Differentiating benign and malignant parotid gland tumors using CT images and machine learning algorithms

Correctly diagnosing parotid gland tumors before surgery is of great significance for clinicians formulating surgical plans, as they are related to patient prognosis. This study evaluated the value of computed tomography (CT) images combined with machine learning algorithms in the differential diagnosis of benign tumors and malignant tumors (BTs and MTs) of the parotid gland. A total of 177 CT images of parotid gland tumors were analyzed in this study, including 99 BT images and 78 MT images. First, the morphological and textural features of the tumor area were extracted, then the least absolute shrinkage and selection operator (Lasso) algorithm was used to reduce the dimensionality of the serially fused features, and finally, the support vector machine (SVM) algorithm was selected to build a classification model. The area under the receiver operating characteristic curve (AUC) was used for the model evaluation. The experimental results demonstrated that the accuracy of the SVM model based on the genetic algorithm (GA-SVM) reached 85.42%, the sensitivity was 72.73%, the specificity was 92.97%, and the AUC was 0.8821. The diagnostic model we proposed could assist doctors in preoperative, noninvasive differential diagnoses, which can better guide the clinical treatment selection.