支持向量机(SVM)及随机森林模型,采用 10 折交叉验证优化模型。采用ROC曲线下面积(AUC)、准确率、灵敏度、特异度及 DeLong 检验等方法比较模型性能,校准曲线、DCA 决策曲线评估模型校准度与临床净获益。结果:排除失访患者后,本研究共纳入426例有效病例,其中发生缺血性脑卒中患者102例。血清Hcy(OR=8.331,95%CI:1.122~61.873,P=0.038)、斑块形态不规则型(OR=5.529,95%CI:1.433~21.337,P=0.013)、颈动脉 2 级狭窄(OR=7.691,95%CI:1.594~37.110,P=0.011)、颈动脉3级狭窄(OR=3.633,95%CI:1.337~9.871,P=0.011)、斑块 2 类(OR=5.930,95%CI:1.116~31.498,P=0.037)、斑块3类(OR=5.585,95%CI:1.117~27.915,P=0.036)为缺血性脑卒中的独立危险因素。随机森林模型的预测效能最优(训练集 AUC=0.945,测试集 AUC=0.913),经 DeLong 检验显著优于逻辑回归(训练集 AUC=0.918,测试集 AUC=0.832)和SVM(训练集 AUC=0.833,测试集 AUC=0.786)。校准曲线提示随机森林预测概率与实际发病风险贴合度最佳,逻辑回归仅低风险区间轻微低估,SVM 低、中风险区间校准偏差显著;DCA 显示随机森林在全临床风险阈值区间净获益较高,临床适用范围更广。结论:多模态影像联合血清Hcy的机器学习模型显著提升缺血性脑卒中预测效能,为早期风险分层提供了客观工具。

Objective: To explore the feasibility and effectiveness of a predictive model combining multimodal imaging and serum biomarkers for early prediction of ischemic stroke in patients with carotid plaque. Methods: A total of 458 patients with carotid plaque who visited our hospital from June 2021 to May 2023 were consecutively included. All patients completed carotid ultrasound, carotid computed tomography angiography(CTA) examination, and serum homocysteine(Hcy) detection at the baseline(within 1 week before the start of follow-up), and the typical imaging signs and serum Hcy data at that baseline were extracted. A 24-month follow-up was conducted for all included patients. The primary research endpoint was the occurrence of ischemic stroke during the follow-up period. Single-factor and multi-factor Logistic regression models were used to screen independent predictors. The data were randomly divided into a training set and a test set in a 8:2 ratio. Logistic regression(LR), support vector machine(SVM), and random forest(RF) models were constructed respectively, and 10-fold cross-validation was used to optimize the models. Model performance was compared using area under the ROC curve(AUC), accuracy, sensitivity, specificity, and DeLong test. Calibration curves and decision curve analysis(DCA) decision curves were used to evaluate the calibration degree and clinical net benefit of the models. Results: After excluding patients lost to follow-up, 426 valid cases were finally included, of whom 102 developed ischemic stroke. The study found that serum Hcy(OR=8.331,95%CI:1.122-61.873, P=0.038), irregular plaque morphology(OR=5.529,95%CI:1.433-21.337, P=0.013), grade 2 carotid artery stenosis(OR=7.691,95%CI:1.594-37.110, P=0.011), grade 3 carotid artery stenosis(OR=3.633,95%CI:1.337-9.871, P=0.011), plaque type 2(OR=5.930,95%CI:1.116-31.498, P=0.037), and plaque type 3(OR=5.585,95%CI:1.117-27.915, P=0.036) were independent risk factors for ischemic stroke. The RF model had the best predictive performance(training set AUC=0.945, test set AUC=0.913), and it was significantly superior to Logistic regression(training set AUC=0.918, test set AUC=0.832) and SVM(training set AUC=0.833, test set AUC=0.786) as verified by the DeLong test. The calibration curve indicates that the RF predicted probability had the best correlation with the actual stroke risk. RF slightly underestimated the low-risk interval, while SVM showed significant calibration bias in both low and medium-risk intervals. DCA demonstrated that the RF model yielded higher net benefit in the entire clinical risk threshold range and had a wider clinical applicability. Conclusion: The machine learning model integrating multimodal imaging and serum Hcy significantly improves the predictive efficacy for ischemic stroke, providing an objective tool for early risk stratification

