ľ�ϸ���Ӱ��

At an MRI conference, he was captivated by the possibilities of artificial intelligence in healthcare. Now, ten years later, Professor Nico van den Berg coordinates the AI Lab that is making medical imaging faster and smarter, using AI. “AI enables us to work faster and more accurately, reducing the strain on patients as well as the workload for healthcare staff.”

Granted, at times, Professor Nico van den Berg, expert in Computational Imaging and coordinator of the , feels a slight unease at the hype surrounding AI. “There’s a big difference between large language models like ChatGPT and medical-use AI. LLMs have been trained on the entire internet, using vast quantities of textual data. Plus, language follows certain rules—grammar, spelling, semantics—so patterns are easier to spot. In the medical field, gathering sufficient training data is far more difficult. Furthermore, the data we do have is often unstructured. You can’t just let AI have a go at some images and expect it to work automatically.”

It’s a disclaimer against overly inflated expectations of AI in healthcare. Nonetheless, Van den Berg saw the very real potential of AI a decade ago at an MRI conference. “It was the annual congress of the International Society of Magnetic Resonance in Medicine. They recognised the potential of AI very early on. I was flabbergasted. I immediately realised: this is going to change healthcare—we need to act on it.”

Medical imaging is one of the most impactful applications of AI in medicine, Van den Berg explains. “Medical images—such as MRIs and CT scans—contain vast amounts of information. The human eye can extract a lot, but AI can often do so faster and more deeply.”

A Van Gogh Painting

After the conference, Van den Berg wasted no time getting started with his team in the Radiotherapy department. “The first thing we developed with AI was an application to convert MRI scans into CT scans. A bit like using AI to turn a photo of a landscape into a Van Gogh painting. That used to be simply impossible. Thanks to this app, a second scan was no longer needed and we were able  to start patients on treatment much sooner.”

Many more AI applications followed, such as automatically outlining at-risk organs on CT and MRI scans for radiotherapy. This saved radiographers thousands of hours of manual work. But according to Van den Berg, the most important example is the use of AI in the , an invention developed in Utrecht that allows you to view the treatment area in real time—enabling a much more targeted therapy.

Anything we can do to relieve the strain on healthcare professionals is vital

Initially, however, that targeted radiotherapy came at a cost: patients spent more time in the machine because the tumour had to be manually outlined by staff. “That’s quite taxing for patients, having to lie in a cramped, very noisy machine.” Van den Berg and his colleagues subsequently trained an AI-assisted system to perform this outlining automatically. “This method combines relevant data with specific domain expertise. The radiographer acts as a supervisor, simply reviewing the outline. That saves a huge amount of time—we were able to reduce the treatment duration from 45 minutes to 30 minutes. That is a lot less discomfort for the patient, as well as valuable time saved for staff.”

According to Van den Berg, this possibility to save time is the key value of AI in healthcare. “AI enables us to work faster and more accurately, reducing the strain on patients as well as the workload for healthcare staff.” We know that over the next ten years, the number of new cancer diagnoses will only rise, as will the amount of available data. Where we used to perform five PET scans a day, we now do thirty. But we don’t have six times as many medical professionals to handle that workload. On the contrary, staffing levels are decreasing significantly. So anything we can do to relieve the strain on healthcare professionals is vital.”

The use of AI isn’t limited to image analysis, either. It can also help reduce the workload of radiologists, Van den Berg explains. “Radiologists spend an enormous amount of time on prep work: reviewing previous reports, checking queries, interpreting data. I envision a future where they have a virtual AI assistant—a kind of orchestrator to quickly gather, filter and present all relevant information in a clear overview. That frees up time for the doctor to practise pure medicine. And to me, that means human interaction—the relationship between patient and healthcare provider.”

Interplay

The AI Labs work closely with industry partners. Van den Berg: “We truly need each other. Yes, we can develop AI methodologies and create lab prototypes. But when it comes to scaling up, commercial support is crucial. There’s far greater engineering capacity in industry than in hospitals. Then we come back into the picture for clinical trials and evaluation. So it’s really an interplay.”

A good example of this is the IMAGINE project, in which the AI Lab for Imaging and Image-Guided Interventions collaborates with other universities, companies like Philips and Elekta, and hospitals to develop the latest imaging and image-guided technologies. The goal is to make certain treatments minimally invasive—or even entirely non-invasive. “This results in less strain on the patient, fewer side effects, and a reduced need for aftercare.”

Shared Challenges

The consolidation of AI-related research at UMC Utrecht under the umbrella of the Health AI Labs means Van den Berg now works more closely with colleagues from other departments who are also developing and applying AI. “Thanks to the AI Lab, I’ve been able to explore to a range of specialisations, such as cardiology and pathology. Interestingly, these fields present very similar challenges: imaging, workflow, data processing. By sharing knowledge, we can build shared infrastructures and train AI collaboratively, resulting in synergy rather than isolated initiatives.”

Which is exactly why Van den Berg says the AI Lab is needed: to ensure good ideas progress beyond one-off pilots. In turn, that also requires greater recognition for those working behind the scenes. “In the Netherlands, we really need to cherish our hospital ‘computer nerds’ more. If we want to keep up with new technological developments and truly reap the benefits, we need to realise how valuable these technical experts are—and invest in them. Because while there’s a lot of talk about AI, someone still actually has to do it.”