EP-1119: Comparison between T staging and FDG-PET for predicting outcomes of maxillary sinus carcinoma

Purpose or ObjectivePredicting distant metastasis after definitive treatment in nasopharyngeal carcinoma is problematic but critical for the decision of following treatment.This study aims to evaluate simulating CT features in predicting distant metastasis in nasopharyngeal carcinoma treated with definitive intensity modulated radiotherapy (IMRT) with or without systemic chemotherapy.Material and Methods 218 consecutive patients (between June 2011 and June 2015) receiving definitive IMRT were enrolled.The simulating CT images with intravenous contrast were collected.A total of 1039 imaging features, analyzed using both volume-averaged and voxelized methods, were extracted for each patients.Machine-learning methods were used to select best sets to predict distant metastasis in NPCs treated with definitive treatment. ResultsA total of 218 patients were enrolled, with a median age of 51 years old (10-73 years).162 were male and 56 were female.All the patients received definitive IMRT, 151 patients (69%) treated with induction chemotherapy, and 140 (65%) patients with concurrent chemotherapy.With a median follow-up time of 32 months, 2 year distant metastasis free survival was 82%.A total of 1039 features were extracted, then 6 classifiers (SVM support vector machine, Logistic regression, KNN, decision tree, random forest, XGBoost) were used to select predicting sets and XGBoost performed best.The prediction accuracy, sensitivity and specificity were 0.83, 0.50 and 0.85. ConclusionOur results suggested that through a systematic analysis of CT features, imaging radiomics play a role in predicting distant metastasis in nasopharyngeal carcinoma, and should be further explored to help tailoring individually treatment.