[Clinical study of cervical lymph node metastasis in oral tongue squamous carcinoma by a machine learning model based on contrast-enhanced CT radiomics].

To investigate the value of machine learning model based on enhanced CT imaging features and clinical parameters in predicting cervical lymph node metastasis in patients with tongue squamous cell carcinoma (TSCC). A total of 75 patients with TSCC who were treated in the Affiliated Stomatology Hospital of Nanjing Medical University from January 2015 to July 2022 were collected. All patients had complete clinical data, enhanced CT image data and postoperative cervical lymph node pathological examination results. All cases were randomly assigned to the training group (n=60) and the validation group (n=15) in a ratio of 8∶2. A total of 1 833 radiomics features were extracted from the venous phase image data of enhanced CT. Correlation coefficient selection and LASSO method were used for feature selection and dimensionality reduction to select the optimal combination of radiomics features. Multiple machine learning algorithm models(LR, KNN, Random Forest, Extra Trees, XGBoost and LightGBM) were used to predict cervical lymph node metastasis on the selected radiomics and clinical features. The performance of the model was evaluated by receiver operating characteristic(ROC) curve and decision curve analysis(DCA). SPSS 21.0 software package was used for data analysis. After screening and dimensionality reduction, totally 14 optimal feature combinations were obtained, and a variety of prediction models were established based on them. Among them, the KNN model showed a more balanced fitting effect in the training group and the test group, with AUC values of 0.869 and 0.861, respectively. To further improve the efficiency of the model, we integrated imaging features with patient clinical features, and the AUC value of this comprehensive model was increased to 0.893 and 0.880 in the training group and the test group, respectively. The DCA decision curve showed that compared with the simple radiomic model, the image-clinical model with the integration of clinical features showed a higher predictive effect and clinical application value. The prediction model based on enhanced CT image omics features combined with clinical parameters can effectively estimate cervical lymph node metastasis in patients with TSCC. This approach facilitates risk stratification of patients with TSCC and optimizes clinical decisions to improve treatment strategies and patient outcomes.