摘要
目的探究食管癌新辅助治疗后CT影像特征动态变化与术后病理完全缓解(pCR)的相关性,构建基于CT影像组学特征的预测模型。方法回顾性纳入2020年6月至2024年5月在无锡市人民医院接受新辅助治疗及食管癌根治术的178例患者。根据术后病理结果将患者分为病理完全缓解组(pCR组,即ypT0N0,n=53)和非病理完全缓解组(非pCR组,n=125)。采用256层螺旋CT于新辅助治疗前及治疗后4周进行双期增强扫描,通过Elastix5.0软件进行非刚性配准校正体位差异,ITK-SNAP3.8.0勾画肿瘤三维容积,基于PyRadiomics3.0从平扫、动脉期及静脉期图像中提取1426个影像组学特征,经一致性检验、差异分析、LASSO回归三级筛选流程确定CT影像组学核心特征,采用Logistic回归构建基础预测模型,生成标准化影像组学评分(Rad-score)。同时整合临床独立预测因子构建联合模型,通过分层随机抽样(7∶3)划分训练集(124例)与验证集(54例),对不同模型的预测效能进行评估。结果两组患者在cT分期、cN分期、肿瘤分化程度及化疗周期完成率上存在显著差异(P<0.05)。与非pCR组相比,pCR组中cT2分期占比更高,cN0比例更优,高分化肿瘤比例显著,化疗周期完成率更高。pCR组在肿瘤体积缩小率、动脉期CT值变化、淋巴结短径缩小率、动脉期灰度共生矩阵(GLCM)能量值和长径缩小率均显著高于非pCR组;静脉期肿瘤与肌肉CT值比值(T/M)明显低于非pCR组,差异有统计学意义(P<0.001)。多因素Logistic回归分析发现,在基线临床特征中,仅化疗周期完成率是pCR的独立预测因子(P=0.008)。CT影像特征中,淋巴结短径缩小率、动脉期GLCM能量值、肿瘤体积缩小率与pCR呈现较强的正相关性(P<0.05),长径缩小率、动脉期CT值变化与pCR亦呈显著正相关(P<0.05),静脉期T/M比值与pCR则呈现显著负相关(P<0.05)。经三级筛选流程,最终确定6个影像组学核心特征,涵盖形态学参数(如肿瘤体积与表面形态)、一阶灰度统计特征(反映强化程度分布)以及高阶纹理特征(表征肿瘤异质性)。CT影像组学模型在训练集和验证集的AUC分别为0.868和0.841,联合模型显著提升至0.909和0.914(Z=2.470、2.891,P=0.013、0.004)。联合模型在验证集的敏感度88.2%,特异度83.8%,显著优于单一影像模型(敏感度67.6%)。Bootstrap验证显示联合模型95%CI更窄(0.841~0.987),表明其稳定性更高。结论食管癌患者在接受新辅助治疗后,肿瘤体积缩小、动脉期强化、静脉期T/M比值降低等CT影像特征动态变化与pCR显著相关,联合化疗周期完成率的预测模型可为术前精准评估病理缓解提供量化工具。
Objective To investigate the correlation between dynamic changes in CT imaging features after neoadjuvant therapy and postoperative pathologic complete response(pCR)in esophageal cancer and to construct a prediction model based on CT radiomic features.Methods A total of 178 patients who underwent neoadjuvant therapy and radical esophagectomy at Wuxi People's Hospital were retrospectively enrolled between June 2020 and May 2024.According to the postoperative pathological results(ypT0N0 was defined as pCR),the patients were divided into pCR group(n=53)and non-pCR group(n=125).Philips 256-slice spiral CT was used for dual-phase enhanced scans before and 4 weeks after neoadjuvant therapy.Elastix5.0 software performed non-rigid registration to correct positional differences,and ITK-SNAP3.8.0 was utilized for 3D tumor volume delineation.A total of 1426 radiomic features were extracted from non-contrast,arterial,and venous phase images using PyRadiomics3.0.Core CT radiomic features were identified through a three-step screening process(consistency testing,differential analysis,LASSO regression).A logistic regression-based predictive model was established to generate a standardized radiomics score(Rad-score).A combined model integrating clinical independent predictors was further developed.Patients were divided into training(124 cases)and validation(54 cases)sets via stratified random sampling(7:3),and the predictive performance of the models was evaluated.Results Significant differences were observed between groups in cT stage,cN stage,tumor differentiation,and chemotherapy cycle completion rate(P<0.05).Compared to the non-pCR group,the pCR group had a higher proportion of cT2 stage,cN0 status,well-differentiated tumors,and higher chemotherapy completion rates.The pCR group exhibited significantly greater reductions in tumor volume,arterial phase CT value changes,lymph node short-axis reduction rate,arterial phase gray-level co-occurrence matrix(GLCM)energy value,and long-axis reduction rate;the ratio of CT value of venous tumor to muscle ratio(T/M)was significantly lower than that of the non-pCR group,with statistical significance(P<0.001).Multivariate logistic regression revealed that among the baseline clinical characteristics,only the chemotherapy cycle completion rate was an independent clinical predictor of pCR(P=0.008).Among CT imaging features,lymph node short-axis reduction rate,arterial phase GLCM energy value,tumor volume reduction rate,long-axis reduction rate,and arterial CT value changes showed strong positive correlations with pCR(P<0.05),whereas venous phase T/M ratio correlated negatively(P<0.05).After a three-level screening process,six core radiomic features were identified,encompassing morphology(such as tumor volume and surface characteristics),first-order gray-level statistics(reflecting enhancement distribution),and high-order texture features(tumor heterogeneity).The CT radiomics model achieved AUCs of 0.868(training set)and 0.841(validation set),while the combined model improved significantly to 0.909 and 0.914,respectively(Z=2.470,2.891;P=0.013,0.004).In the validation set,the combined model demonstrated 88.2%sensitivity and 83.8%specificity,outperforming the radiomics-only model(88.2%vs.67.6%sensitivity).Bootstrap validation confirmed higher stability for the combined model(95%CI:0.841-0.987).Conclusion Dynamic changes in CT imaging features—including tumor volume reduction,arterial phase enhancement,and decreased venous T/M ratio—are strongly associated with pCR after neoadjuvant therapy in esophageal cancer.The combined model integrating chemotherapy cycle completion rate provides a quantitative tool for preoperative assessment of pathologic response.
作者
胡斌
柳林
Hu Bin;Liu Lin(Department of Thoracic Surgery,Wuxi People's Hospital,Wuxi 214000,China)
出处
《中华消化病与影像杂志(电子版)》
2025年第6期627-634,共8页
Chinese Journal of Digestion and Medical Imageology(Electronic Edition)
关键词
食管鳞状细胞癌
新辅助治疗
病理完全缓解
体层摄影术
X线计算机
影像组学
Esophageal squamous cell carcinoma
Neoadjuvant therapy
Pathologic complete response
Tomography
X-ray computer
Radiomics
