Challenges of AI in healthcare

What is the biggest challenge of using AI in healthcare?

RAIVO KOLDE: I think the biggest problem is still the data, because every AI system is only as good as the data that it’s trained on. And there are a few persistent problems with medical data that are really hard to solve, even though we have large medical data sets around already from all over the world.

PETER VAN OOIJEN: I think that one of the biggest challenges of using AI in healthcare is the actual clinical implementation. And that has one aspect of acceptance. Users have to accept it, the medical doctor. But also the patient has to trust the system, has to be able to use it, and also, the integration into the clinical workflow. It’s a tool. So it has to be a tool that is easily integrated and that fits within the normal work of the medical doctor.

ANNA LEONTJEVA: The first challenge is, of course, around machine learning models, lack of transparency in them, and building trust between practitioners and model scores. The second challenge is around ethical biases in those models, how you build the model that doesn’t propagate them further.

ERIK RANSCHAERT: For me, there’s no doubt that the main challenge is to gain trust from all the users. The users have to be convinced that the algorithm is accurate, that it’s working fine, and that they can rely on the results.

ANGEL ALBERRICH-BAYYARI: The biggest challenge we have nowadays of using AI in health care is a challenge of integration. So once the algorithms are trained, I would say that the biggest efforts go for a smooth and seamless integration into clinical routine.

TANZILA MUKHTAR: My personal concern is only with respect to the data security with respect to no data could be mishandled, or there’s always a threat of identity theft. And I think there’s a dire need that, whenever a lot of patient data is involved, the different levels of security have to be strengthened.

RENATO CUOCOLO: So the biggest challenges in the implementation of AI in clinical practice are, first of all, from a legal and liability point of view. It is still very unclear who should take responsibility for errors made while using AI tools in clinical practice. And then there is the question of trust, which is still has to be built between the practitioners, the end users, the patients, and the systems in and of themselves, as the AI tools often are very hard to interpret in their actions both for the end user but also for the developers. So that’s the big problem for now.