The Radiomic Analysis Based On CT For Differentiating Benign And Malignant Parotid Tumors

PurposeTo explore the value of radiomics based on CT in differentiating benign and malignant parotid gland tumors.MethodsThe clinical data and CT manifestations of 313 patients(73 malignant parotid tumors and 240 benign parotid tumors)proved by the pathology after surgery were retrospectively analyzed.Non-contrast and contrast-enhanced CT examinations were performed on all the patients who were randomly divided into a training cohort(n=219)and a validation cohort(n=94)in a ratio of 7:3.The clinical features were analyzed with SPSS 23.0.Radiomics features were extracted by Python software.Radiomics features were selected by Pearson correlation coefficient(PCC)and Relief algorithm to construct a radiomics model with Support vector machine(SVM).Afterwards,integrated model incorporating clinical features was developed.The performance of these models were evaluated and compared using the area under the receiver operating characteristic(ROC)curve(AUC).Results1218 radiomics features were obtained and 15 features were finally selected.And 4 clinical features were selected at the same time.AUC,accuracy,sensitivity,specificity of training cohort and validation cohort in radiomics model were 0.89(95%CI[0.85,0.93]),0.87,0.93,0.85 and0.83(95% CI[0.76,0.87]),0.87,0.93,0.85,respectively.While the AUC of the training and validation cohorts in integrated model increased to0.94(95%CI[0.89,0.97])and 0.90(95%CI[0.86,0.95]),respectively.ConclusionThe radiomics model based on CT showed great potential in differentiating benign and malignant parotid gland tumors,and it can be further improved when combining with clinical features.