Value of radiomic model based on plain CT in distinguishing parotid pleomorphic adenoma from adenolymphoma

CHU Xiangle; LIU Haiyan; SUN Haitao; ZHANG Yingli; WANG Zhifang; HUANG Yonghua

doi:10.20258/j.cnki.1006-9011.2025.09.010

Abstract：Objective To explore the clinical value of machine learning models based on plain CT radiomics characteristics in differentiating parotid pleomorphic adenoma(PA)from adenolymphoma(AL).Methods The clinical and imaging data of 161 patients with parotid gland tumors confirmed by pathology were retrospectively collected,including 77 cases of PA and 84 cases of AL.The patients were randomly divided into the training set and validation set according to the ratio of 7∶3.The regions of interest of plain CT images were sketched after preprocessing and the radiomic features were extracted,the optimal features were screened by various dimensionality reduction methods,and the machine learning models were established.Receiver operat-ing characteristic(ROC)curve was used to evaluate the differential efficacy of each model.DeLong test was utilized for compar-ing the AUCs.Calibration curve and decision curve were used to evaluate the accuracy and clinical usefulness of the models.Results 8 best radiomic features were selected and used to construct LR,KNN,SVM,RF,MLP and XGB machine learning models.All the models had high efficiency in differentiating the two groups of tumors.The AUC of the validation set ranged from 0.892 to 0.915,and there was no significant difference in AUC among all models by DeLong test(P>0.05),the MLP model had the largest AUC,the sensitivity,and specificity of MLP were 91.3%and 80.8%,respectively.Calibration curve showed that each model had good prediction accuracy,and decision curve showed that each model had high clinical net benefit.Conclusion The radiomics model based on plain CT is helpful to distinguish parotid pleomorphic adenoma from adenolymphoma.