3434 - A Longitudinal CT-Based Subregional Radiomics Nomogram for Predicting Local Recurrence-Free Survival in Esophageal Squamous Cell Carcinoma Treated with Definitive Chemoradiotherapy: A Multicenter Study
Presenter(s)
J. Gong1, Q. Wang2, and L. Zhao3; 1Department of Radiation Oncology, First Affiliated Hospital of Air Force Medical University, Xi'an, Shaanxi, China, 2Department of Radiation Oncology, Sichuan Cancer Hospital and Institution, Sichuan Cancer Center, School of Medicine, University of Electronic Science and Technology of China, Radiation Oncology Key Laboratory of Sichuan Province, Chengdu, China, 3Department of Radiation Oncology, Xijing Hospital, Air Force Medical University, xi an, shan xi, China
Purpose/Objective(s): Despite definitive chemoradiotherapy (dCRT) being the standard treatment for locally advanced esophageal squamous cell carcinoma (ESCC), more than 50% of patients eventually developed local recurrence. Current prognostic models predominantly rely on pretreatment static clinical factors, which fail to account for tumor and peritumoral microenvironment infiltration and therapy-induced spatiotemporal heterogeneity. This study aimed to develop and validate a longitudinal CT-based subregional radiomics nomogram for predicting local recurrence-free survival (LRFS) in ESCC patients receiving dCRT.
Materials/Methods: This multicenter retrospective study enrolled 371 ESCC patients from Xijing Hospital (training: 211; internal validation: 70) and Sichuan Cancer Hospital (external validation: 90). Habitat analysis was performed on tumor and peritumoral regions using pre- and post-treatment contrast-enhanced CT to identify biologically distinct subregions. A total of 1,316 radiomic features were extracted separately from tumor regions, peritumoral regions, and their respective subregions. Delta-radiomic features were further computed to quantify longitudinal changes in tumor heterogeneity between pre- and post-treatment phases. Feature selection involved correlation analysis, univariate Cox regression, followed by LASSO Cox regression to establish the longitudinal CT-based subregional radiomic signature (LS-RS). Its performance was compared against the pretreatment CT-based conventional whole-tumor radiomic signature (PT-RS). A nomogram integrating the LS-RS with clinical risk factors was constructed and validated, and its performance was evaluated by C-index, calibration curve and decision curve. SHapley Additive exPlanations (SHAP) method was employed for model interpretability.
Results: Three tumor-based subregions (TS1, TS2, TS3) and three peritumor-based subregions (PS1, PS2, PS3) were identified. The LS-RS constructed by 2 pre-tumor, 1 pre-TS1, 1 pre-TS3, 1 pre-PS1, 2 post-TS2, 2 post-TS3, 1 delta-PS3, 1 delta-PS2 features outperformed PT-RS in all sets (C-index in training: 0.73 vs 0.68; internal validation: 0.67 vs 0.63; external validation: 0.68 vs 0.62). The final nomogram integrating LS-RS with T-stage and concurrent chemotherapy demonstrated superior performance compared to clinical models (C-index in training: 0.75 vs 0.68; internal: 0.73 vs 0.64; external: 0.70 vs 0.62). The SHAP analysis revealed greater predictive contributions from LS-RS than T-stage and chemotherapy.
Conclusion: The novel nomogram provides accurate LRFS prediction for ESCC patients receiving dCRT, significantly outperforming conventional radiomics model and clinical model. This study introduces the first longitudinal spatiotemporal predictive model integrating pre-/post-treatment tumor and peritumoral subregional radiomics to decode tumor evolutionary trajectories and refine personalized prognostic risk stratification in EC.