ESTRO 2025 - Abstract Book

S1740

Clinical – Upper GI

ESTRO 2025

with locally advanced esophageal squamous cell carcinoma (ESCC) remains a significant clinical challenge. This study aims to develop a radiomics-based model to predict the major pathological response (MPR) of LNs.

Material/Methods: This retrospective study included patients with locally advanced ESCC who underwent NCRT followed by surgery at Sichuan Cancer Hospital. The lymph node tumor regression grade (LN-TRG) was assessed based on the percentage of residual tumor area within the tumor bed of the LNs. LNs with a residual cancer rate of <10% were classified as achieving MPR (LN-MPR), while those with ≥10% were classified as non-MPR (non-MPR). Pre-treatment computed tomography (CT) images were used to delineate radiographically positive LNs, and LN-TRG was assessed for each pathologically confirmed positive LN. These radiographically positive LNs were matched to the pathologically confirmed LNs based on anatomical location. Deep learning and traditional radiomics features were then extracted from all positive LNs. Three models were developed: a deep learning (DL) model, a traditional radiomics (Rad) model, and a combined deep learning and radiomics (DL + Rad) model, to predict LN-MPR. The predictive performance of the models was evaluated using the area under the receiver operating characteristic curve (AUC), and calibration and decision curve analyses were used to assess model performance.

Results: