CT影像组学模型在儿童节细胞神经母细胞瘤和神经母细胞瘤鉴别诊断中的价值  

Application of CT image omics model in the differential diagnosis of ganglioneuroblastoma and neuroblastoma in childhood

作　　者：李海燕 李志强 赵伟 全帅 张思琦 徐树明 Li Haiyan;Li Zhiqiang;Zhao Wei;Quan Shuai;Zhang Siqi;Xu Shuming(CT Room,Shanxi Children's Hospital(Shanxi Maternal and Child Health Hospital),Taiyuan 030013,China;Department of Radiology,Taiyuan Children's Hospital(Taiyuan Maternal and Child Health Care Hospital),Taiyuan 030025,China;GE Healthcare(Shanghai)Co.Ltd.Medical Affairs,Shanghai 210000,China)

机构地区：[1]山西省儿童医院(山西省妇幼保健院)CT室,太原030013 [2]太原市儿童医院(太原市妇幼保健院)放射科,太原030025 [3]通用电气药业(上海)有限公司医学部,上海210000

出　　处：《肿瘤研究与临床》2024年第11期858-862,共5页Cancer Research and Clinic

摘　　要：目的探讨CT影像组学模型在儿童节细胞神经母细胞瘤(GNB)和神经母细胞瘤(NB)鉴别诊断中的价值。方法回顾性病例系列研究。收集2013年1月至12月在山西省儿童医院行手术并经病理检查证实为NB(23例)和GNB(23例)患儿的临床及影像学资料。以DICOM格式从PACS系统中导出所有患儿术前CT平扫期、动脉期及静脉期原始图像,应用ITK-SNAP(3.4.0版)软件对每例患儿术前各期的病灶进行逐层手动勾画并提取影像组学特征,经最小绝对收缩选择算子和逐步多因素logistic回归方法进行特征筛选,筛选出各期的有效特征,使用logistic模型建立对应期相影像组学模型。通过受试者工作特征曲线、校准曲线和决策曲线分析评价各期影像组学模型的诊断效能。结果从CT三期原始图像中共提取出1361个影像组学特征,经多因素logistic回归最终筛选特征建立模型,分别提取出平扫期4个、动脉期2个、静脉期3个及三期联合7个特征。平扫期模型的曲线下面积(AUC)为0.940,准确度89.1%、灵敏度91.3%、特异度87.0%;动脉期模型的AUC为0.923,准确度84.8%、灵敏度82.6%、特异度87.0%;静脉期模型的AUC为0.949,准确度87.8%、灵敏度83.3%、特异度91.3%;三期联合模型的AUC为0.964,准确度95.1%、灵敏度94.7%、特异度95.5%;单一期相影像组学模型对于儿童NB和GNB鉴别诊断均具有良好效能,三期联合影像组学模型的AUC、准确度、灵敏度及特异度均高于各单一期相影像组学模型。校准曲线和决策曲线显示,三期联合模型鉴别诊断儿童NB和GNB的概率与观察值有较高一致性,可获得较好净收益。结论基于CT的影像组学在鉴别诊断NB和GNB具有较高临床应用价值。Objective:To investigate the application of CT image omics model in the differential diagnosis of ganglioneuroblastoma(GNB)and neuroblastoma(NB)in childhood.Methods:A retrospective case series study was performed.The clinical and imaging data of 23 NB and 23 GNB pediatric patients confirmed by surgery and pathology in Shanxi Children's Hospital from January 2013 to December 2013 were collected.The original CT images in the normal scan phase,arterial phase and venous phase of all the children before operation were extracted from the PACS system in DICOM format.ITK-SNAP(ver.3.4.0)software was applied to manually outline and extract the image omics features layer by layer of the lesions in the normal scan phase,arterial phase and venous phase of each patient before surgery.The minimum absolute contraction selection operator and stepwise multi-factor logistic regression method were used to screen out effective features in different scan phases.The corresponding phase image omics model was established by using logistic model.The diagnostic efficiency of each phase of the image omics model was evaluated by using the receiver operating characteristic curve,calibration curve and decision curve.Results:A total of 1361 image omics features were extracted from the original CT images in the 3 phases.The model was established by using multi-factor logistic regression to extract 4 features in the normal scan phase,2 features in the arterial phase,3 features in the venous phase and 7 features in the combination of the 3 phases.The area under the curve(AUC)of the model in the normal scan phase was 0.940,the accuracy was 89.1%,the sensitivity was 91.3%and the specificity was 87.0%;the AUC of the model in the arterial phase was 0.923,the accuracy was 84.8%,the sensitivity was 82.6%,and the specificity was 87.0%;the AUC of the model in the venous phase was 0.949,the accuracy was 87.8%,the sensitivity was 83.3%,and the specificity was 91.3%;the AUC of 3 phases combined model was 0.964,the accuracy was 95.1%,the sensitivity was 94.7%

关 键 词：影像组学 儿童 节细胞神经母细胞瘤 神经母细胞瘤 诊断 鉴别 

分 类 号：R739.4[医药卫生—肿瘤] R730.44[医药卫生—临床医学]