Integrated tumor habitat and metabolic hot spot displacement analysis on [¹⁸F]FDG PET/CT for predicting treatment response and progression in unresectable locally advanced NSCLC: a two-center competing risk study.

Abstract

To evaluate the predictive value of [¹⁸F]FDG PET/CT derived tumor habitats and metabolic hot spot displacement (MHSD) for treatment response and disease progression in patients with unresectable locally advanced non-small cell lung cancer (LA-NSCLC). This retrospective two-center study included 403 patients with unresectable LA-NSCLC who underwent baseline [¹⁸F]FDG PET/CT before antitumor therapy. The cohort was randomly divided into training (n = 241, 60%), validation (n = 81, 20%), and held-out testing (n = 81, 20%) cohorts. Four metabolic-density habitats were delineated using the Otsu thresholding algorithm to generate a habitat score. MHSD and the edge proximity score (EPS) were quantified to describe the spatial localization of metabolic dominance. The habitat model was compared with whole-tumor radiomics and clinical PET/CT models by the area under the receiver operating characteristic curve (AUC). Independent predictors of radiographically confirmed progression were evaluated using Fine-Gray competing risk regression. The habitat model achieved AUCs of 0.843 (95% confidence interval [CI], 0.789-0.895), 0.822 (95% CI, 0.725-0.904), and 0.847 (95% CI, 0.744-0.935) in the training, validation, and held-out testing cohorts, respectively, and outperformed the radiomics and clinical PET/CT models. In multivariable Fine-Gray analysis, elevated pro-gastrin-releasing peptide (subdistribution hazard ratio [sHR], 1.86; 95% CI, 1.23-2.81; P = 0.004), higher Habitat Score per 0.1-unit increase (sHR, 1.49; 95% CI, 1.31-1.70; P < 0.001), and EPS-SUVmax (sHR, 1.80; 95% CI, 1.13-2.88; P = 0.014) were independent predictors of radiographically confirmed progression. Integrated assessment of tumor habitats and MHSD on [¹⁸F]FDG PET/CT may improve treatment-response prediction and progression-risk stratification in unresectable LA-NSCLC. These imaging biomarkers provide prognostic information beyond conventional radiomics and clinical PET/CT indices, though their specific predictive utility across distinct therapeutic regimens warrants prospective validation.