Early Prediction of Death Risk in Progressive Nasopharyngeal Carcinoma Using Radiomics Nomogram Based on Clinical Semantic Multi-Parameter Magnetic Resonance Imaging.

Abstract

Aims/Background Patients with recurrent or/and metastatic nasopharyngeal carcinoma (NPC) have a notably low survival rate. Our primary objective in this study is to establish a comprehensive nomogram model based on clinical factors, semantic features, and multi-parameter magnetic resonance imaging (MRI) radiomic features, and to predict the risk of mortality in patients with progressive NPC following intensity-modulated radiation therapy. Methods A retrospective study, including 110 patients with recurrent or/and metastatic NPC who underwent treatment at the Zhejiang Cancer Hospital between June 2012 and December 2016, was conducted. Comprehensive reviews of clinical and pre-treatment MRI data were undertaken. Patients were categorized into two groups based on their mortality status within a 5 year-frame: the non-death group (54 cases) and the death group (56 cases). Radiomic features were extracted from patients' MRIs and the best feature set was selected. Each patient was assigned a radiomic score (Rad-Score). A combined model was constructed using multivariate binary logistic regression, incorporating Rad-Score, semantic features, and clinical data. Receiver operating characteristic (ROC) curves and calibration plots were generated to evaluate the predictive performance of the radiomic feature model, the clinical-semantic feature model, and the combined model for predicting death risk in patients with progressive NPC. A nomogram based on the combined model was constructed. Results Gender, invasion of the carotid sheath by the primary tumour, tumour volume, and progression time showed statistically significant differences between the two groups (p < 0.05). There were statistically significant differences between the three models in the death and non-death groups (p < 0.001). The area under the curve (AUC) value for the radiomic feature model was 0.861 (95% confidence interval [CI]: 0.783-0.920), while the AUC value for the clinical-semantic feature model was 0.797 (95% CI: 0.709-0.868). The combined model demonstrated the highest efficacy for predicting death risk in NPC patients, with an AUC value of 0.904 (95% CI: 0.832-0.952), accuracy of 0.818, sensitivity of 0.857, specificity of 0.870, negative predictive value of 0.778, and positive predictive value of 0.857. Conclusion The combined model incorporating clinical features, semantic features and multi-parameter MRI radiomic features is a highly valuable tool for predicting death risk in patients with progressive NPC, providing a quantitative approach to aiding in early clinical intervention and treatment.