In his DLD Health presentation, Prof. Bernhard Kainz – a computer scientist at FAU Erlangen-Nuremberg and Imperial College London – explores how medical AI is moving beyond simple prediction toward normative learning, synthetic patient data, foundation models, and agentic healthcare AI.
Kainz explains how deep learning shifted medical image analysis from hand-coded algorithms to data-driven systems, enabling real-time analysis of complex modalities such as ultrasound, pathology, and 3D imaging.
He also highlights a major limitation: healthcare cannot label every disease, especially rare or unknown conditions. That’s where normative learning comes in: training AI on “normal” or single-class data so it can detect when a new patient falls outside an expected distribution, similar to how blood-test reference ranges work.
Watch the video to explore this talk in detail.
