| Sumario: | This study aimed to develop and validate 18F-FDG positron emission tomography/computed tomography (PET/CT) radiomics-based models for predicting cervical lymph node metastasis (LNM) and prognosis (overall survival [OS] and disease-free survival [DFS]) in patients with oral squamous cell carcinoma (OSCC). We retrospectively enrolled 130 patients with OSCC who underwent surgery. A total of 107 radiomics features were extracted from the primary tumor. Least absolute shrinkage and selection operator (LASSO) regression and random forest algorithms were applied for feature selection. Multivariable logistic regression (for LNM) and Cox regression (for OS/DFS) models integrating selected radiomics and clinical features were constructed. Model performance was evaluated using receiver operating characteristic (ROC) curves, concordance index (C-index), calibration curves, and decision curve. The LNM diagnostic model, which incorporated three radiomics features (GLCM_Correlation, GLCM_Imc1, and GLCM_JointEnergy) and two clinical features (cT and cN), achieved an area under the curve (AUC) of 0.856 and an accuracy of 81.5 %, outperforming conventional visual PET/CT assessment (accuracy = 74.6 %). For prognosis, the combined radiomic-clinical models for OS and DFS yielded C-indexes of 0.716 and 0.683, respectively, both significantly higher than those of TNM stage (0.630 and 0.594). Both prognostic models successfully stratified patients into high- and low-risk groups with significant survival differences (p < 0.01). PET/CT radiomics significantly improved the diagnostic accuracy for cervical LNM and possessed the potential as a supplementary component for the TNM staging system. PET/CT radiomics possesses considerable potential for clinical translation in lymph node diagnosis and prognosis prediction.
|
|---|