Training AI on Health Data Without Compromising Privacy
RISE develops privacy-preserving AI techniques like federated learning and homomorphic encryption, enabling healthcare AI to learn from sensitive data without exposing patient information.
Healthcare offers enormous potential for AI, but patient privacy creates fundamental constraints. RISE is developing advanced privacy technologies that allow AI to learn from sensitive medical data without ever exposing that data.
The Privacy Challenge
As senior researcher Rickard Brannvall explains: “The laws exist to protect individuals’ privacy… we must comply with them.” Privacy isn’t just a legal requirement - it’s essential for maintaining patient trust in healthcare systems.
Federated Learning
The primary approach is federated learning, where AI models are trained without centralizing sensitive data:
- Organizations jointly train AI models by exchanging model updates rather than raw data
- “Algorithms trained on data held by different organisations without the data leaving their IT systems”
- Each hospital or clinic retains full control of patient information
This means a cancer detection AI could learn from thousands of hospitals across Europe without any patient record ever leaving its home institution.
Homomorphic Encryption
For additional security, RISE explores homomorphic encryption, which allows “calculations on encrypted data” without ever decrypting it. When combined with federated learning, this provides layered protection.
Practical Applications
The techniques enable powerful health AI applications:
- Combining wellness data from mobile phones with healthcare records
- Identifying behavioral changes that predict health issues
- Enabling early intervention before clinical symptoms appear
- Supporting personalized medicine at scale
RISE’s Support Services
RISE helps organizations implement privacy-preserving AI through:
- Platform construction for federated learning
- Advisory services on privacy technology choices
- Testing through the Cyber Range testbed
- Compliance guidance for health data regulations
The Path Forward
Privacy-preserving AI isn’t just about compliance - it’s about unlocking the full potential of healthcare data while maintaining the trust that makes modern medicine possible.
