Abstract
This study aimed to develop a prognostic model utilizing intratumoral and peritumoral radiomics from simulated localization CT images to predict overall survival (OS) in patients with advanced esophageal cancer, while evaluating its clinical applicability. We conducted a retrospective cohort study involving 151 patients with esophageal cancer who underwent radical radiotherapy between January 2017 and January 2023 (144 men, 7 women). Participants were randomly assigned to a training cohort (n = 105) and a validation cohort (n = 46) at a 7:3 ratio. The primary outcome measured was OS. We extracted 851 radiomic features from the radiotherapy target area of localized CT images. Univariate Cox and LASSO-Cox models were employed to identify features associated with OS. We developed four Cox proportional hazards regression models: a clinical model, a GTV radiomics model combined with the clinical model, a peritumoral radiomics model combined with the clinical model, and a comprehensive radiomics-clinical model. Model performance was assessed using receiver operating characteristic (ROC) curves, Kaplan-Meier survival curves, and nomograms. The median follow-up period was 22 months (range: 6-101). The clinical model exhibited C-index values of 0.540 and 0.590 for predicting OS in the training and validation cohorts, respectively. The GTV radiomics combined with the clinical model demonstrated improved performance with C-index values of 0.753 and 0.677. The peritumoral radiomics combined with the clinical model yielded C-index values of 0.662 and 0.587. The total radiomics-clinical model showed the best predictive capability, with C-index values of 0.762 and 0.704 in the training and validation cohorts. Calibration curves validated the accuracy and clinical relevance of the total radiomics-clinical model, which effectively stratified patient risk categories (p < 0.001). The total radiomics-clinical model, developed from simulated localization CT images, demonstrates a robust ability to predict overall survival (OS) in patients with advanced esophageal cancer. By accurately identifying high- and low-risk patients, this model empowers clinicians to tailor treatment strategies to individual patient profiles. This personalized approach enhances clinical decision-making, enabling more effective allocation of resources and interventions based on the unique prognostic factors of each patient.