基于临床与头颅CT特征构建的积分模型预测高血压患者颅内微出血灶的程度  

A scoring model based on clinical and brain CT features for predicting cerebral microbleeds degree in hypertension

作　　者：汪鑫斌 邱勇刚 董浩 楼存诚[1] 胡亦程 张菁 陈狄洪[1] 徐建霞 余日胜[2] WANG Xinbin;QIU Yonggang;DONG Hao;LOU Cuncheng;HU Yicheng;ZHANG Jing;CHEN Dihong;XU Jianxia;YU Risheng(Department of Radiology,First People’s Hospital of Xiaoshan District,Hangzhou 311200,China;Department of Radiology,Second Hospital Affiliated of Medicine School of Zhejiang University,Hangzhou 310009,China;Department of Radiology,Second Hospital Affiliated of Zhejiang Chinese Medical University,Hangzhou 310005,China)

机构地区：[1]杭州市萧山区第一人民医院放射科,浙江杭州311200 [2]浙江大学医学院附属第二医院放射科,浙江杭州310009 [3]浙江中医药大学附属第二医院放射科,浙江杭州310005

出　　处：《中国中西医结合影像学杂志》2025年第2期202-207,共6页Chinese Imaging Journal of Integrated Traditional and Western Medicine

基　　金：杭州市医药卫生科技计划一般项目(B20210068);浙江省医药卫生科技计划一般项目(2024KY250)。

摘　　要：目的:探讨基于临床与头颅CT特征的积分模型预测高血压患者颅内微出血灶(CMBs)程度的价值。方法:回顾性分析170例高血压患者的临床、头颅CT及MRI资料。根据MRI结果,分为CMBs≤5个组和>5个组。按7∶3比例随机分为训练集120例和验证集50例。应用单因素及二元logistic回归分析比较训练集CMBs>5个组与≤5个组的临床及头颅CT特征,并筛选组间差异有统计学意义的因素建立logistic回归模型,进一步加权赋分得到积分模型。绘制ROC曲线评估模型预测效能。采用DeLong检验比较logistic回归模型与积分模型的AUC。同时将积分模型划分为3个积分段。结果:logistic回归分析显示,高血压病程、同型半胱氨酸水平、腔隙性脑梗死(LI)分级和脑白质疏松(LA)分级是鉴别CMBs≤5个与>5个的独立因素。积分模型包括高血压病程≥10年(2分)、同型半胱氨酸水平升高(2分)、LI2~3级(4分)、LA2~3级(2分),其AUC为0.857(95%CI0.782~0.914),敏感度与特异度分别为78.3%、79.7%,截断值为3分。将积分模型划分为3个积分段:0~1分,2~3分,4~10分。随着积分增加,训练集、验证集各积分段CMBs>5个的发生率逐步增高。结论:基于临床与头颅CT特征构建的积分模型对预测高血压患者CMBs程度具有较高价值,可为临床诊疗提供依据。Objective:To explore the value of a scoring model based on clinical and brain CT features to predict the degree of cerebral microbleeds(CMBs)in hypertensive patients.Methods:Clinical,brain CT and MRI data of 170 hypertensive patients were retrospectively analyzed.According to the CMBs count on MRI images,170 cases were divided into two groups,a group of CMBs≤5 and a group of CMBs>5.They were randomly divided into training cohort(120 cases)and validation cohort(50 cases)at a 7∶3 ratio.Univariate and binary logistic regression were used to analyze clinical and brain CT features between the group of CMBs≤5 and the group of CBMs>5 in the training cohort,and the features with statistical differences were screened to build a logistic regression model.A weighted scoring system was then used to develop a scoring model.The predictive efficacy of the model was assessed using ROC curves.DeLong test was used to compare the AUC of the logistic regression model and the scoring model.Finally,the scoring model was classified into three integral intervals.Results:Logistic regression analysis showed that hypertension duration,homocysteine,lacunar cerebral infarction(LI)grade,and cerebral leukoaraiosis(LA)grade were independent factors in identifying the group of CMBs≤5 and CMBs>5.The scoring model included hypertension duration≥10 years(2 points),elevated homocysteine(2 points),LI grade 2—3(4 points),and LA grade 2—3(2 points),and the ROC curve had the AUC of 0.857(95%CI 0.782—0.914),the sensitivity of 78.3%,the specificity of 79.7%,the cutoff value of 3 points.The scoring model was divided into three integral intervals,0—1,2—3,and 4—10 points.The incidence of CMBs>5 in each integral interval of the training cohort and validation cohort gradually increased with increasing score.Conclusions:The scoring model based on the clinical and brain CT features has a high value for predicting the CMBs degree in hypertensive patients,and can provide a basis for clinical diagnosis and treatment.

关 键 词：高血压 颅内微出血灶 体层摄影术 X线计算机 积分模型 

分 类 号：R74[医药卫生—神经病学与精神病学]