Objective:

To investigate the value of radiomics models based on computed tomography (CT) in differentiating parotid pleomorphic adenoma (PA) from Warthin tumor (WT).

Methods:

A total of 28 patients with PA and 25 patients with WT

which were confirmed by pathology

were collected. MaZda was used to extract 5 kinds of radiomics features

including histogram analysis (HA)

gray-level co-occurrence matrix (GLCM)

gray-level run length matrix (GLRLM)

absolute gradient (AG) and autoregressive model (AR)

of tumors in plain CT scan images. Intraclass correlation coefficient (ICC) was used to analyze the radiomics

the characteristics with ICC＞0.75 in PA and WT groups were selected. Then

the feature parameters with statistically significant difference between the two groups were selected and further screened by least absolute shrinkage and selection operator (LASSO) regression analysis. Random forest (RF)

logistic regression (LR) and support vector machine (SVM) classifier models were established

using the feature parameters which were finally selected

and their effectiveness were evaluated using the receiver operating characteristic (ROC) curve.

Results:

Among the features with statistically significant difference between PA and WT

the number of GLCM was the largest. Finally

11 feature parameters were selected as subset to establish three models of RF

LR and SVM

among which RF had the best performance

corresponding accuracy

sensitivity

specificity and area under curve (AUC) were 83.3%

78.6%

88.0% and 0.882

respectively.

Conclusion:

Radiomics models based on CT have good classification efficiency

which can distinguish PA and WT effectively before operation.