参考文献：

[1] 乔丽敏,王明慧,赵雅宁,等.基于贝叶斯网络的老年缺血性脑卒中患者衰弱的影响因素分析[J].现代医学,2025,53(4):515-522.
[2] XU Y T,YIN Q Q,GUO Z R.Common carotid artery thrombosis and malignant middle cerebral artery infarction following ovarian hyperstimulation syndrome:a case report[J].World J Clin Cases,2022,10(26):9478-9483.
[3] 刘蕾,何春渝,杨嘉欣,等.脑卒中患者衰弱发生率与不良结局的系统评价[J].现代医学,2023,51(6):823-830.
[4] KUMAGAWA T,NEGISHI H,YAMAMURO S,et al.Recurrentembolic stroke caused by a thrombosed common carotid artery aneurysm:an uncommon and initially indeterminate source of embolism[J].Cureus,2025,17(8):90525.
[5] TANG Y.Heart,brain and neck integrated multimodal ultrasound revealed stroke etiology:carotid sub-web thrombus[J].Med Ultrason,2023,25(3):361.
[6] SABA L,CAU R,MURGIA A,et al.Carotid plaque-RADS a novel stroke risk classification system[J].JACC Cardiovasc Imag,2024,17(1):62-75.
[7] SABA L,CAU R,VERGALLO R,et al.Carotid artery atherosclerosis:mechanisms of instability and clinical implications[J].Eur Heart J,2025,46(10):904-921.
[8] 王增武.《基层冠心病与缺血性脑卒中共患管理专家共识2022》解读[J].中国心血管杂志,2022,27(5):411-416.
[9] HUANG Z,CHENG X Q,LU R R,et al.Incremental prognostic value of carotid plaque-RADS over stenosis degree in relation to stroke risk[J].JACC Cardiovasc Imag,2025,18(1):77-89.
[10] HATHOUT G M,FINK J R,EL-SADEN S M,et al.Sonographic NASCET index:a new Doppler parameter for assessment of internal carotid artery stenosis[J].AJNR Am J Neuroradiol,2005,26(1):68-75.
[11] YANG Y,DONG M.Exploring the role of oxidative stress in carotid atherosclerosis:insights from transcriptomic data and single-cell sequencing combined with machine learning[J].Biol Direct,2025,20(1):15.
[12] JAKUBOWSKI H,WITUCKI Ł.Homocysteine metabolites,endothelial dysfunction,and cardiovascular disease[J].Int J Mol Sci,2025,26(2):746.
[13] MA S,MA F,DING N,et al.Homocysteine promotes the pathogenesis of atherosclerosis through the circ-PIAS1-5/miR-219a-2-3p/TEAD1 axis[J].Adv Sci,2025,12(18):2415563.
[14] SHAH N M,SAROJA A O,NAIK K R,et al.Carotid health in patients with cerebral venous thrombosis[J].Neurol Ind,2022,70(5):1971-1975.
[15] FARHAT W,PARIENTE A,MIJAHED R.Extensive cerebral venous thrombosis secondary to recreational nitrous oxide abuse[J].Cerebrovasc Dis,2022,51(1):114-117.
[16] ZHANG L,XIANG Y,CAO C,et al.Ciliary neurotrophic factor promotes the development of homocysteine-induced vascular endothelial injury through inflammation mediated by the JAK2/STAT3 signaling pathway[J].Exp Cell Res,2024,440(1):114103.
[17] HURJUI L L,TARNICERIU C C,SERBAN D N,et al.Homocysteine attack on vascular endothelium-old and new features[J].Int J Mol Sci,2025,26(13):6298.
[18] HENNING R J,HOH B L.The diagnosis and treatment of asymptomatic and symptomatic patients with carotid artery stenosis[J].Curr Probl Cardiol,2025,50(6):102992.
[19] LI Y,ZHENG S,ZHANG J,et al.Multimodal ultrasound parameters aided carotid plaque risk stratification in patients with asymptomatic carotid stenosis[J].Acta Radiol,2022,63(2):278-286.
[20] CHENG L,ZHENG S,ZHANG J,et al.Multimodal ultrasound-based carotid plaque risk biomarkers predict poor functional outcome in patients with ischemic stroke or TIA[J].BMC Neurol,2023,23(1):13.
[21] 卢然,吴春艳,许海娜,等.正常同型半胱氨酸绝经后中老年女性发生颈动脉斑块风险的阈值研究[J].中国全科医学,2024,27(11):1320-1325.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】

提示：您还未登录，请登录！点此登录