Node-RADS v1.0 on chest CT for lung cancer: Reproducibility and diagnostic performance

Purpose To evaluate intra- and inter-reader agreement of Node-RADS v1.0 for mediastinal lymph node assessment on chest CT in patients with stage I–III non–small cell lung cancer (NSCLC) and to assess its diagnostic performance across radiologists with different levels of expertise. Methods This retrospective, single-center study included 46 patients (mean age, 71 ± 8 years; 38 adenocarcinomas, 8 squamous cell carcinomas) with 158 pathologically confirmed mediastinal lymph nodes (22 malignant, 136 benign). Four radiologists (two experts, two juniors) independently assigned Node-RADS scores and descriptors (“size” and “configuration”) in two sessions, three weeks apart. Agreement was calculated using Gwet’s AC2 statistics. Diagnostic performance was assessed by ROC analysis; sensitivity, specificity, and predictive values were calculated at a Node-RADS score ≥ 3 threshold. Results Inter-reader agreement for Node-RADS scores was almost perfect for experts (AC2 = 0.97; 95%CI: 0.96–0.99) and juniors (AC2 = 0.95; 95%CI: 0.93–0.97). Intra-reader agreement AC2 values ranged from 0.95 to 0.99. Descriptor agreement was similarly high (AC2 ≥ 0.85). ROC AUCs ranged from 0.71 to 0.76 for experts and 0.68–0.84 for juniors. At the ≥ 3 threshold, specificity and negative predictive value were consistently ≥ 90%, while sensitivity remained limited (<64%). Conclusions Node-RADS v1.0 demonstrated excellent reproducibility among radiologists with different levels of expertise for mediastinal lymph node assessment on chest CT in NSCLC. The consistently high specificity and negative predictive value support its role as a standardized framework for structured lymph node reporting and training, while the limited sensitivity underscores the need for complementary diagnostic tools.