Intra- and inter-reader agreement of the Node-RADS v1.0 score for axillary lymph node assessment on preoperative breast MRI

Purpose To assess intra- and inter-reader agreement of the Node-RADS v1.0 score for axillary lymph node evaluation in breast cancer (BC) patients undergoing preoperative breast MRI (bMRI), and to explore its patient-level association with pathological nodal involvement. Contrast-enhanced CT was evaluated as a subgroup exploratory analysis. Methods This retrospective study included 100 women with biopsy-proven BC who underwent preoperative bMRI; 25 also underwent contrast-enhanced CT. Four readers (two expert breast radiologists and two junior readers) independently assigned Node-RADS v1.0 scores to the most suspicious axillary lymph node on MRI (n = 100) and CT (n = 25). Thirty bMRI examinations were re-evaluated after a 3-week washout to assess intra-reader agreement. Agreement was assessed using percent agreement and quadratic-weighted Cohen's κ. Exploratory patient-level diagnostic performance for pathological nodal involvement was evaluated using ROC analysis and AUC values. Results Inter-reader agreement on bMRI was moderate among junior readers (κ = 0.59; 95%CI: 0.39–0.79) and almost perfect among expert readers (κ = 0.98; 95%CI: 0.96–1.00). Intra-reader agreement ranged from substantial to almost perfect. Diagnostic performance on bMRI was moderate for most readers (AUC 0.62–0.64), with higher AUC in one junior reader (AUC 0.79). In the clinically selected CT subgroup, AUC values ranged from 0.74 to 0.90. A Node-RADS threshold >2 showed the most balanced sensitivity-specificity trade-off in this cohort. Conclusion Node-RADS v1.0 showed good to excellent reproducibility for axillary lymph node assessment, particularly among expert readers. Diagnostic performance was exploratory and variable , reflecting the patient-level reference standard and the one-node-per-axilla approach. A Node-RADS threshold >2 may represent a pragmatic exploratory cut-off for patient-level risk stratification.