PET/MR for predicting extranodal extension of head and neck cancer.

Objective: To analyze the diagnostic accuracy of multiparametric FDG-PET/MR in identifying pathologic extranodal extension (pENE) of lymph node metastases (LNM) in head and neck squamous cell carcinoma (HNSCC) patients.

Methods: Retrospective analysis of 57 HNSCC patients who underwent preoperative FDG-PET/MR imaging. PET parameters of LNM SUVmax and MTV, lymph node size as well as MR parameters flare sign, shaggy margin sign and vanishing border sign were analyzed. Histopathological assessment of neck dissection specimens served as standard of reference.

Results: A logistic regression model consisting of lymph node size (p = 0.029), shaggy margin sign (p = 0.031) and MTV (p = 0.035) proved that all three parameters significantly contributed to the prediction of pENE (χ²(3) = 54.23, p < 0.001). A second model without the reader-dependent parameter shaggy margin sign yielded similar results (χ²(2) = 45.36, p < 0.001), with every increase in lymph node size (p = 0.006) by 1 mm increasing the likelihood of pENE by a factor of 1.41 (95%-CI[1.11, 1.81]), and every increase in MTV (p = 0.023) by 1 cm3 increasing the likelihood of pENE by a factor of 1.64 (95%-CI[1.07, 2.50]). This model yielded an accuracy of 94.7% (95%-CI [85.4, 98.9]) for predicting pENE, with a specificity of 97.3% (95%-CI [85.8, 99.9]) and a sensitivity of 90.0% (95%-CI [68.3, 98.8]). Internal validation using a test dataset confirmed high accuracy of this model.

Conclusions: PET/MR-based multivariate binomial logistic regression models consisting of MTV, lymph node size and/or shaggy lymph node margins predict pENE with high accuracy.