Abstract
BACKGROUND: Programmed cell death protein-ligand 1 (PD-L1) expression is an important marker for immunotherapy in locally advanced non-small cell lung cancer (LA-NSCLC). PD-L1 expression has a bi-directional positive feedback relationship with glycolysis status. OBJECTIVE: This study aimed to develop a metabolic habitat model based on (18)F-fluorodeoxyglucose positron emission tomography/computed tomography ((18)F-FDG PET/CT) images to predict PD-L1 expression levels in patients with LA-NSCLC, and to explore relevant biological characteristics. METHODS: We included 219 patients from two independent centers and divided them into the training (n = 175) and testing (n = 44) cohorts. Tumors were segmented into four spatially distinct, biologically similar metabolic habitat subregions using the Otsu method. Radiomic characteristics and metabolic parameters were extracted from each habitat and used to generate multiple predictive models based on the Extra Trees classifier. Data from 1043 patients in The Cancer Genome Atlas database were used to analyze the genes associated with PD-L1 expression in NSCLC. RESULTS: The metabolic habitat model exhibited the highest performance, with area under the curve values of 0.833 and 0.786 in the training and testing cohorts, respectively, outperforming other models. Subregion analysis revealed that high-glycolytic/high-density habitats (PET(High)-CT(High)) exhibited the highest metabolic characteristics, and their spatial distribution correlated positively with PD-L1 expression. Four genes (IFNG, IL2RA, HK3, and MYCN) were associated with PD-L1 expression in glycolysis gene correlation analysis. CONCLUSIONS: The metabolic habitat model based on (18)F-FDG PET/CT enables noninvasive prediction of PD-L1 expression in LA-NSCLC. Its interpretability is enhanced by spatial habitat distribution, thereby advancing its potential for clinical translation.