Node-RADS in cervical cancer: a multi-reader agreement and diagnostic performance study

To evaluate the agreement and diagnostic performance of the Node Reporting and Data System 1.0 (Node-RADS) for preoperative lymph node staging in cervical cancer across readers with different experience levels.

This retrospective study enrolled 439 consecutive cervical cancer patients who underwent preoperative MRI and lymph node dissection. Target nodes were pre-specified by a most experienced consultant radiologist to unify the assessment objects. Four readers (two senior with 9-11 years of experience, two junior with 3-5 years of experience) independently assigned Node-RADS scores, blinded to histopathology. Inter-reader agreement and between-group agreement were assessed using weighted and Cohen's kappa. Diagnostic performance was evaluated against histopathology as reference standard.

Senior readers achieved near-perfect agreement for Node-RADS scores (k = 0.988), nodal status (k = 0.959) and all individual morphological parameters (k = 0.847-0.967). Junior readers showed moderate to substantial agreement (k = 0.554-0.754). Although junior readers requiring consensus more frequently (7.1% vs. 1.4%, p < 0.01), the between-group agreement for nodal status was substantial (k = 0.783). For nodal metastasis detection, senior readers demonstrated higher diagnostic performance (AUC 0.868; sensitivity 76.4%; specificity 97.3%) compared to junior readers (AUC 0.758; sensitivity 54.6%; specificity 97.0%), and both groups presented favorable diagnostic efficacy overall.

Inter-reader agreement for nodal status was almost perfect among senior readers and substantial among junior readers. Diagnostic performance was slightly better in the senior group, suggesting an influence of reader experience.