Department of Nuclear Medicine, the First Affiliated Hospital of Guangzhou University of Chinese Medicine, Guangzhou, Guangdong 510405, China , nfyyysx@126.com
Abstract: (23 Views)
Background:We aimed to accurately and efficiently evaluate Programmed Death Ligand-1 (PD-L1) expression by relevant radiomic studies of fluoro-18-fluorodeoxyglucose (18F-FDG) Positron Emission Tomography/Computed Tomography (PET/CT) images in nasopharyngeal carcinoma (NPC) patients. Materials and Methods: This retrospective study included 60 untreated NPC patients had PET/CT. Cohorts of training and validation were randomly selected among the patients. The CT and PET radiomic features from the training cohort were obtained and screened, to construct CT, PET and combined models. Finally, verification and comparative analysis were performed. Results: According to the analysis, the maximum Standardized Uptake Value (SUVmax) alone was the standalone predictive indicator of PD-L1 presence level, thus incorporated into the combined model. Among our training cohort, the CT, PET, and combined models’ Area under Curve (AUC) values respectively were 0.837, 0.852, and 0.948, demonstrating excellent discrimination and calibration. However, the combined model had higher AUC values in the cohorts of training and validation, reaching AUCs of 0.948 and 0.802, respectively. Clinical decision curve analysis (DCA) further illustrated Combined model surpassed both the CT and PET models, attaining a benefit threshold probability of more than 5% and a net benefit (NB) of 0.450 at the optimal threshold probability. Conclusion: The combined predictive model based on relevant radiomic studies of PET/CT scans performed better than other models in assessing individualized PD-L1 expression in NPC.