The goal of the session ‘Artificial Intelligence in medicine development - without the hot air’ was to dig into the real opportunities for AI for medicine development, with machine learning as a specific example of current use in clinical trials.
Speakers at the session were:
- Dr. Pawel Widera - Research Associate, Interdisciplinary Computing and Complex BioSystems group at Newcastle University, UK
- Prof. Bram van Ginneken - Professor of Medical Image Analysis at Radboud University Medical Center, the Netherlands; chair of the Diagnostic Image Analysis Group
- Prof. dr. Bert Leufkens, Professor of Pharmaceutical Policy and Regulatory Science at Utrecht University, former Chair of the College ter Beoordeling van Geneesmiddelen (Medicines Evaluation Board), and former CHMP member at the EMA.
During the session it was discussed that machine learning will provide many opportunities, but that better access to and use of data will be the actual challenge if one wants to capture the full potential value of AI methodologies. How will we organise access to data in the case of medical data? Who actually owns and protects the data? These questions will become more and more urgent in the future.
Additionally, there is a contentious debate around the role of doctors: will doctors become data clerks and machine operators? Or will AI create space for real human connections, a doctor who listens and a patient who is heard? AI can transform everything doctors do (note taking, scans, diagnosis, probably even treatment, cutting down the cost of medicine), but what the end-result of this transformation will be is still unclear.
AI tools and techniques will provide new challenges for regulatory authorities; they will have to assess tools and techniques of a totally different nature than the current state of affairs. At the same time, AI may challenge the boundaries between different regulatory bodies or uncover gaps that are not currently being addressed.
After the presentation there was a lively discussion with the audience. Topics discussed were the need for healthy (non-patient) data for AI algorithms to function properly, which is often not available. Furthermore, the issue of harmonisation of unstructured data was discussed, which is an area that is otherwise challenging and requires further work for its improvement.