A triple-classification radiomics model for the differentiation of pleomorphic adenoma, Warthin tumour, and malignant salivary gland tumours on the basis of diffusion-weighted imaging. Issue 6 (June 2021)

• Record Type: Journal Article
• Authors: Shao, S.; Zheng, N.; Mao, N.; Xue, X.; Cui, J.; Gao, P.; Wang, B.
• Is Part Of: Clinical radiology. Volume 76:Issue 6(2021)
• Page Start: 472.e11
• Page End: 472.e18
• Publication Date: 2021-06
• Subjects: Medical radiology -- Periodicals; Radiotherapy -- Periodicals; Radiotherapy -- Periodicals; Radiology -- Periodicals; Societies, Medical -- Periodicals; Medical radiology; Radiotherapy; Electronic journals; Periodicals; 616.0757
• DOI: 10.1016/j.crad.2020.10.019
• Languages: English
• ISSNs: 0009-9260

Abstract

Abstract : AIM: To develop and validate a triple-classification radiomics model for the preoperative differentiation of pleomorphic adenoma (PA), Warthin tumour (WT), and malignant salivary gland tumour (MSGT) based on diffusion-weighted imaging (DWI). MATERIALS AND METHODS: Data from 217 patients with histopathologically confirmed salivary gland tumours (100 PAs, 68 WTs, and 49 MSGTs) from January 2015 to March 2019 were analysed retrospectively and divided into a training set ( n= 173), and a validation set ( n= 44). A total of 396 radiomic features were extracted from the DWI of all patients. Analysis of variance (ANOVA) and least absolute shrinkage and selection operator (LASSO) regression were used to select radiomic features, which were then constructed using three classification models, namely, logistic regression method (LR), support vector machine (SVM), and K-nearest neighbor (KNN). The diagnostic performance of the radiomics model was quantified by the receiver operating characteristic (ROC) curve and area under the ROC curve (AUC) of the training and validation data sets. RESULTS: The 20 most valuable features were investigated based on the LASSO regression. LR and SVM methods exhibited better diagnostic ability than KNN for multiclass classification. LR and SVM had the best performance and yielded the AUC values of 0.857 and 0.824, respectively, in the training data set and the AUC values of 0.932 and 0.912, respectively, in the validation data set of MSGT diagnosis. CONCLUSION: DWI-based triple-classification radiomics model has predictive value in distinguishing PA, WT, and MSGT, which can be used for preoperative auxiliary diagnosis in clinical practice. Highlights: DWI-based radiomics model can differentiate PA, WT and MSGT. Radiomics features yield highest AUC value for differentiation of MSGT. Logistic regression and support vector machine have the best performance.