Privacy-Safe Synthetic Data Generation

What it is: AIE generates privacy-preserving synthetic biomedical datasets that reproduce the statistical properties and cross-modal correlations of real patient data, enabling data sharing, AI training, and validation without exposing personal data. 

Duration: Typically 4-10 weeks. Targets and sectors: EIC Pathfinder and Transition beneficiaries and Seal of Excellence holders in health, biotechnology, and digital health that are constrained by GDPR, small cohorts, or data-sharing barriers. 

Goal: To unlock the value of restricted or scarce patient data, so innovators can train models, share representative data with partners, and support regulatory submissions.  

Phases and activities: 

• (1) assessment of source data and privacy constraints; 
• (2) selection of generation method (CTGAN, copula-based, or foundation-model approaches); 
• (3) synthetic data generation, including patent-pending multi-modal patient vectors; 
• (4) formal privacy-risk assessment aligned with GDPR Article 35 (DPIA); 
• (5) utility benchmarking of synthetic against original data. Expected outcome: A validated synthetic dataset delivered with documented privacy-risk and utility metrics, ready for sharing or model development. 

Implementation: Delivered remotely by AIE's in-house team; source data processed under secure, GDPR-compliant conditions;synthetic outputs and documentation delivered with a review session. Track record: AIE leads synthetic data generation (WP6) in the DTRIP4H digital-twin research infrastructure, with benchmarks showing 18-40% better correlation preservation than standard methods.