Multimodal imaging based on MRI to distinguish benign and malignant tongue tumors and T stage

Abstract

Abstract Background: The differentiation of benign and malignant tumors and the stage of malignant tumors is very important to determine the treatment plan and evaluate the prognosis of tumor patients. At present, the application of MRI multimodal imaging to the accurate diagnosis of tongue tumors is not sufficient. Objective: To construct an optimal prediction model based on MRI multimodal imaging to distinguish benign and malignant tongue tumors and T-stage. Methods: This study retrospectively analyzed 124 patients (100 malignant and 24 benign) with tongue tumors who underwent enhanced MRI scans before surgery from January 2017 to December 2022.The surgical tissue was confirmed by pathological examination and was included in the predictive T stage cohort study, which classified T1 and T2 as T1-2 and T3 and T4 as T3-4. The radiomic features were extracted by cross-sectional T2-weighted imaging (T2), weighted diffusion imaging (DWI) and enhanced T1-weighted imaging (CET1).After reduction and selection, support vector machine (SVM) and logistic regression (LR) were used to construct the radiomics model. The clinical model was established by screening independent risk factors using single multifactor analysis. Combined with radiomics and clinical features, a combined model was constructed and a nomogram was constructed. Using ROC curve analysis to evaluate performance compare model, using decision curve analysis (DCA) decisioncurveanalysis, comparative evaluation the clinical value of each model. Results: In the prediction of benign and malignant tongue tumors, the AUC values of the imaging model training set were: CET1 was 0.885, T2 was 0.870, DWI was 0.827, and ALL was 0.993.The test set AUC was 0.720 for CET1, 0.778 for T2, 0.724 for DWI, and 0.793 for ALL. AUC value of clinical omics model: training set 0.885, test set 0.750; AUC value of the columnium: training set 0.889, test set 0.938.In the identification of T stage of tongue cancer, the AUC values of the training set were: CET1 0.815, T2 0.822, DWI 0.883, Clinic 0.566, nomogram 0.932, and the AUC values of the test set were: CET1 is 0.616, T2 is 0.505, DWI is 0.343, Clinic is 0.611, nomogram is 0.646. Conclusion: The nomogram constructed with the combination of multi-modal imaging features and clinical features is better than the single mode or single clinical model in differentiating the benign and malignant and T stage of tongue tumors. The multi-modal imaging model based on MRI can provide a non-invasive and effective aid for clinical decision-making of tongue tumors, and provide a valuable reference for clinicians in individual clinical decision-making system.