Development and temporal validation of an ultrasound-adapted station 104 risk stratification framework for
Yi Zhu1, Yuting Fan1, Yuke Mo1
1Department of Ultrasound, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China (UESTC), Chengdu, China.
Purpose:
To develop and temporally validate LN-104-RADS, an ultrasound-based risk stratification system for station 104 supraclavicular lymph nodes (SCLNs) in esophageal squamous cell carcinoma (ESCC).
Methods:
This retrospective diagnostic study included consecutive patients with histopathologically confirmed ESCC who underwent pre-treatment supraclavicular ultrasound. Patients were assigned to a development cohort (January 2023 to June 2025) or a temporal validation cohort (July 2025 to January 2026). LN-104-RADS was developed by adapting the Node-RADS size-configuration framework to ultrasound assessment of station 104 lymph nodes. Metastatic nodes were defined by histopathological or cytological confirmation, while non-metastatic nodes were defined by negative pathology/cytology without subsequent progression or by radiologic stability, regression, or disappearance during at least 6 months of follow-up. The prespecified cutoff of category ≥ 4 was applied unchanged in the validation cohort. Diagnostic performance, calibration, clinical utility, and interobserver agreement were assessed using ROC analysis, calibration curves, decision curve analysis, and Cohen's kappa.
Results:
The development cohort included 1806 patients, contributing 1832 station 104 lymph node observations, including 263 metastatic and 1569 non-metastatic nodes. Using LN-104-RADS category ≥ 4, the AUC, sensitivity, specificity, and accuracy were 0.968, 95.8%, 97.8%, and 97.5%, respectively. The metastatic rate increased stepwise from 0% in categories 1 and 2 to 4.8% in category 3, 81.6% in category 4, and 100% in category 5. In the temporal validation cohort of 500 patients, LN-104-RADS retained favorable performance, with an AUC of 0.920, sensitivity of 90.9%, specificity of 93.0%, and accuracy of 92.6%. Calibration analysis showed acceptable agreement in the development cohort, although some deviation was observed in the temporal validation cohort. Decision curve analysis demonstrated greater net clinical benefit for LN-104-RADS than the 'treat-all' or 'treat-none' strategies across clinically relevant threshold probabilities. Interobserver agreement for LN-104-RADS categorization was excellent (κ = 0.841).
Conclusion:
LN-104-RADS showed high diagnostic performance, reproducibility, and temporal robustness for ultrasound-based station 104 nodal assessment in ESCC. This framework may support standardized pre-treatment risk stratification and help identify patients who may warrant further diagnostic evaluation for supraclavicular nodal staging.