Pretreatment CT And MR Image Features Improve The Prediction Of Moderate To Severe Late Xerostomia After Radiotherapy In Head And Neck Cancer

Objective: To analyze whether radiomics parameters in planning CT and pre-treatment MR can improve the early prediction of moderate to severe late xerostomia in patients with head and neck cancer receiving radiotherapy,compare the prediction performance of different parameter combinations,search for the optimal parameter combinations and explore its clinical significance.Methods: A retrospective analysis was performed on 49 patients with head and neck tumor who were treated at the First Hospital of China Medical University from January2013 to December 2017.The planning CT and Contrast-enhanced MR within 1 month before radiotherapy were collected.The end point of the study was moderate to severe xerostomia 12 months after radiotherapy.The parotid glands were manually segmented in CT and MR images,The image texture features of bilateral parotid glands were extracted by mathematical algorithm using software Python.Univariate analysis was performed to calculate the correlation between texture features,collected clinical features and moderate to severe late xerostomia.Parameters with statistical significance will be analyzed by Logistic regression,and calculated the area under the receiver-operating characteristic curve(ROC AUC)to evaluate the predictive performance.Compare the predictive performance of combination clinical,CT,MR,CT+MR,clinical + CT,clinical+MR,clinical+CT +MR respectively.Results: A total of 1 clinical parameter,7 CT parameters and 54 MR parameters were selected by univariate analysis,and 7 parameters were selected by multivariate regression analysis: Dmax-contra,P90,GLV,GLN,SAE,LAE and SDE.The ROC-AUC of clinical,CT,MR,clinical +CT,CT+MR and clinical +CT+MR combinations were 0.869,0.762,0.945,0.976,0.786,0.964,respectively.The combination of MR,CT+MR and clinical+CT+MR could improve the AUC value of clinical combination,and the AUC value of clinical +CT-ROC was significantly different from that of clinical +CT+MR-ROC(P=0.016).Conclusion: Clinical parameters combined with both CT and MR parameters can improve the prediction of moderate and severe advanced xerostomia.Adding MR parameters to the combination of clinical+CT parameters can improve the prediction of moderate and severe advanced xerostomia.