Development and Validation of CT-Based Dose-Volume-Radiomics Nomogram for Radiation Induced Hypothyroidism in Nasopharyngeal Carcinoma

Abstract

Several studies reported hypothyroidism occurred in 40-50% of patients who were treated with neck irradiation. Post-radiation hypothyroidism impairs quality of life, increases the risk of cardiac complications, and requires lifelong thyroxine replacement in affected patients. At present, radiation dose-volume constraints of thyroid gland are used to predict thyroid function outcomes in patients with nasopharyngeal carcinoma (NPC). However, it is limited by (a) inferior predictive power, (b) a lack of analyzing individualized thyroid characteristics as a categoriad to predict radiation induced hypothyroidism (RIHT). In this study, we firstly developed and validated CT-based dose-volume-radiomics nomogram to predict RIHT in patients with NPC. A total of 451 NPC patients who underwent definitive radiotherapy were randomly assigned into the training (n = 338) and validation set (n = 113) in a 3:1 ratio. Dose-volume parameters, including the thyroid volume, mean dose (Dmean), percentage of the volume that received xGy of radiation (Vx), and the absolute volume that was spared from xGy of radiation (Vsx), were collected from radiotherapy planning databases. We defined primary hypothyroidism as an elevated TSH serum level (> 4.94 mIU/L) in combination with a normal or low serum FT4 level, regardless of symptoms. 1316 CT radiomic features were extracted and selected to construct the radiomics signature (RS). A CT-based nomogram was established by integrating clinical factors, dose-volume parameters and radiomics signature in training set and was tested in validation set. With a median follow-up period of 68 months, 301 (66.7%) patients developed RIHT. Compared with other dose-volume parameters including thyroid volume, V30, V50, Dmean, Vs45, Vs50, the thyroid volume spared from 60Gy (Vs60) had best power to predict RIHT. The radiomics signature constructed by 8 selected radiomic features showed better prognostic performance than Vs60 for predicting RIHT in training set (RIHT vs. Vs60, C-index: 0.69 vs. 0.58) and internal validation set (C-index: 0.65 vs. 0.55). Patients were stratified into high- and low-risk groups by median radiomic signature. Patients in high-risk group had higher rate of RIHT than patients in low-risk group (training set:61% vs.39%, P<0.05; validation set: 73% vs.32%, P<0.05). The nomogram established by integrating radiomics signature with Vs60 showed optimal prognostic performance with C-index of 0.71 in training, 0.66 in validation set. Calibration curves showed good agreement. CT-based dose-volume-radiomics nomogram provided an excellent prognostic tool for predict incidence rate of RITH in patients with NPC received definitive radiotherapy.