Node-RADS on Preoperative MRI Predicts Lymph Node-Driven Survival in Treatment-Naïve Rectal Cancer Patients: A...

RATIONALE AND OBJECTIVES: To evaluate the diagnostic and prognostic utility of the Node Reporting and Data System (Node-RADS) using preoperative MRI.

This was a retrospective, single-center cohort study of 126 consecutive patients with newly diagnosed rectal adenocarcinoma who underwent preoperative pelvic MRI and curative-intent surgery between 2017 and 2023. Regional nodes were graded using Node-RADS. Standard MRI descriptors (T stage, tumor length, and tumor location) and high-risk features-including MRI-detected extramural venous invasion (mrEMVI) and circumferential resection margin (CRM) involvement-were recorded. Serum tumor markers (CEA and CA19-9) obtained proximate to imaging were assessed. Associations with oncologic outcomes (RFS and OS) were examined using multivariable models adjusted for clinicopathologic covariates; nomogram model performance was quantified by discrimination, calibration, and decision-curve analysis. SHapley Additive exPlanations (SHAP) was used to quantify variable contributions and enhance model interpretability.

In the assessment of regional lymph node status, Node-RADS achieved an AUC of 0.861. When a Node-RADS score ≥4 was used as the positivity threshold, the diagnostic accuracy reached 0.841. During a median follow-up of 56.2 months, the Kaplan-Meier estimated 3-year RFS and OS were 85.0% and 88.2%, and the corresponding 5-year RFS and OS were 82.0% and 78.6%, respectively. Multivariable analysis revealed that Node-RADS, age, and CA19-9 were independent predictors of OS, whereas Node-RADS, clinical EMVI, and CA19-9 were independent predictors of RFS. Based on these factors, nomograms for OS and RFS prediction were developed. For OS prediction, the 3-year and 5-year AUCs were 0.825 and 0.831, respectively, with a C-index of 0.821 being observed. For RFS prediction, the 3-year and 5-year AUCs were 0.741 and 0.786, respectively, with a C-index of 0.726 being observed. SHAP analysis ranked Node-RADS as the primary contributor to RFS predictions (mean |ϕ|=0.6) and the secondary contributor to OS predictions (mean |ϕ|=0.4).

Preoperative MRI-based Node-RADS has diagnostic and prognostic utility and may serve as a standardized imaging biomarker for preoperative risk stratification, supporting individualized treatment and surveillance.