Federated virtual twins for privacy-preserving personalised outcome prediction of type 2 diabetes treatment
Almost one in ten adults worldwide suffers from type 2 diabetes, making the development of personalized treatment approaches increasingly relevant. However, while healthcare providers are becoming increasingly effective at addressing diabetes risk factors (e.g. diet or exercise), there is still a lack of clear guidelines for the predicted outcomes of specific treatments in individual patients. The overall goal of the „dAIbetes“ project is thus to develop a model for personalized prediction of treatment outcomes in type 2 diabetes while respecting data protection and patient privacy. Using novel methods to privacy-preserving machine learning such as federated learning, the project aims to resolve the contradiction between data protection and big data in cross-national diabetes research.
Project in detail
Diabetes is associated with an increased risk of mortality and morbidity. It is linked to cardiovascular and renal diseases and is associated with an increased rate of retinopathy, neuropathy, and stroke.
The economic cost to society related to diabetes is dramatic when one considers the expenses stemming from medications to manage the disorder and the financial impact of hospitalizations and inability to work. From the patient perspective there are strong negative consequences on the quality of life. The “dAIbetes” project is dedicated to researching the use of “virtual twins” as innovative prognostic tools for personalized disease management. However, since training these models is a data-intensive undertaking and therefore falls directly within the scope of the General Data Protection Regulation (GDPR), the project focuses on data protection-friendly techniques such as “Federated Learning”. This approach enables the effective use of big data while protecting sensitive patient data, which was proven in the recently successfully completed predecessor project FeatureCloud, which SBA contributed its expertise to.
The “dAIbetes” project is therefore building on this technology to develop a federated health data platform for the clinical application of the first internationally trained “federated virtual twin models” with the primary goal of personalized prediction of treatment recommendations and outcomes. The consortium behind “dAIbetes” combines expertise in the areas of artificial intelligence, software development, cybersecurity and diabetes research and treatment.
SBA Research plays an important role within the project, particularly with its expertise in privacy-preserving machine learning and cybersecurity, in particular security of machine learning systems.
SBA Research is among other aspects responsible for:
- The overall data security and privacy aspects within the project
- The technical parts of the data protection impact assessment
- Making the digital twins explainable
- The project is led by the Universität Hamburg, Institute for Computational Systems Biology
- SBA Research is a project partner.
- Project website