DI Europe spoke with Prof. Elmar Kotter, immediate Past-President of the European Society of Medical Imaging Informatics (EuSoMII), about the current and future trends in AI and the content dedicated to the technology at ECR 2023.
As Chair of the eHealth and Informatics Subcommittee of the European Society of Radiology (ESR), you oversaw the AI content at ECR. Which sessions stood out?
AI has become one of the main topics at ECR because it plays an increasingly important role in radiology.
During ECR 2023, 84 sessions were labeled AI and machine learning. That’s a lot! The sessions addressed the three different levels of the European training curriculum. Besides, we had 25 industry sessions in the AI Theatre that attracted crowds of delegates.
Some sessions particularly generated interest from the audience, for example the session on data sharing fueling AI development, where Luis Martí Bonmatí, Laure Fournier, Rick Abramson, Martin Willemink, Peter Van Ooijen and I discussed how important data sharing is to AI development.
He is also Chair of the eHealth and Informatics Subcommittee of the ESR and the immediate Past-President of EuSoMII.
Another session tackled the EUropean Federation for CAncer IMages (EUCAIM) initiative, which aims to foster innovation and deployment of digital technologies in cancer treatment and care.
We also had content on AI for dose optimization and management; quality control; and on ethics and sustainability, to name a few. In my term as EuSoMII President, I initiated a partnership with the European Federation of Radiographers’ Societies, as radiographers will play an important role in the future of AI.
Another really interesting session was the one organized by the European School of Radiology (ESOR), on how to adapt education in radiology. The session was chaired by ECR 2023 President Adrian Brady and Valérie Vilgrain, Head of ESOR, and my talk focussed on how to teach radiologists about AI.
How educated are radiologists when it comes to AI?
It’s very difficult to assess the knowledge of AI radiologists have. Some have very good knowledge, others do not. Most radiologists probably are on the application side, but even then, you need some basic knowledge of AI. As an analogy, you do not need to be a physicist to run an MRI scanner, but you need to understand the basic principles to reduce examination time and be able to explain the problems that may arise when using the technology.
The same goes for AI. You need to know the basic principles and what kind of errors may occur. There is no perfect system. AI, just like other systems, will make errors. You need to check the tool and that takes time. Then also comes the question on how you are going to finance AI.
We know that AI works in radiology, but how are we going to pay for it? That remains a difficult discussion in Europe as most countries don’t have any dedicated reimbursement scheme. To help radiologists get a broader picture of AI, we have just launched the ESR Master Class in AI, which comprises of 70 lectures that are organized into five modules.
What should radiologists do when AI makes a mistake?
Before you buy a solution, you need to test it over a few months to see if it runs well. Once AI is running, you need to monitor its performance. That is something very difficult to implement and most institutions do not do it. And then a time will come when the radiologist does not agree with the tool and decides to override it. Then you need to tell the company about the problem.
There are only very few AI systems that allow us to deliver feedback to their developpers in a direct and easy fashion, like pressing a button. Typically, you need to call the company or write an email. This is time consuming, and we are already short on time.
AI companies should thus offer an option for users to communicate their feedback more easily. Algorithms make errors and and it is important to involve the radiologist in the loop.
There was a session on return on investment (ROI) in AI in the AI Theatre. Why is that an important topic?
It is very hard to show how much money can be saved buying an AI tool. It is easier if there is no radiologist onsite, for example in the emergency department, and AI acts as a safety net, or when you have a general lack of radiologists.
But for other situations, it is an unresolved question and the answer also depends on the country. In the United States, you already have reimbursement for some applications. In Europe, there are ongoing discussions with insurance companies and the payers to know who should pay for AI. One possibility would be to prove that you can treat patients earlier, faster, and thus reduce hospital stay duration; or save time for radiologists.
Most people believe that the radiologist should be the first reader and AI should take a second look. Sometimes AI will detect nodules you have missed in a lung CT scan, but it does not allow you to work faster. If we want to save time, then maybe AI should read the scans first. There are many applications outside diagnostic imaging, to improve for example logistics in radiology and optimize work lists. Regarding reimbursement, another possibility to reimburse AI would be not through the healthcare system or the hospital, but through the pharmaceutical industry. Imagine a pharma company has a new medication and finds an AI company who identifies the pathologies for which this medication works. This would save so much time and money.
Bayer has recently bought Blackford Analysis and more pharmaceutical companies will buy AI developers in the future. Some AI companies are changing their orientation and do not just try to sell to hospitals but the pharmaceutical industry as well. This trend will continue in the next few years.
