From Magnetic Resonance Imaging Radiomics to the Prediction of Regional Nodal Metastasis in Laryngeal Cancer

To investigate the potential of magnetic resonance imaging (MRI) radiomics-based machine learning (ML) models in predicting local lymph node metastasis (LNM) at initial diagnosis in patients with laryngeal squamous cell carcinoma (LSCC).

This retrospective single-center study included 192 patients with pathologically confirmed LSCC who underwent preoperative contrast-enhanced T1-weighted neck MRI followed by curative surgical resection and lymphadenectomy. Tumor volumes were manually segmented, and 107 standardized radiomic features, along with clinical variables, were extracted. After collinearity reduction and feature selection, the most predictive features were used to train a random forest classifier. Model robustness was assessed through repeated cross-validation, and diagnostic performance was evaluated on an independent test set using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy.

LNM was present in 78 patients (40.6%). T stage and supraglottic involvement showed significant associations with LNM (P<0.001 and P=0.020, respectively). Feature selection consistently identified 6 radiomic features, with "Sphericity" emerging as the most predictive across all runs. The final radiomics model achieved a mean AUC of 0.90 (95% CI: 0.85-0.94), with sensitivities of 0.96, specificities of 0.82, and an accuracy of 0.87.

MRI-based radiomics demonstrates potential for the preoperative prediction of regional LNM in laryngeal cancer by capturing quantitative imaging features related to tumor heterogeneity, shape, and morphology. However, these findings should be interpreted with caution, given the retrospective single-center design and lack of external validation, and require confirmation in larger multicenter studies.