In the Netherlands, half of university hospitals now have dedicated AI teams—up from none just a few years ago.
And while AI is transforming many industries, healthcare stands out for its potential to directly improve people's lives, from automated procedures to personalized treatments. Across the Dutch healthcare system, it is quickly emerging as a key tool for making healthcare more accessible, efficient and effective.
To understand this transformation, we spoke with five leading experts at different intersections of the field - from hospital AI teams pioneering new implementations to startups pushing technical boundaries to industry veterans shaping policy.
What You'll Learn:
"AI in healthcare isn't new," explains Nicky Hekster, Delegate of the Netherlands AI Coalition, Health & Care, who spent 35 years at IBM, including on Watson Health. "Even in the 90s, we were working on knowledge-based systems. What's different now is the scale, accessibility, quality of tools, and the amount of data available."
"In the medical field, we've been developing statistical models for many years to determine whether certain medications work," adds Greet Vink, Director of Research Development at Erasmus Medical Center (Erasmus MC). "Of course, now AI is completely hyped and every model is called AI, which isn't completely correct. Nevertheless, with the way we can access big, fast data sets and the internal data of patients, we are using more advanced models."
This evolution is clear to see in how some hospitals are approaching AI. Many university medical centres in the Netherlands now have fully dedicated AI teams. For example, Leiden University Medical Center (Leiden UMC) has a seven-person AI team, including machine learning engineers, data engineers, product managers and implementation specialists.
"Within University Hospitals, around half have AI teams at this point," explains Marieke van Buchem, Innovation Manager AI at Leiden UMC. "They all started a couple of years ago at most. Within non-University Hospitals, it's much more difficult to get funding for a dedicated team. You might see one or two people gaining AI expertise, but it's not standard yet."
The journey to establishing these teams reveals an interesting evolution. "About eight years ago, we started with a program called Applied Data Analytics in Medicine (ADAM)," explains Ilse Kant, Team Lead AI for Health at Utrecht Medical Center (UMC Utrecht). "It was about finding out how to move forward with data science in healthcare." This phase led to the creation of their Digital Health department, which now has three focus areas: e-health apps, continuous patient monitoring (in-hospital and at home), and AI.
"In the last few years, we're getting to more of a maturity level. While healthcare might still be behind some other sectors when it comes to AI, we're now making that leap forward," Ilse notes.
At Erasmus MC, the adoption of AI has been accelerated by their modern facilities. "We moved to a new hospital seven years ago, which gave us new opportunities," notes Vink. "From that point, it went fast - patient files, clinical laboratories, pharmacy systems - it's all automated with AI and robotics now."
Dutch hospitals are increasingly working together to avoid duplicate efforts and speed up innovation. A key to this is the Knowledge Network AI Implementation in Healthcare. This brings together over 30 partners, including hospitals and knowledge institutions. "We meet three times a year to exchange knowledge on actual implementation," explains Ilse, who co-leads the network. "Not the newspaper headlines about what's possible with AI, but the actual work needed to get AI to the bedside. We do workshops together to see what others are doing rather than struggling on our own."
This collaborative approach marks a shift in mindset. "Especially in academic hospitals, there's traditionally been a lot of competition," Ilse notes. "But if you're focusing on how to improve healthcare, then we should be able to work together. That's something we're doing more and more in the last couple of years."
The speed of AI evolution is being harnessed by startups too, of course. "In the early days, we started with traditional image processing algorithms, low pass filtering, small feature detections," explains Pascal Wolkotte, Software Group Lead at Vitestro. "But more recently, we’ve discovered just how powerful AI has become. Our models have become more complex and are able to handle larger scales. What's changing is the maturity of all the tool sets - which makes it easier to apply AI into the overall process, from annotation to training to improving the effectiveness of the model."
AI has many diverse applications in healthcare. Our experts highlighted several key areas where it is being implemented and where it is having the most impact in the Netherlands:
In radiology, AI has moved beyond hype to deliver tangible benefits. "Sometimes AI can see certain spots that the human eye will not see, because it can be very precise," says Vink from Erasmus MC.
The impact is particularly noticeable in diagnosis speed. While patients usually have to wait days for radiology results, AI-assisted analysis can deliver insights within hours. But Vink stresses that AI complements, not replaces, human expertise. "Of course, you can get positive and negative results. The human eye always takes another look."
Vitestro demonstrates how AI can revolutionise routine medical procedures. The company has developed Europe's first autonomous blood drawing device. It uses neural networks to interpret sensor data.
"The main AI we use is deep learning neural networks," Wolkotte explains. "Specifically, we use computer vision models for image processing and feature detection, running neural networks directly on the device for real-time inference. Our approach combines traditional computer vision techniques with deep learning models for more complex pattern recognition. The system needs to process multiple sensor inputs simultaneously."
At Leiden UMC, the AI team developed a system to predict which emergency room patients will need to be admitted to the hospital. The system analyses various patient factors, including medical history, underlying conditions and current presenting symptoms, to determine the likelihood of hospital admission.
"We use data that's already collected in the electronic health record from the moment the patient presents at the hospital," van Buchem explains. "While we deliberately chose a simpler machine learning model for reliability and interpretability, the engineering challenge was the real-time data processing pipeline. We needed to aggregate predictions continuously and ensure the system could handle multiple concurrent inputs while maintaining accuracy."
This not only helps with resource planning, it had an unexpected benefit: "The dashboard improved communication and trust between the emergency room and admitting departments," notes van Buchem.
