The role of ultrasound texture analysis in the discrimination of pleomorphic adenoma and Warthin tumor in subjects with well-defined tumor borders.
Authors
Affiliations (3)
Affiliations (3)
- Department of Otorhinolaryngology, Head and Neck Surgery, The Fifth Affiliated Hospital, Southern Medical University, Guangzhou, China.
- Department of Ultrasound Medicine, Beijing Chaoyang Hospital, Capital Medical University, Beijing, China.
- Department of Otorhinolaryngology, Head and Neck Surgery, Beijing Chaoyang Hospital, Capital Medical University, Beijing, China.
Abstract
Pleomorphic adenoma (PA) and Warthin tumor (WT) are the two most common benign parotid tumors, and their distinct management strategies necessitate accurate preoperative discrimination. Therefore, this study aimed to investigate the value of ultrasound texture analysis in differentiating between them. This retrospective study analyzed ultrasound images from patients with pathologically confirmed PA or WT. Following lesion segmentation, texture features were extracted from the regions of interest (ROI). Feature selection was subsequently performed to identify the most discriminative feature subset. Machine learning (ML) models were constructed based on the selected features and evaluated on an independent test set using metrics including accuracy, sensitivity, specificity, and the area under the curve (AUC). Finally, SHapley Additive exPlanations (SHAP) was employed to interpret the model's predictions and quantify the contribution of key features, thereby linking them to sonographic characteristics of the tumors. In addition, the cut-off value of each feature was calculated. Eight key texture features were identified. PA showed more homogeneous and regular patterns, whereas WT appeared more heterogeneous and random. A discriminative model using these features achieved good performance on the test set: AUC 0.958 [95% confidence interval (CI): 0.899-0.996], sensitivity 90.5% (95% CI: 76.9-100.0%), specificity 83.3% (95% CI: 66.7-96.3%). On the independent external validation cohort, the model achieved accuracy 85.4% (95% CI: 73.2-95.1%), sensitivity 72.2% (95% CI: 50.0-92.9%), specificity 95.7% (95% CI: 86.4-100.0%), and AUC 82.4% (95% CI: 67.7-95.7%). This study identified and validated eight texture features that may help distinguish PA from WT in the parotid gland. Integrated into an ML model, these features showed discriminative potential, offering a useful adjunct for preoperative assessment of parotid masses.