Radiation-Induced Xerostomia in Patients with Head and Neck Cancer Treated with Comprehensive Salivary Gland-Sparing Helical Tomotherapy Technique: Based on a Backpropagation Artificial Neural Network

Abstract

Abstract Background The aim of this study was to create a backpropagation artificial neural network (BPANN) model for gauging the risk of developing xerostomia (dry mouth) due to targeted radiotherapy in patients with head and neck cancer (HNC) who underwent comprehensive salivary gland-sparing helical tomotherapy (HT).Methods A total of 246 HNC patients treated with salivary gland-sparing HT were included in this study prospectively from February 2016 to August 2018. The baseline characteristics and clinical data of 222 patients were collected and analyzed. The potential variables included age, sex, tumor type, radiation dose to the salivary glands, and xerostomia questionnaire score. These variables were adjusted using multivariate linear regression. The BPANN model was constructed to predict the likelihood and severity of xerostomia at both 1 and 2 years after radiotherapy. Model evaluation was based on the confusion matrix table and the area under the receiver operating characteristic curve (AUC of the ROC).Results The BPANN model revealed that the risk of radiation-induced xerostomia could be evaluated by evaluating the age, sex, tumor type, and radiation dose applied to specific salivary glands (parotid glands, submandibular glands, oral cavity, and tongue glands). Multivariate analysis indicated that age, sex, and submandibular gland dose were the primary influencing factors for xerostomia. Both prediction models demonstrated strong performance, as reflected in the confusion matrix table and the AUC of ROC curve.Conclusions The BPANN represents a potential and recommended predictive tool for assessing the likelihood of xerostomia induced by salivary gland-sparing helical tomotherapy.Trial registration: This study was registered with the Chinese Clinical Trial Registry (ChiCTR-ONN − 17010597).