#007 - Quality healthcare for everyone using AI

October 27 2020/30 min read

We are joined by Stefan Vlachos, Head of the Center for Innovation at Karolinska University Hospital. Stefan has a solid background in engineering and innovation, both from heavy industry as well as the TIME sector. In this interview, we take part in Stefan’s insights into the hurdles the public sector needs to overcome to enable AI solutions to save lives. You might also get some clues to answer the question: How can we make the best healthcare available to everyone everywhere?

This discussion is in English.

Key takeaways from this episode

Why AI innovation projects are different from general innovation projects in the public sector:

  • You need to share data. The legislation within Sweden and the EU limits the possibilities for what can be done.
  • There are also procurement issues. How do you define something that learns? From a procurement perspective, the next iteration of your learning system is generally not considered an upgrade.
  • A huge platform vendor has no specialization in Healthcare, but they know how to make it through a public procurement process. 
  • A lot of the new development is driven by small and innovative startups that don't know how, or that don’t have the stamina, to go through such a public procurement process with us.

What spurs an AI innovation project:

  • It often starts with an engaged clinician. Often someone whose competence bridges the gap between tech and clinician. 
  • You often want to start by doing something that a human is already doing and improve by adding an algorithm, hence we become more efficient.
Computer vision is the most common healthcare innovation project at the moment, but there are still always humans in the loop:

  • We have seen that the field of image recognition is the most active at this point.
  • We have seen no examples of taking the human out of the loop. Quality demands make systems that require this difficult.
Karolinska University Hospital's most interesting AI projects: 

  • We have set up a center for health data (Centrum för hälsodata). Here we have built up a database for all clinical data we have. We have also worked a lot with making it widely accessible. 
  • I-AID Integrated AI Diagnostics. A Vinnova funded project. The point is to showcase a structured and systematic approach to launch AI projects for healthcare. It aims to overcome all the hurdles such as data collection, anonymization, procurement, how to store data etc. It also defines a best practice business model for vendors. 
  • Recently we published a RFI for an AI pathology project. We have made prototypes that show that it is possible to find cancer with the bare minimum cells to analyze.
The hardest obstacles to overcome in governmental innovation projects are:

  • To create a burning platform to change. When the regular market mechanisms are not at play (as intended in the public sector) change is not necessarily driven by new, cost-efficient technology. 
Politicians should think about this to accelerate the adoption of AI in healthcare:

  • Help us be very specific with what data we can share and how we can share it. Today many high risk projects never see the light of day due to the unclear legislation.
  • Use economic incentives to push the healthcare system to work more on preventing illness than on just treatment.

You can find more information on what Stefan and his team are doing here: https://www.karolinska.se/en/innovation