Abstract / Summary
This retrospective study investigates whether pre-treatment 18 F-FDG PET/CT radiomics enhances prediction of 5-year relapse risk in diffuse large B-cell lymphoma (DLBCL) patients achieving complete response (CR). Between July 2011 and February 2019, 147 CR-DLBCL patients were divided into recurrence ( n = 39) and non-recurrence ( n = 108) groups. Three prediction models were constructed: Model 1 (clinical/conventional PET parameters), Model 2 (radiomics features), and Model 3 (integrated features). Random forest analysis was used for feature selection and model building. Performance was evaluated via 5-fold cross-validation (internal test) and an external validation cohort ( n = 28, 10 recurrences). The area under the receiver operating characteristic curve (AUC) was calculated, and differences between models were compared using the Delong test. Kaplan–Meier estimates and log-rank tests assessed progression-free survival (PFS). Internal testing showed Model 1 had a mean AUC of 0.579 ± 0.09, while Model 2 achieved 0.733 ± 0.06. In external validation, Model 3 improved AUC from 0.625 (Model 1) to 0.828 and yielded a higher hazard ratio (HR = 10.981) for PFS prediction compared to Model 2 (HR = 3.492). Pre-treatment 18 F-FDG PET/CT radiomics may provide complementary information for predicting 5-year relapse risk in patients with CR-DLBCL.