【作者】 李骞；

【导师】 宋燕；

【作者基本信息】 郑州大学 ， 外科学（普外科）（专业学位）， 2024， 硕士

【摘要】 研究背景动脉粥样硬化性颈动脉狭窄(Carotid artery stenosis,CAS)是一种常见的慢性、炎性疾病,其在临床上是导致患者发生脑卒中的重要原因之一。CAS的发生通常与动脉内壁的斑块形成和动脉管腔的狭窄有关。这种斑块主要由胆固醇、钙盐、纤维组织和炎性细胞组成,形成于颈动脉的内膜下层。随着斑块的不断积累和炎症反应的加剧,颈动脉管腔逐渐变窄,从而影响了血液的正常流动。如果斑块不稳定或者发生血栓形成,可能会导致血流受阻,从而引发脑卒中等严重并发症。根据临床表现可将其分为症状性和无症状性颈动脉狭窄(Asymptomatic carotid artery stenosis,AC AS)。症状性重度狭窄患者常表现为严重的头晕、头痛、感觉运动功能障碍等,严重影响生活质量且卒中风险较高,所以一经检出即需要尽快进行手术干预;而无症状重度颈动脉狭窄患者的临床表现较为隐匿,仅在影像学上显示为颈动脉管腔狭窄,目前国际上针对这类患者的诊治指南尚未达成共识,导致一些高卒中风险的ACAS患者未及时接受正确的手术干预,而部分低卒中风险的患者接受了不必要的侵入性治疗,故如何识别出ACAS患者中的高卒中风险人群从而针对性的选择合适的治疗方式是亟待解决的问题。本文提出了一种基于颈动脉斑块影像组学联合颈动脉血管周围脂肪衰减值(perivascular fat attenuation index,PFAI)及传统颈动脉CTA特征的风险预测模型。回顾性收集2020年6月—2023年6月经郑州大学第一附属医院临床诊断的颈动脉重度狭窄患者的临床资料及颈部CTA信息,从临床资料、传统CT A特征及颈动脉斑块影像组学特征中,以相关统计学方法筛选出缺血性脑卒中的独立危险因素,并分别建立逻辑回归模型,后验证对比三种预测模型得出最优预测模型,辅助临床医生识别高危患者,在对无症状颈动脉重度狭窄患者制定诊疗计划时提供参考价值。研究目的1.筛选出无症状性颈动脉重度狭窄患者发生脑卒中的独立危险因素。2.分别建立传统CTA模型、斑块影像组学标签模型及联合影像组学模型并验证对比得出最优模型,帮助临床医生做诊疗决策。研究方法回顾性收集2020年6月至2023年6月经郑州大学第一附属医院临床诊断的颈动脉重度狭窄患者的临床资料及颈动脉CTA信息。后进一步处理颈动脉CTA图像,分别提取传统影像特征及影像组学特征,对传统影像组学特征进行单因素logistic回归与多因素logistic回归分析筛选出独立危险因素。对影像组学特征数据通过组内相关系数(intraclass correlation coefficient,ICC)、组间相关系数(Interclass correlation coefficient,ICC)、Spearman 相关性分析以及最小绝对收缩与选择算法(least absolute shrinkage and selection operator,LASS O)回归分析,筛选放射组学特征并影像组学评分,以预测患者是否会发生缺血性脑血管症状。通过R软件将纳入患者随机按7:3比例分为训练集与验证集。通过筛得数据分别构建传统CTA特征模型、影像组学模型及联合影像组学模型。通过受试者工作曲线(receiver operating characteristic curve,ROC)、ROC 曲线下面积(area under curve,AUC)、校准曲线、Hosmer-Lemeshow拟合优度检验及决策曲线(decision curve analysis,DCA)来综合评价模型的预测效能,最终将最优性能模型进行列线图可视化。研究结果共有106例ACAS患者被纳入本研究,其中症状组58例,无症状组48例。将所有数据随机分为70%的训练集(74例)和30%验证集(32例)。其中临床基础资料中有16个变量,传统CTA特征有10个变量被纳入研究,经过单因素和多因素差异分析后得到ACAS患者发生卒中的独立危险因素2个:斑块厚度、血管周围脂肪组织衰减值。经影像组学提取出来的变量1218个,经相关性检验、冗余性分析后剩余116个变量,将剩余116个变量纳入L ASSO回归分析后,得出8个影像组学特征。分别利用以上变量构建3组逻辑回归模型。分别为传统CTA模型、影像组学模型、联合影像组学模型。得出传统CTA模型、影像组学模型及联合影像组学模型的AUC值分别为0.733、0.834及0.850。并经过校准曲线、DCA曲线等进行综合验证得出联合影像组学模型性能优于其他模型。研究结论传统CTA特征模型、影像组学模型及联合影像组学模型均表现出较好的效能,其中联合影像组学模型表现最优,可作为一种无创性的高效能工具,辅助临床医生对该类患者进行临床诊断和决策时提高帮助。更多还原

