The study of nomogram model based on CT radiomics and clinical features for histological classification of parotid gland tumors

Accurate preoperative identification of pathological types of parotid tumors is essential for the formulation of treatment decisions. The study aims to develop and validate a CT-based radiomics nomogram combining radiomics signature and clinical factors, and evaluate the effectiveness of different models in the classification of parotid gland tumors. A total of 427 patients with parotid gland tumors were randomly divided into a training set and a test set at a ratio of 7:3. Radiomic features were selected using the ANOVA and the LASSO regression. Three-step machine learning models were constructed using three common classifiers (LR, SVM and XGBoost) to classify the parotid gland tumors into four subtypes. The radiomics signature was constructed using the optimal radiomics model, and a radiomics score (Rad-score) was calculated. Clinical data and CT features were evaluated to build a clinical factor model. A radiomics nomogram incorporating the independent clinical factors and Rad-score was constructed. The evaluation of those models’ performance was executed by using receiver operator characteristics (ROC) curves (AUC) and calibration curves, and the clinical usefulness of these models was evaluated by decision curve analysis (DCA). In each step of the three-step procedure, twenty-seven, twelve, and thirteen valuable features were selected, respectively. And the radiomics model based on the LR, SVM, and LR classifiers obtained the highest AUC in differentiating BPGTs from MPGTs (AUC = 0.838), PA from WT PA vs. WT WT vs. BCA, AUC = 0.925). The calibration curve and the DCA demonstrated that the combined nomogram showed superior predictive performance than radiomics model and clinical factor model. The proposed nomogram of radiomics combined with clinical models has high clinical value for the preoperative classification of parotid gland tumors, which might hold promise in assisting clinicians in the exact preoperative diagnosis and formulation of personalized treatment strategy.