- Multimodal imaging improved model performance: Combining PET/CT and MRI features consistently outperformed single-modality approaches, with the full model achieving the highest AUCs and the greatest gains in reclassification and discrimination.
- External validation showed moderate to high discrimination: The model maintained performance across two independent cohorts (AUC ~0.82 and 0.89), supporting generalizability within similar academic-center settings.
- PET features contributed most strongly to predictions: Feature importance analyses demonstrated that PET-derived radiomic features were the dominant drivers of model output, with MRI features providing complementary information.
- Automated segmentation was feasible: Deep learning–based prostate segmentation yielded similar diagnostic performance to expert annotation, with ~80% of radiomic features showing good agreement between methods.
- Clinical role remains adjunctive: Despite strong discrimination, the model’s moderate negative predictive value and retrospective design limit its immediate use for biopsy deferral; it should be considered a risk stratification tool requiring prospective validation.
Daily News
Stay up to date with the latest clinical headlines and other information tailored to your specialty.
