Radiomics-based prediction of radiation-induced hypothyroidism in nasopharyngeal cancer patients

Abstract

When planning radiation therapy, late-effect complications due to radiotherapy should be considered. One of the most common complications of head and neck radiotherapy is hypothyroidism. Although clinical and dosimetry data are usually used to assess the risk of hypothyroidism after radiation for nasopharyngeal cancer, the outcome is still unsatisfactory. Medical imaging can provide additional information and increase prediction accuracy. The aim of this study was to predict hypothyroidism in patients with nasopharyngeal cancer using CT radiomics combined with clinical and dosimetric data. The study included 220 participants who were diagnosed with hypothyroidism within 2 years after radiotherapy. Manual segmentation covered the thyroid gland, and feature extractions were performed from pretreatment CT images. All radiomics features were analyzed with clinical and dosimetry information, and the model was constructed using logistic regression, random forest, and gradient boosting. In addition to the radiomics model, conventional, and combined models were built based on the tree-based predictive algorithms. The findings of the study demonstrated that the combined model had the highest validation performance, as indicated by AUCs of 0.80 ± 0.06 and 0.81 ± 0.06 in logistic regression and random forest, respectively, which were greater than the conventional mode with the AUCs of 0.68 ± 0.07 and 0.71 ± 0.06 (p-value < 0.05). The combined model used in this study used radiomics features, with the majority of these features coming from texture-based classes and filtered-based classes, while the important clinical and dose factors were bilateral neck metastasis, pretreatment TSH level, age, TR V40, and TR mean. In conclusion, the combination of CT radiomics with clinical and dose information can predict the RIH in nasopharyngeal cancers and significantly improve the performance of prediction models compared to the conventional method. We contend that pretreatment thyroid images contain valuable information that can be used to predict the risk of hypothyroidism after nasopharyngeal radiotherapy.