Samenvatting

Artificial Intelligence (AI) offers promising opportunities for innovating the field of medical imaging, yet its practical implementation remains limited. Two important reasons for this are strict legislation and the unavailability of data, partly due to patient privacy. Synthetic data (SD) could offer a solution to the latter. Synthetic data is a machine-learning (ML) technique that learns the distributions and correlations of a real dataset and generates a synthetic dataset with the same characteristics, without containing real data. Synthetic data could facilitate multi-centre collaborations enabling the training of ML-models with data that are now limited. Our aim was to investigate the effects of replacing real data with synthetic data on ML-model performance using SD-generators for the prediction of metastases on 18F-PSMA-1007 PET/CT using a tabular dataset.