MRI Dark–Light–Dark Sign and Node Reporting Score: An Integrated Model for Predicting Lymph Node Metastasis in Rectal Cancer

Objectives: To evaluate the diagnostic performance and incremental value of the dark-light-dark (DLD) sign combined with the Node Reporting and Data System (Node-RADS) for the preoperative prediction of lymph node metastasis in rectal cancer. Methods: This retrospective single-center study included 166 patients with histopathologically confirmed rectal adenocarcinoma who underwent preoperative pelvic MRI and subsequent histopathological nodal assessment. Node-RADS score and DLD status were independently assessed by three blinded radiologists, followed by consensus review. Multivariable logistic regression was used to construct a combined DLD + Node-RADS model. Model discrimination was evaluated using receiver operating characteristic analysis and DeLong’s test. Calibration, bootstrap internal validation, and decision curve analyses were performed to assess the model’s reliability and potential clinical utility. Results: Histopathological lymph node metastasis was observed in 122/166 patients (73.5%). Both the Node-RADS score and DLD-negative status were significantly associated with nodal metastasis. In multivariable analysis, the Node-RADS score remained independently associated with lymph node metastasis (adjusted OR, 2.054; 95% CI, 1.455–2.899; p < 0.001), and DLD-negative status was an independent predictor after adjustment for Node-RADS (adjusted OR, 6.635; 95% CI, 2.873–15.319; p < 0.001). The combined model achieved higher discriminatory performance (AUC, 0.795; 95% CI, 0.715–0.875) than Node-RADS alone (AUC, 0.740) and DLD status alone (AUC, 0.713), with significant improvements according to DeLong analysis. Calibration analysis demonstrated good agreement between predicted and observed outcomes (Brier score, 0.142; Hosmer–Lemeshow p = 0.101), and bootstrap internal validation suggested stable performance. Decision curve analysis showed a greater net benefit for the combined model across clinically relevant threshold probabilities. Conclusions: The integration of DLD status with Node-RADS showed potential incremental value for preoperative prediction of lymph node metastasis compared with either approach alone. The combined model demonstrated good calibration, stable internal performance, and potential clinical utility. This imaging-based strategy may provide complementary risk stratification, particularly in indeterminate nodal categories, although external validation is required before routine clinical implementation.