Data are considered the new gold in healthcare. However, making use of the full potential of the existing data is related to various obstacles, mainly due to intellectual property, privacy, and legal issues. Therefore, new approaches are needed which support usage of distributed data without requiring to transfer huge amounts of sensitive data.
We are developing solutions which support privacy-preserving technologies. Depending on the respective scenario, different technologies are applied.
Privacy-Preserving Record Linkage
Privacy-Preserving Record Linkage (PPRL) concerns the linking of different datasets, without disclosing the participant’s identity. Our European Patient Identity (EUPID) Services have been designed to facilitate secondary use of datasets in biomedical research and healthcare by addressing the following major requirements:
- prevent duplicate registration of patients
- avoid creating a transparent universal patient ID
- provide distinct pseudonyms for patients in different contexts
- preserve the possibility for re-identification by a trusted third party
- keep a protected link between the different pseudonyms in the background
- support the creation of merged datasets for secondary use
The EUPID Services are currently used primarily in rare disease applications, including paediatric oncology infrastructures. Additionally, they are a core component of the virtual platform, connecting rare disease research across Europe, as currently developed in the European Joint Programme.
Federated Learning
Federated learning has gained a great momentum since 2016 based an algorithm developed by Google AI researchers, where deep learning networks could be trained on multiple end-users' smart phones. We have applied these algorithms in the health care setting and developed algorithms which can handle huge datasets from different healthcare providers, without the need to pool these sensible datasets in a central data pool.
Privacy-Preserving Research Infrastructures
We are developing infrastructures which enable different types of privacy-preserving AI, including PPRL and federated learning. Additionally, together with our partners, we are also investigating other privacy-preserving technologies, including Secure Multi-Party Computation (SMPC), synthetic data generation, homomorphic encryption, etc.
Selected Projects
| ||||||||||||||||||||||||||
