Comparative analysis of CT radiomics features between cephalic para-carcinoma tissues and resection margin in esophageal squamous cell carcinoma after neoadjuvant chemotherapy and immunotherapy

GOU Yueqin; GAO Dan; OU Jing; CHEN Tianwu

doi:10.19300/j.2025.L21920

Abstract：Objective To explore the feasibility of radiomics models based on contrast-enhanced computed tomography(CECT)in distinguishing cephalic para-carcinoma tissues and resection margin in esophageal squamous cell carcinoma(ESCC)following neoadjuvant chemotherapy and immunotherapy(NACI).Methods This retrospective study included 188 pathologically confirmed ESCC patients who underwent NACI,recruited from two medical centers.A total of 138 patients from Center A were randomly divided into a training set(97 cases)and an internal validation set(41 cases)at a 7∶3 ratio,while 50 patients from Center B served as an external validation set.Using an open-source software 3D-Slicer,four regions of interest(ROIs)representing cephalic para-carcinoma tissues(P1,P2,P3,and P4)at 1 cm,2 cm,3 cm,and 4 cm above the tumor margin,respectively,and one ROI for resection margin tissue(P5,5 cm above the tumor)were delineated on CECT images.Radiomics features were extracted using the Pyradiomics package.The radiomics features obtained from four cephalic para-carcinoma tissues were individually paired with those of resection margin tissue to differentiate between them,which were designated as groups P1,P2,P3,and P4,respectively.Univariate analysis and the least absolute shrinkage and selection operator(LASSO)method were employed to select optimal radiomics features in the training sets,and logistic regression models were constructed.The area under the receiver operating characteristic(ROC)curve(AUC)was used to assess the discriminatory performance of the radiomics models.Results The AUCs of the P1 model in the training,internal validation,and external validation sets were 0.831,0.820,and 0.787,respectively.The AUCs of the P2 model were 0.809,0.797,and 0.769,respectively.Both the P1 and P2 models demonstrated good discriminatory performance(AUC＞0.76),with the P1 model achieving higher AUC values than the P2 model in all datasets.Conclusion The CECT-based radiomics model demonstrates high efficacy in distinguishing cephalad peritumoral(P1 and P2)and resection margin tissues in ESCC following NACI.