【Abstract】 ObjectiveCarotid artery stenosis(CAS)due to atherosclerosis is a common chronic and i nflammatory disease and is considered a significant cause of stroke in clinical practi ce.The occurrence of CAS is typically associated with plaque formation within the a rterial wall and narrowing of the arterial lumen.These plaques are primarily compos ed of cholesterol,calcium salts,fibrous tissue,and inflammatory cells,forming bene ath the intima of the carotid arteries.As plaques accumulate and inflammatory reacti ons intensify,the lumen of the carotid artery gradually narrows,affecting normal blo od flow.If plaques become unstable or thrombosis occurs,it may lead to blood flow obstruction,resulting in serious complications such as stroke.Based on clinical pres entations,CAS can be classified into symptomatic and asymptomatic carotid artery s tenosis(ACAS).Patients with symptomatic severe stenosis typically present with se vere dizziness,headache,sensory and motor dysfunction,significantly impacting qu ality of life and posing a high risk of stroke.Thus,prompt surgical intervention is re quired upon detection.However,clinical guidelines for the diagnosis and treatment of asymptomatic severe carotid artery stenosis lack consensus internationally,leadin g to some high-risk ACAS patients not receiving timely appropriate surgical interven tion,while some low-risk patients undergo unnecessary invasive treatments.Therefo re,identifying high-risk individuals among ACAS patients and selecting appropriate treatment strategies is an urgent issue.Purpose1.To identify independent risk factors for stroke occurrence in patients with asymptomatic severe carotid artery stenosis.2.To establish and validate separate traditional CTA models,plaque-based rad iomics label models,and a combined radiomics model,aiming to assist clin icians in making diagnostic and therapeutic decisions.MethodsClinical data and neck CTA information of patients diagnosed with severe carot id artery stenosis at The First Affiliated Hospital of Zhengzhou University from June 2020 to June 2023 were retrospectively collected.Subsequently,neck CTA images were further processed,and traditional imaging features and radiomics features were extracted.Univariate and multivariate logistic regression analyses were performed o n traditional radiological features to identify independent risk factors for ischemic st roke.Radiomics features data were subjected to intra-class correlation coefficient(I CC),interclass correlation coefficient(ICC),Spearman correlation analysis,and leas t absolute shrinkage and selection operator(LASSO)regression analysis to screen ra diomics features and generate radiomics scores to predict the occurrence of ischemic cerebrovascular symptoms.Patients were randomly divided into training and validat ion sets at a ratio of 7:3 using R software.Traditional CTA feature models,radiomic s models,and combined radiomics models were separately constructed using the scr eened data.Receiver operating characteristic curve(ROC),area under the ROC curv e(AUC),calibration curve,Hosmer-Lemeshow goodness-of-fit test,and decision cu rve analysis(DCA)were used to comprehensively evaluate the predictive performan ce of the models.ResultsA total of 106 ACAS patients were included in this study,comprising 58 sympt omatic and 48 asymptomatic cases.All data were randomly divided into a training se t(74 cases,70%)and a validation set(32 cases,30%).Among the clinical baseline v ariables,16 were considered,while 10 traditional CTA features were included in the analysis.After univariate and multivariate analyses,two independent risk factors for stroke occurrence in ACAS patients were identified:plaque thickness and perivascul ar adipose tissue density.A total of 1,218 variables were extracted from radiomics fe atures.Following correlation and redundancy analyses,116 variables remained,whi ch were further subjected to LASSO regression analysis,resulting in the identificatio n of 8 radiomics features.Subsequently,three logistic regression models were constr ucted using these variables,namely,the traditional CTA model,radiomics model,an d combined radiomics model.The AUC values for the traditional CTA model,radio mics model,and combined radiomics model were 0.733,0.834,and 0.850,respectiv ely.Comprehensive validation through calibration curves,DCA curves,and other m ethods demonstrated superior performance of the combined radiomics model compa red to the other models.ConclusionTraditional CTA feature models,radiomics models,and combined radiomics m odels all demonstrated good performance,with the combined radiomics model exhib iting the most superior performance.It may serve as a non-invasive,high-efficiency tool to assist clinicians in clinical diagnosis and decision-making for this patient pop ulation.更多还原