Prognostic Impact of 18F-FDG PET/CT Radiomics in Patients with Small Cell Lung Cancer

Purpose: The prognostic value of PET-derived radiomic texture features in SCLC remains limited. This study evaluates whether texture-based radiomic features from ¹⁸ F-FDG PET/CT can predict survival in small cell lung cancer (SCLC) patients. Methods: This retrospective cohort study analyzed 104 SCLC patients undergoing staging ¹⁸ F-FDG PET/CT. PyRadiomics was used to extract 123 primary tumor features. Robust prognostic features were selected via a Bootstrap-LASSO approach. LASSO Cox regression was then employed to construct clinical, radiomics, and combined models, stratifying patients into risk groups to evaluate overall survival. Results: Among 104 patients included in the analysis, Bootstrap-LASSO identified ECOG performance status, age, and metastasis status as stable clinical predictors, while Gray-Level Size Zone Matrix (GLSZM) entropy and Interpolated Minimum Intensity (ImI) emerged as stable radiomic predictors. In multivariate Cox regression, poor ECOG status (HR: 3.375, P  = 0.005), older age (HR: 1.890, P  = 0.014), higher GLSZM entropy (HR: 1.782, P  = 0.035), and lower ImI (HR: 1.880, P  = 0.014) were independently associated with shorter overall survival. The combined model, integrating both clinical and radiomic parameters, achieved the highest prognostic accuracy (C-index: 0.705, P  = 0.0000021), outperforming both the clinical-only (C-index: 0.676, P  = 0.000012) and Radiomic-only (C-index: 0.65, P  = 0.005) models. Subgroup analysis demonstrated that the combined model retained significant predictive power across both limited (P  < 0.001) and extensive stages ( P  = 0.026). Conclusion: By integrating GLSZM entropy and ImI to capture metabolic heterogeneity, the combined model provided consistent risk stratification across the entire disease spectrum. It maintained significant predictive power in both limited and extensive stages.