Karolinska AEGIS Framework Addresses Post-Market Updates for Adaptive Medical AI Under US-EU Rules
AEGIS offers a structured method for deciding medical AI updates under existing US and EU rules. It prioritizes predefined governance over continuous learning but requires further validation to confirm safety benefits. The framework highlights unresolved questions of accountability and trust in adaptive clinical AI.
The AEGIS system was demonstrated on two models: one detecting sepsis from ICU records and another segmenting brain tumors on MRI. It monitors incoming data streams against an approved change control plan rather than retraining models continuously. This approach aligns with FDA and EU MDR requirements that mandate validation before any update reaches patients, yet leaves implementation details unspecified in current regulations.
EMA will issue guidance incorporating AEGIS-like change control plans by Q4 2027.
Sources (3)
- [1]Primary Source(https://www.nature.com/articles/s41746-026-03290-1)
- [2]Supporting Source(https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device)
- [3]Supporting Source(https://www.ema.europa.eu/en/documents/scientific-guideline/reflection-paper-use-artificial-intelligence-ai-machine-learning-ml-develop-medicinal-products_en.pdf)