基于PET影像组学预测食管癌脉管及神经侵犯  

Prediction of Vascular and Neural Invasion in Esophageal Cancer Based on PET Radiomics

作　　者：李昂 季仲友[1] 蒋韩 林莉莉 Li Ang;Ji Zhongyou;Jiang Han;Lin Lili(PET/CT Room,Union Hospital of Fujian Mediccl University,Fuzhou,Fujian 350001)

机构地区：[1]福建医科大学附属协和医院PET/CT室,福建福州350001

出　　处：《现代医用影像学》2021年第12期2216-2219,共4页Modern Medical Imageology

摘　　要：目的:探讨基于PET影像组学方法构建食管癌脉管及神经侵犯预测模型的可行性。方法:回顾性分析收集2015年至2020年在福建医科大学附属协和医院收治的143例食管癌胸腔镜根治术患者的病理资料及术前PET/CT资料。用3D Slicer软件,提取PET影像组学参数构建模型。训练组(70%)用于模型构建及特征筛选,验证组(30%)用于模型验证。通过ROC曲线评估模型的曲线下面积(AUC)、灵敏度、特异度及准确度。结果:143例食管癌患者,合并癌栓者40例,伴有神经侵犯者39例。通过筛选得到脉管瘤栓PET影像组学特征5个以及神经侵犯PET影像组学特征5个。Model_(thum)包含1个形态特征:Sphericity;1个GLDM特征:Dependence Non Uniformity;2个GLSZM特征:Small Area Low Gray Level Emphasis、Zone Percentage。Model_(neo)包含2个一阶特征Energy、Total Energy。Model_(thum)模型的曲线下面积(AUC)为0.7206(CI95%:0.4618~0.9794),敏感度为0.525,特异度为0.8529,准确度为0.8095。Model_(neo)的AUC为0.7185(CI95%:0.5482~0.8887),敏感度为0.6774,特异度为0.7272,准确度为0.6904。结论:PET影像组学方法可以作为预测食管癌脉管及神经侵犯的可用方法。Objective:To explore the feasibility of building a prediction model of vascular and neural invasion of esophageal cancer based on PET radiomics.Methods:the pathological data and preoperative PET/CT data of 143 patients with esophageal cancer undergoing VATS in Fujian Medical University Union Hospital from 2015 to 2020 were retrospectively analyzed.3 D slicer software was used to extract PET imaging parameters to construct the model.The training group(70%)was used for model construction and feature selection,and the validation group(30%)was used for model validation.Area under curve(AUC),sensitivity,specificity and accuracy of the model were evaluated by ROC curve.Results:among 143 cases of esophageal cancer,40 cases had tumor thrombus and 39 cases had nerve invasion.Through screening,5 PET imaging features of vascular tumor thrombus and 5 PET imaging features of nerve invasion were obtained.Model_(thum)has one shape feature:Sphericity;One GLDM feature:Dependencenonuniformity;Two glszm features:small area low gray level emphasis and zone percentage.Model_(thum)includes two first-order features Eergy and Total Energy.The area under the curve(AUC)of model;model was 0.7206(CI95%:0.4618-0.9794),the sensitivity was 0.525,the specificity was 0.8529,and the accuracy was 0.8095.AUC of model_(neo)was 0.7185(CI95%:0.5482-0.8887),sensitivity was 0.6774,specificity was 0.7272,accuracy was 0.6904.Conclusion:PET radiomics method can be used to predict vascular and neural invasion of esophageal cancer.

关 键 词：PET 影像组学 食管癌 脉管瘤栓 神经侵犯 

分 类 号：R735.1[医药卫生—肿瘤] R730.44[医药卫生—临床医学]