基于瘤内及瘤周放射组学特征联合临床特征、能谱CT定量参数的列线图术前预测胃癌周围神经侵犯  

A Nomogram Based on Intratumoral and Peritumoral Radiomic Features Combined with Clinical Features and Spectral CT Quantitative Parameters for Preoperative Prediction of Perineural Invasion in Gastric Cancer

作　　者：卢洲 张思维 王泽楷 杨晓 吴振东[1] 胡曙东 孙宗琼[1] LU Zhou;ZHANG Siwei;WANG Zekai(Department of Imaging,Affiliated Hospital of Jiangnan University,Wuxi,Jiangsu Province 214000,P.R.China)

机构地区：[1]江南大学附属医院影像科,无锡214000

出　　处：《临床放射学杂志》2025年第3期472-478,共7页Journal of Clinical Radiology

基　　金：江南大学附属医院临床医学研究与转化课题资助项目(编号:YJY202305)。

摘　　要：目的探讨基于碘密度图瘤内及瘤周放射组学特征联合临床特征、肿瘤能谱CT定量参数的列线图术前预测胃癌周围神经侵犯(PNI)的价值。方法回顾性搜集经胃镜证实为胃腺癌并行手术切除的190例患者,根据术后病理分为PNI阳性组(124例)和PNI阴性组(66例)。于术前1周内行能谱CT增强动脉期(AP)和静脉期(VP)扫描。基于能谱CT图测量胃癌病灶的能谱参数,包括Z有效原子序数(Z-Eff)、碘密度(IoD)、标准化碘密度(NIoD)、能谱曲线斜率(λ)。采用3D Slicer包在静脉期IoD(IoDVP)图上勾画感兴趣区(ROI),使用Pyradiomics包提取瘤内及瘤周ROI的放射组学特征,按照7∶3的比例将数据随机分为训练集(133例)和验证集(57例)。采用组内相关系数(ICC)、最大相关最小冗余(MRMR)以及最小绝对收缩和选择算子(LASSO)算法进行特征筛选,通过Logistic回归构建每例患者的瘤内、瘤周及瘤内+瘤周放射组学标签(Radscore)。搜集患者临床病理信息,包括患者性别、年龄、CT-T分期、CT-N分期、肿瘤最长径、Borrmann分型、Lauren分型等。采用独立样本t检验或Mann-Whitney U检验比较两组间计量资料的差异,采用χ2检验或Fisher确切概率法比较分类资料的差异。采用多因素Logistic回归分析筛选胃癌PNI的独立危险因素,并建立联合预测模型,用列线图呈现。采用受试者操作特征曲线评估各参数或模型的预测效能,曲线下面积(AUC)的比较采用DeLong检验。结果在训练集,PNI阳性组和PNI阴性组间临床病理因素CT-T分期、CT-N分期、肿瘤最长径、Borrmann分型、Lauren分型和分化程度差异均有统计学意义(P<0.05)。PNI阳性组胃癌病灶的能谱参数Z-Eff VP、IoD VP、NIoD VP、λVP及放射组学标签瘤内、瘤周、瘤内+瘤周Radscore值均高于PNI阴性组,差异有统计学意义(均P<0.05),其预测PNI的AUC(95%CI)分别为0.647(0.560~0.728)、0.740(0.656~0.812)、0.664(0.577~0.743)、0.611(0.522~0.694)、0Objective To explore the value of a nomogram based on intratumoral and peritumoral radiomic features of iodine density maps combined with clinical features and tumor spectral CT quantitative parameters for preoperative prediction of perineural invasion(PNI)in gastric cancer.Methods A total of 190 patients with gastric adenocarcinoma confirmed by gastroscopy were retrospectively collected and divided into PNI-positive(124 cases)and PNI-negative(66 cases)groups according to postoperative pathology.Spectral CT arterial phase(AP)and venous phase(VP)scans were performed within 1 week before surgery.Spectral parameters,including effective atomic number(Z-Eff),iodine density(IoD),normalized iodine density(NIoD),and spectral curve slope(λ),were measured based on spectral CT images.The 3D Slicer package was used to outline regions of interest(ROIs)on IoD VP maps,and the Pyradiomics package was used to extract intratumoral and peritumoral radiomic features from these ROIs.The data were randomly divided into a training set(133 cases)and a validation set(57 cases)in a 7∶3 ratio.Intra-class correlation coefficient(ICC),maximum relevance minimum redundancy(MRMR),and least absolute shrinkage and selection operator(LASSO)algorithms were used for feature selection.Logistic regression was used to construct intratumoral,peritumoral,and combined intratumoral+peritumoral radiomic scores(Radscores)for each patient.Clinicopathological information,including patient gender,age,CT-T stage,CT-N stage,tumor longest diameter,Borrmann classification,and Lauren classification,was collected.Independent samples t-tests or Mann-Whitney U tests were used to compare quantitative data between the two groups,and chi-square tests or Fisher's exact tests were used for categorical data.Multivariate logistic regression analysis was used to identify independent risk factors for gastric cancer PNI,and a joint prediction model was established and presented as a nomogram.The predictive performance of each parameter or model was evaluated using receiver op

关 键 词：胃癌 能谱CT 瘤内 瘤周 放射组学 周围神经侵犯 

分 类 号：R73[医药卫生—肿瘤]