To Predict Advanced Radiation Xerostomia after Radiotherapy In NPC Patients Based on Radiological Features Extracted from T 2 WI Images
Abstract
Purpose
To explore the value of radiological features extracted from T2-weighted imaging (T 2 WI) images of the parotid gland of patients with nasopharyngeal cancer (NPC) in predicting advanced radiation xerostomia after radiotherapy.
Methods
A retrospective analysis of 123 patients with nasopharyngeal carcinoma (nasopharyngeal carcinoma, NPC) who received radiotherapy from January 2019 to March 2021 was conducted. All patients had MRI scans before and after radiotherapy. Patients were divided into the training set (98) and the test set (25). The ipsilateral parotid (iPG) and contralateral parotid (cPG) were outlined as regions of interest on T 2 WI images (ROI).851 radiographic features, including shape features, first-order intensity histogram (IH) features, statistical matrix (SM) features, and wavelet features, were extracted using the PyRadiomics package.
Results
Of the preprocessing features, 20 features were selected; 19 features and 20 features were selected.The AUC of the pretreated radiology model was 0.902 and 0.740 in the training and test sets, respectively; the AUC of the post-treatment radiology model was 0.761 and 0.701, respectively; and the AUC of the differential radiology model was 0.867 and 0.851, respectively.Pretreatment and post-processing changes in cPG features have high accuracy in predicting late radiation xerostomia.
Conclusion
A significant correlation between these features and advanced radioactive xerostomia was verified by analyzing radiographic features in T 2 WI images of the parotid gland before and after radiotherapy in NPC patients. In particular, pre-and post-processing changes in cPG features have high accuracy in predicting advanced radiation xerostomia. The results show that radiological features based on MRI T2WI images can be used to evaluate and predict advanced radioactive xerostomia after radiotherapy in NPC patients, providing a new predictive tool in the clinic.