基于胃充盈超声造影下的影像组学构建胃间质瘤NIH危险度分级预测模型  

Prediction model of NIH risk stratification for gastrointestinal stromal tumor based on ultrasonographic radiomics by oral contrast enhanced ultrasonography

作　　者：杨凡 刘春伟[2] 张岱 赵利辉 毛怡然 穆洁 王海玲 魏玺 Yang Fan;Liu Chunwei;Zhang Dai;Zhao Lihui;Mao Yiran;Mu Jie;Wang Hailing;Wei Xi(Department of Ultrasound,Diagnosis and Treatment,Tianjin Medical University Cancer Institute and Hospital,National Clinical Research Center for Cancer,Tianjin′s Clinical Research Center for Cancer,Key Laboratory of Cancer Prevention and Therapy,Tianjin 300060,China;Department of Cardiology,Tianjin Chest Hospital,Tianjin 300222,China)

机构地区：[1]天津医科大学肿瘤医院超声诊疗科国家恶性肿瘤临床医学研究中心天津市肿瘤防治重点实验室天津市恶性肿瘤临床医学研究中心,天津300060 [2]天津市胸科医院心内科,天津300222

出　　处：《中华超声影像学杂志》2023年第12期1062-1069,共8页Chinese Journal of Ultrasonography

基　　金：国家自然科学基金面上项目(82272008);天津市医学重点学科(专科)建设项目(TJYXZDXK-009A);天津市健康科研项目(ZD20018)。

摘　　要：目的基于胃充盈超声造影下的影像组学构建胃间质瘤美国国立卫生研究院(NIH)危险度分级的预测模型,包括临床超声模型、超声影像组学模型以及两者的联合模型,分别探讨三种模型对胃间质瘤NIH危险度分级的预测效果。方法回顾性分析2021年6月至2022年6月于天津医科大学肿瘤医院接受手术治疗且病理证实为胃间质瘤的患者共204例,收集其临床及超声影像资料,其中NIH危险度分级为高危险度及中危险度的患者共101例,纳入高危组;NIH危险度分级为低危险度及极低危险度的患者共103例,纳入低危组。通过ITK-SNAP软件对胃间质瘤最大径线的超声图像进行手动分割,应用Python 3.8.7中的Pyradiomics(v3.0.1)模块对所分割的感兴趣区(ROI)图像进行影像组学特征提取。将患者按照7∶3的比例随机分为训练集和测试集。应用Sklearn模块,通过XGBoost算法构建临床超声模型、超声影像组学模型以及两者的联合模型,评估ROC曲线下面积(AUC)、敏感性、特异性及准确性;通过Delong检验比较三种模型的预测能力;应用校准曲线评价模型性能,应用临床决策曲线确定患者的净获益。结果从ROI中共提取578个影像组学特征,经回归降维处理,最终保留8个超声影像组学特征用于建模。最终,测试结果显示临床超声模型、超声影像组学模型以及联合模型的AUC、敏感性、特异性及准确性分别为0.75、69.3%、68.9%、69.1%,0.87、79.2%、81.6%、80.4%,0.91、80.2%、83.5%、81.9%。Delong检验结果显示,对于胃间质瘤NIH危险度分级预测的ROC曲线,超声影像组学模型与临床超声模型AUC间的差异有统计学意义(Z=2.698,P<0.001),联合模型明显优于临床超声模型(Z=4.062,P<0.001)及超声影像组学模型(Z=2.225,P=0.026)。校准曲线显示出联合模型具备较高性能,决策曲线同样显示出联合模型具有优越的临床实用性。结论基于胃充盈超声造影下的影像组学�Objective To investigate the prediction of National Institute of Healthy(NIH)risk stratification of gastrointestinal stromal tumor(GIST)based on clinical ultrasound model,ultrasonographic radiomics model and combined model by oral contrast enhanced ultrasonography.Methods The clinical and ultrasound imaging data of 204 gastric GIST patients attending Tianjin Medical University Cancer Institute and Hospital from June 2021 to June 2022 were retrospectively analyzed,among whom a total of 101 patients with high and moderate NIH risk stratification GIST confirmed by postoperative pathology were included in the high risk group,and a total of 103 patients with low and extremely low NIH risk stratification GIST were in the low risk group.The ultrasound images of the largest diameter of the GIST were manually segmented by ITK-SNAP software,and Pyradiomics(v3.0.1)module in Python 3.8.7 was applied to extract ultrasonographic radiomics features from the ROI segmented images.The patients were randomly divided into training and validation sets in the ratio of 7∶3.The XGBoost of Sklearn module was applied to construct the clinical ultrasound imaging model,ultrasonographic radiomics model,and combined model.Then the area under ROC curve(AUC),sensitivity,specificity,and accuracy were evaluated;the predictive ability of the three models was compared by Delong test.Calibration Curve was applied to evaluate the model performance,and the clinical Decision Curve Analysis was applied to determine the net benefit to patients.Results A total of 578 ultrasonographic radiomics features were extracted from ROI,and 8 ultrasonographic radiomics features were finally retained for modeling after regression and dimensionality reduction.Finally,test results showed that AUC,sensitivity,specificity and accuracy of clinical ultrasound imaging model,ultrasonographic radiomics model and combined model were 0.75,69.3%,68.9%,69.1%;0.87,79.2%,81.6%,80.4%;0.91,80.2%,83.5%,81.9%,respectively.Delong test showed that the difference of AUC between ultrason

关 键 词：胃肠道间质瘤 超声检查 影像组学 危险度分级 

分 类 号：R735[医药卫生—肿瘤]