Modifying node-RADS in diagnosing parotid lymph node metastasis of nasopharyngeal carcinoma

This study aimed to establish a modified Node Reporting and Data System (mNode-RADS) for differentiating metastatic from benign parotid lymph nodes (PLNs) in nasopharyngeal carcinoma (NPC).

166 and 60 NPC patients with PLN from two institutions were retrospectively reviewed. The diagnostic performance of initial imaging characteristics for metastatic PLN (PLNM) was compared using the area under the curve (AUC). Two top-performing variables, lesion area and short axis, were used to modify Node-RADS (la-mNode-RADS and sa-mNoder-RADS, respectively). The AUC, sensitivity, specificity, and accuracy were compared among short axis, conventional and modified Node-RADS criteria.

In the internal cohort, the lesion area showed the highest diagnostic performance (AUC = 0.877; cutoff = 60.69 mm²), followed by the short axis (AUC = 0.869; cutoff = 7.5 mm). Both la-mNode-RADS and sa-mNode-RADS achieved higher AUCs than Node-RADS (0.912 or 0.911 vs. 0.870; all p < 0.05). Dichotomized la-mNode-RADS and sa-mNode-RADS (scores 1-2 vs. 3-5) showed greater sensitivity than Node-RADS or short axis (94.8 % vs. 92.7 % vs. 76.0 % vs. 77.1 %; all p < 0.05), with similar specificity (78.7 % vs. 79.8 % vs. 84.3 % vs. 86.5 %; all p > 0.05). Both la-mNode-RADS and sa-mNode-RADS had higher accuracy than Node-RADS (87.0 % or 86.5 % vs. 80.0 %; all p < 0.05). No significant differences were observed between la-mNode-RADS and sa-mNode-RADS. The superior diagnostic performance of short-axis threshold of 7.5 mm and its corresponding mNode-RADS was also confirmed in external cohort.

The user-friendly sa-mNode-RADS shows superior diagnostic performance over traditional Node-RADS in detecting PLNM in NPC.