Prediction Of Xerostomia Caused By Radiotherapy For Nasopharyngeal Carcinoma Based On Artificial Intelligence And Radiomics

Objective:Radiation-induced xerostomia remains one of common side effects for radiotherapy（RT）of nasopharyngeal carcinoma（NPC）and severely affects the quality of life.Early prediction of radiation-induced xerostomia is extremely important,but few multimodal radiomics studies mention it.The purpose of this study is to establish an early prediction model of xerostomia caused by radiotherapy in patients with nasopharyngeal carcinoma and guide clinical practice.Methods:Pathologically confirmed NPC patients in our hospital were prospectively enrolled from May 2020 to August 2021.All patients received concurrent chemoradiotherapy.All patients underwent MRI and DWI scans（b=0,500,800s/mm~2）at pre-RT,the fifth(5^th)fraction,the fifteenth(15^th)fraction and post-RT,respectively.Bilateral parotid glands and submandibular glands were delineated as the regions of interest（ROI）on contrast-enhanced T1-weighted imaging（CE-T1WI）,T2-weighted imaging（T2WI）and diffusion-weighted imaging（DWI）.And the mean apparent diffusion coefficient（ADC）value of ROI was measured by MIM software.Radiation therapy oncology group（RTOG）criteria was used to evaluate xerostomia at 3 months post-RT.The clinical basic parameters of the patients and the radiotherapy dosimetry parameters of the ROI were collected.Binary logistic regression analysis was used to screen the predictors of xerostomia.We extract a variety of features from all the images and use matlab 2017 to process the data and train the model.Then,we use variance analysis to select the optimal feature combination and build the prediction model by support vector machine.Receiver-operating characteristic（ROC）curve was used to evaluate predictive performance.Results:A total of 75 patients were included,with an average age of 49 years（23-79 years）.24 were female and 51 were male.The mean ADC values of bilateral parotid glands and bilateral submandibular glands increased gradually with the increase of radiotherapy dose,and the mean ADC values began to increase at the 5^th fraction.The volume of bilateral parotid glands and submandibular glands decreased gradually during radiotherapy,and the degree of reduction was not significant at the 5^th fraction.RTOG evaluation showed that 41 patients（54.7%）with xerostomia in G0-G2 and 34 patients（45.3%）with xerostomia in G3-G4.Binary logistic regression analysis showed that mean ADC changes in bilateral parotid glands and submandibular glands were significant predictors of xerostomia.We built the models containing only radiomics features（models A1-4,B1-4 and C1-4）and the models combining radiomics features,clinical parameters and dosimetry parameters（models A5,B5 and C5）based on MRI at pre-RT,at the 5^th and their delta-radiomics features,respectively.Among them,model C5 which included delta-radiomic features between pre-RT and the 5^th fraction radiotherapy,was the best predictive model,with an area under the receiver operating characteristic（ROC）curve of0.7641,a precision of 0.7444,a sensitivity of 0.7238,and a specificity of 0.8424.Conclusion:The mean ADC value of bilateral parotid glands and submandibular glands in patients with nasopharyngeal carcinoma gradually increased during radiotherapy and began to change independently of volume in the early stage of radiotherapy.The comprehensive prediction model including delta-radiomics features,clinical parameters and dosimetry parameters has the best prediction effect,which is helpful to early identify patients with severe xerostomia after radiotherapy.