Development and validation of prognostic models based on F-FDG PET radiomics, metabolic parameters, and clinical factors for elderly DLBCL patients.

This study aimed to assess the predictive value of baseline F-FDG PET radiomics features, metabolic parameters, and clinical factors for PFS and OS in elderly DLBCL patients. Using LASSO COX regression, we derived Radscore from PET radiomics features. We constructed and externally validated prognostic models, evaluating their performance through various metrics. From 341 training set patients and 83 external validation set patients revealed significant correlations between PET radiomics features and survival outcomes. Multivariate COX analysis identified associations of radiomics features (Radscore), metabolic parameters (TMTV, Dmax), and clinical factors (ECOG PS, hemoglobin level) with PFS and OS. In external validation, the combined model incorporating radiomic features, metabolic parameters, and clinical factors showed superior predictive performance for PFS and OS compared to other models. The combined model had higher C-index values for both PFS and OS, and its td-ROC curves exhibited significantly higher AUCs. Calibration curves demonstrated good consistency, and DCA revealed a higher net benefit for the combined model. In conclusion, the combined model that incorporated F-FDG PET radiomics features, metabolic parameters, and clinical factors demonstrated superior prognostic predictive ability, providing a useful tool for personalized treatment decisions in elderly DLBCL patients.