Most clinical AI models never make it from research into clinical use.
Friday October 9th, the INDICATE team sat down with David van Laere, member of the INDICATE expert advisory board and CEO of Innocens, and his team at Innocens in Antwerp, Belgium, to look at how INDICATE and Innocens can close that gap together.
What we saw was a relay, not a handoff: people built on each other’s work rather than simply passing responsibility along:
→ INDICATE delivers a federated research pipeline across European intensive care units. It produces models ready for clinical evaluation, without patient data leaving the hospital.
→ Innocens brings data annotation at scale, clinical validation, and a route through the Medical Device Regulation to CE marking and real-time use at the bedside.
We share three challenges:
- Certification. Under the Medical Device Regulation, any software change can trigger a new certification. A modular design that keeps changes separate from the intended use helps, but notified bodies still assess the system as a whole.
- Real-time data. Research runs on a data warehouse, while bedside predictions need streaming HL7 messages. Hospitals need both.
- Retraining. To improve a model after go-live, hospitals have to keep the raw data and preprocessing scripts. That lets the data flow back to the manufacturer for retraining.
For hospitals this means choice. They can join for research infrastructure, clinical AI infrastructure, or both, and turn their data into an asset that delivers value.
Next step: a joint use case. More to follow.
