We caught up with Peter van Ooijen – professor of AI in Radiotherapy and coordinator of the Machine Learning Lab at University Medical Center Groningen, and former president of the European Society of Medical Imaging Informatics (EuSoMII) – for a recent interview, to talk about new technologies and future directions for medical imaging and radiotherapy.
Offering a brief introduction to his current role and background, Peter shared that he studied computer science at the Delft university of technology in the Netherlands before moving into the field of computer graphics, completing his Master’s thesis on radiology data, which led him into the wider arena of medical imaging. “I’m from a technical background,” he said, “but I’ve spent my whole career working within hospitals, focusing on using medical imaging data to improve patient care.”
In his current role at the University Medical Center Groningen, Peter is “on the one hand a professor of AI in Radiotherapy, and on the other hand the machine learning expert at the Data Science Center in Health (DASH)”, he told us. “DASH is trying to improve the level of data science throughout the University Medical Center,” he went on, “so we’re a group of people who work across different departments but dedicate our time to work on that at a central level.” That includes supporting people in a variety of projects, helping locate the right people, and working “a lot on education” which is something highlighted by the newly introduced AI Act as “a very important part now, for medical students, but also for biomedical engineering, and our employees from healthcare professionals to PhD students – everybody needs to learn about AI and how to use it”.
Going on to talk about his voluntary work for EuSoMII, Peter said: “We’re focused on medical imaging, but it’s a bit broader than radiology, and we try to have a very diverse group of people involved, so we’ve got technical and medical people represented to get that interaction going, because we think it’s of vital importance to work together to get the most out of the technology that’s there.” At the moment, the “most promising” technology is AI, he said, “and we need to work together to build responsible AI that also has real clinical value”.
As well as the society’s annual meetings, EuSoMII also runs challenges, according to Peter, “where teams of three people, including at least one clinical and one technical person, work on a challenge for a day – they get the data, they get a platform where they can train models, and they have to train models to perform certain tasks with AI and that clinical data. That helps them start to talk to each other and understand the different perspectives, and then they can learn from each other about the technical parameters and the clinical implications – we also try to get more ethical and legal people involved, as well as patients.”
Sharing some of the research he’s been involved in recently using AI in medical imaging, Peter told us that his PhD students are working on a range of projects, looking at things like tumour segmentation and how to improve quality there, as well as explainable AI, “working on how we can build models in such a way that they’re also acceptable for medical experts who have to use them”. That can include developing attention maps or heatmaps, which help illustrate what the model does and how it produced certain outcomes. “We also need to show the uncertainty of the model, perhaps mapping that into the result, allowing people to better assess how certain that model is of a decision and to understand where they should maybe reconsider the AI decision or make changes to the result,” Peter continued, “so at what point are they not going with the model any more because it becomes too uncertain.”
Research is also being carried out on predictive models, Peter shared, looking at using AI to “try to predict the future”. Whilst that is “a bit further away from clinical practice”, he explained, “we try to predict the outcomes of a specific treatment based on the data we have, what kind of toxicities the patient might develop, and so on”. The idea is that this information could be used to optimise treatment planning, he went on, “so you could even make multiple plans and select the best one based on those predictions, and when treating a patient, you could perhaps even predict the moment at which you need to change the plan to make it more optimal for that patient”.
The challenge, Peter says, is considering how to communicate that process to the user, and how to ensure that information can be trusted. “Nowadays, a lot of AI is still required to have human oversight,” he said, “so there has to be a lot of checking – whether the segmentation was correct, if what the model did was acceptable – for predicting the future that’s quite difficult.” Overcoming these challenges would be worthwhile, however, Peter considered, “as you can really use that to optimise and personalise treatment”.
Commenting on what technologies such as AI and machine learning can offer for the treatment of cancer and other diseases across the globe, Peter said: “If you look at radiotherapy, things like the segmentation of organs at risk and automated plan construction are moving into clinical practice – those things are in use, they’re on the market, you can buy them from different companies, and they perform very well. We all know that healthcare is in a difficult position with more and more people requiring care, and with less and less personnel to provide that care. Technology gives us more data about a patient all the time, but the people to act upon that data are not always available and it will be even worse in the future.”
We need to take action to make healthcare more sustainable, Peter highlighted, “and we believe that technologies like AI can help to partly solve this problem and do the work that needs to be done, by automating steps within the process and allowing people to be faster and more accurate with what they do”. That’s why getting that technology actually implemented and into clinical practice is so important, he continued, “because that’s lagging behind, due to needing to ensure that everything complies with laws and regulations, and that requires some extra effort”.
On what he’s most excited about relating to the use of technology in radiotherapy, Peter told us that “that’s changing all the time”, but that foundation models and large language models (LLMs) offer “so many possibilities” for healthcare. “We also see a lot of possibilities with synthetic data generation,” he went on, “and that’s something we’re still exploring in terms of how we could employ that within clinical practice.” The most exciting thing is “basically AI in itself”, Peter said, “and who would have expected a few years ago that we could do what we can do with LLMs nowadays? That makes our research exciting, and we try new models, we try to apply different things developed with AI into the clinical field, and if you see what is feasible sometimes that’s really very impressive.”
Peter talked about his years of experience in the medical imaging field, stating that the amount of effort that had to be put into doing segmentations with conventional methods was “enormous”. The programming was very difficult, he shared, “and the development of the algorithms, especially when you wanted to segment multiple structures within a dataset, was a lot of work”. Now, we have AI models that can segment all those organs “really fast”, he considered, “and that’s incredible”. It’s impossible to say where medical imaging will be in ten years’ time, Peter said, “but I hope we will be able to do adaptive radiotherapy and optimise the treatment plan at every step of the way, that we can use AI to do that segmentation and everything we need to do in a very short time, so we can offer every patient the most optimal treatment individually”.
Training and education will be central to ensuring that those working in the medical imaging field are prepared to be able to use these models and new technologies in their everyday roles, Peter told us, “and I think training in things like AI is something we have to do on a general scale, and additionally with specialties like radiotherapy we need to do it on a very specific level”. Every role in healthcare will change to some extent as a result of the impact of technology, he noted, and the ways radiotherapists work, the processes that are now in place, “will include more automation, more steps done by a computer”. Whilst AI won’t replace humans in the field, “it’s going to be a tool used to make decisions on treatment and during treatment”, he added.
“I do think that certain skills will disappear,” Peter concluded, “with certain things becoming so good with the use of AI that you won’t need to do them any more yourself – which ones those are going to be is difficult to say, but the example we used of organs at risk segmentation – I think that’s something we won’t have to do any more.” He likened this to the introduction of satellite navigation, saying: “You cannot always rely on it, so you have to be critical about when you use it and the mistakes it could make, but for many people I think that being able to read an actual map is a lost skill. You need to know how to use it, because you need to know how to navigate on Google Maps and search for places; you need to be critical if you see something strange, or if a road it’s telling you to go down is blocked; but you don’t have to know the whole process any more. It’s taken time to get to where we are now with that, and that will be the same in healthcare with technologies like AI.”
We’d like to thank Peter for taking the time to share these insights with us.
