D4.1 Synthetic validation framework

The SYNTHEMA Synthetic Validation Framework (SVF) establishes a rigorous approach for validating synthetic data in clinical research, particularly for diseases like acute myeloid leukaemia (AML) and sickle cell disease (SCD). Designed to ensure data utility while protecting privacy, the SVF supports the safe use of synthetic data in AI healthcare applications. The SVF evaluates synthetic data through three primary dimensions: statistical fidelity, clinical utility, and privacy. Statistical fidelity ensures synthetic data mirrors real data distributions and correlations. Clinical utility assesses the data relevance for insights like survival analysis and mutation frequencies, while privacy metrics evaluate the risk of re-identifying individuals. Data generation approaches were tested, including generative adversarial networks (GANs) and variational autoencoders (VAEs). VAEs showed strong performance in replicating cell characteristics, though rare features require further development. The SVF is essential to SYNTHEMA goal of validating synthetic data for real-world healthcare, ensuring it is accurate, clinically valuable, and privacy compliant. This framework provides a critical foundation for advancing synthetic data use in secure, effective AI-driven medical research.