基于CT平扫的影像组学模型预测基底节区脑出血血肿周围水肿早期扩张风险的价值  

Value of radiomics model based on CT plain scan to predict the risk of early expansion of perihematomal edema in basal ganglia intracerebral hemorrhage

作　　者：吴楚玲 郑旭峰[2] 林少帆[2] 林黛英[2] 马树华[1] WU Chuling;ZHENG Xufeng;LIN Shaofan;LIN Daiying;MA Shuhua(Department of Radiology,The First Affiliated Hospital of Shantou University Medical College,Shantou 515041,China;Department of Radiology,Shantou Central Hospital,Shantou 515041,China)

机构地区：[1]汕头大学医学院第一附属医院放射科,广东汕头515041 [2]汕头市中心医院放射科,广东汕头515041

出　　处：《汕头大学医学院学报》2024年第4期200-206,共7页Journal of Shantou University Medical College

摘　　要：目的:探讨基于CT平扫的影像组学模型预测基底节区脑出血血肿周围水肿(perihematomal edema,PHE)早期扩张风险的价值。方法:回顾性纳入2021年1月—2023年12月在汕头市中心医院住院的267例基底节区脑出血患者,其中男性165例,女性102例,年龄(63±11)岁,早期PHE扩张患者169例,非早期PHE扩张患者98例。将所有研究对象按7∶3比例随机分为训练集(187例)和测试集(80例),分别进行模型建立和内部验证。通过3D-Slicer软件在基线CT平扫图像上分别半自动分割脑血肿和周围水肿2个靶区,再用FAE影像组学软件分别提取2个靶区的影像组学特征。利用logistic回归建立基于临床因素和常规CT征象(临床-常规CT模型)、影像组学特征(影像组学模型)和两者联合(联合模型)的预测模型。采用受试者工作特征曲线下面积(area under curve,AUC)、校准曲线和决策曲线评估模型的预测价值。结果:由年龄、格拉斯哥昏迷评分、血肿体积、岛征和影像组学评分构建的联合模型在训练集和测试集中的AUC值(分别为0.956、0.943)高于临床-常规CT模型(分别为0.732、0.683)和影像组学模型(分别为0.912、0.871)(均P<0.05)。联合模型在训练集和测试集中的校准曲线Brier评分(分别为0.077、0.090)低于临床-常规CT模型(分别为0.194、0.218)和影像组学模型(分别为0.109、0.147)。在决策曲线分析中,联合模型在训练集和测试集中的大部分阈值概率范围内均具有最高的净获益。结论:基于CT平扫的影像组学联合模型在预测基底节区脑出血PHE的早期扩张风险方面具有较好的预测效能。Objective:To investigate the value of radiomics model based on CT plain scan in predicting the risk of early expansion of perihematomal edema(PHE) in basal ganglia intracerebral hemorrhage.Methods:A total of 267patients with basal ganglia intracerebral hemorrhage who were hospitalized in Shantou Central Hospital from January2021 to December 2023 were retrospectively enrolled,including 165 males and 102 females,aged(63±11) years old,169 patients with early PHE expansion and 98 patients without early PHE expansion.All subjects were randomly divided into a training set(187 cases) and a test set(80 cases) in a ratio of 7∶3 for model establishment and internal validation,respectively.The 3D-Slicer software was used to semi-automatically segment the two target areas of cerebral hematoma and surrounding edema on the baseline CT plain scan images,and the FAE radiomics software was used to extract the radiomics features of the two target areas.Logistic regression was used to establish a prediction model based on clinical factors and conventional CT signs(clinical-conventional CT model),radiomics features(radiomic model),and the combination of the two(combined model).The area under the receiver operating characteristic curve(AUC),calibration curve and decision curve were used to evaluate the predictive value of the model.Results:The combined model constructed by age,Glasgow coma score,hematoma volume,island sign and radiomics score had higher AUC values in the training and test sets(0.956 and 0.943,respectively) than the clinical-conventional CT model(0.732 and 0.683,respectively) and the radiomics model(0.912 and 0.871,respectively)(all P<0.05).The calibration curve Brier scores of the combined model in the training and test sets(0.077 and 0.090,respectively) were lower than those of the clinical-conventional CT model(0.194 and 0.218,respectively) and the radiomics model(0.109 and 0.147,respectively).In the decision curve analysis,the combined model had the highest net benefit in most threshold probability ranges in bo

关 键 词：脑出血 基底节区 血肿周围水肿 影像组学 体层摄影术 X线计算机 

分 类 号：R445.3[医药卫生—影像医学与核医学]