The value of Synthetic MRI in discriminating metastatic and non-metastatic lymph nodes in head and neck squamous cell carcinoma, compared with DWI and subjective experience.

OBJECTIVES: To explore the role of Synthetic MRI (SyMRI) histogram parameters in differentiating metastatic from non-metastatic cervical lymph nodes (LNs) in head and neck squamous cell carcinoma (HNSCC) patients, and construct a practical model.

METHODS: A total of 149 pathologically confirmed LNs (metastatic LNs: 58, non-metastatic LNs: 91) were included in the study. LNs were stratified and randomly divided into a training set and an independent validation set in a ratio of 7:3. Histogram parameters derived from SyMRI, ADC values, and the short and long diameters of each LN were obtained. Significantly different parameters between metastatic and non-metastatic LNs were selected in the training set, and logistic regression analysis was adopted to construct different models. ROC analysis and AUC were performed to assess the diagnostic performance of different models and subjective analysis.

RESULTS: The AUCs of the three models were 0.882 (SyMRI_model), 0.755 (DWI), and 0.952 (Combined_model) in the validation set. The Combined_model, constructed based on SyMRI, ADC values, and size, had the highest diagnostic potency in both training and validation sets, with an accuracy of 0.905 and 0.864 in the two sets, respectively. The diagnostic performance of the Combined_model was superior to multi-radiologists' subjective experience, not only in LNs from validation set (AUC: 0.952 vs. 0.705 ∼ 0.845) but also in the cohort of sub-centimeter LNs (AUC: 0.878 vs. 0.429 ∼ 0.628) (all P < 0.001).

CONCLUSION: Histogram parameters derived from SyMRI are feasible in discriminating metastatic from non-metastatic cervical LNs in HNSCC, and the diagnostic efficacy is optimal when combined with DWI and size.