The Challenge of External Generalisability: Insights from the Bicentric Validation of a [[sup.68]Ga]Ga-PSMA-11 PET Based Radiomics Signature for Primary Prostate Cancer Characterisation Using Histopathology as Reference

Despite the promising results of radiomics for the characterisation of primary prostate cancer, its use in clinical practice is still limited due to a lack of results validation. In this study, we validated a previously published radiomics signature in a bicentric setting. Then, we also explored the role of the previously identified radiomics features if models are trained on new data. Furthermore, we investigated the performance of radiomics analysis if feature selection and model training are performed in the bicentric cohort. Results show that radiomics models work well when tested in their development settings, but generalise poorly to new datasets, thus warning about the need to interpret with caution results from single-centre studies or studies lacking proper external validation.