A similar example comes from UMC Utrecht. There, the AI team made a system to predict patient no-shows at outpatient clinics. UMC Utrecht adapted the concept from Erasmus MC's initial work. "The Erasmus Medical Center started this, and we thought okay, we can do this as well," Ilse explains. "We set up our example in a way that we published our code open source and then distributed that knowledge within the Knowledge Network. Then a lot of other people started implementing this model or tweaking and implementing it at their hospitals."
For what AI says are high-risk no-show patients, they call three days in advance. This has greatly reduced no-shows. "After successful pilots, we had statisticians and data scientists from the Julius Center, UMC Utrecht review our model again and suggest improvements before scaling up throughout the whole hospital."
A common theme across our interviews was AI's potential to reduce administrative burden in healthcare. "There's an enormous amount of free text data that's difficult to process and search through," van Buchem explains. "Having better ways to search and access information can save tremendous time and effort."
Vink emphasises how AI tools can improve patient interactions: "AI-powered virtual assistants and toolboxes can pre-analyse test results, allowing doctors to look less at their screens and more at patients. When a patient comes in, you can have blood test results and radiology results together with a pre-analysis. You can show it to patients and say, 'If we compare you with 800 people who came in with similar complaints, we think the causes are this and this.' This gives more time for explaining to patients, instead of going through each individual part."
AI in healthcare raises important ethical and regulatory concerns that require careful navigation.
All of our experts emphasised the critical importance of safety in implementing AI.
"It has to be safe. You cannot afford mistakes - it's health, it's people's lives," emphasises Vink from Erasmus MC. Wolkotte from Vitestro agrees, stressing that when developing AI-driven medical devices or methods, you have to ask yourself: "Would I be happy to have myself or a close relative treated with this?"
The regulatory landscape in Europe is evolving rapidly. Every healthcare company needs to comply with the MDR (Medical Device Regulation) and when it comes to AI, the new EU AI Act. The experts all agree that, though regulation seems burdensome, it is vital. "You want to use things that are evidence-based and high quality," notes Vink.
Hekster from the Dutch AI Coalition provides a broad perspective on what the EU AI Act means for healthcare: "We are today relying on a lot of big tech and medtech vendors from outside Europe. We need to consider how to deal with those applications. The Act protects the EU market and ensures we're not overruled by big tech."
The regulatory environment creates particular challenges for healthcare startups. "They don't have the time or money to bring in regulatory experts," notes Hekster. "That's why we're working with Finland, Germany and Denmark to develop a tool that helps startups and scaleups assess their compliance with the MDR or AI Act."
However, Vitestro's experience shows that regulations don't have to hinder innovation. "The AI Act and MDR give us very good guidance of what we need to do," explains Wolkotte. "If you use the regulations during your design process, then it actually gives you guidance on where to focus. If you see regulation as a separate thing, then it may slow you down, but if you incorporate it into your development process, then it actually helps instead of blocking you."
When it comes to healthcare and AI, data privacy is fundamental to patient trust and ethical implementation. "In terms of privacy and data protection, hospitals are used to working with that, and we have secure means of dealing with data," notes van Buchem.
Companies like Vitestro have developed specific approaches to handle privacy concerns.
"It should not be biased,” notes Vink. “As a hospital, you see many categories of people, and biases may appear." She points to a concrete example in women's health: "All the data, algorithms, and medications where AI has been used to select components - that's mostly done by males. That's why a lot of the resulting medications don't work very well for female patients."
However, Vink sees AI as potentially part of the solution: "If you know this and have better data, then maybe we can use AI to create better medications."
A key ethical issue is who is responsible when AI aids in medical decisions. "Who is responsible when AI makes a mistake?" Vink asks. "Normally, if a mistake is made by a doctor or nurse, like giving the wrong medication, then of course the hospital and staff are responsible. But with AI, it's somewhere in between. It's like self-driving cars - where does the responsibility lie? With the insurance company, the tech company, or the owner?"
"The first challenge is always data," Hekster emphasises. "Opening up databases containing patient data, ensuring proper anonymisation, compliance with legislation and getting patient consent - these are all crucial first steps."
Data quality and relevance are challenges. This is especially true for historical data, which may not reflect current medical practices.
To address these challenges, the Netherlands is developing national initiatives for better data infrastructure. "Data availability remains a big issue in healthcare," notes Ilse. "But initiatives like Health RI and Cumuluz are working to change this. Health RI aims to realise a health research infrastructure for data in the Netherlands, while Cumuluz is a coalition focused on making data available at point of care."
This is particularly important in the Netherlands' distributed healthcare system. "Our whole field is spread out in different organisations that all have their own IT systems that don't communicate well with each other," Ilse explains. "In a few years, I hope we'll see the first examples where we can exchange data at point of care and run AI models that help with tasks like patient triage between hospitals."
Ilse advises referring to the Dutch guidelines for AI in healthcare, developed by the Ministry of Health.
Both hospitals and startups emphasise the importance of choosing the right model architecture for healthcare applications. "In research, people often use very complex techniques," van Buchem notes, "but for clinical practice, it’s a trade-off between performance and practical things like available resources and interpretability. The challenge isn't in implementing the most advanced architecture, but in building robust, reliable systems that healthcare professionals want to work with."
Wolkotte describes Vitestro's approach to AI development: "We have a rigorous, step-by-step approach. We first collect data from a diverse group of patients (to avoid biases). Then we ensure proper annotation by relevant experts. We test models extensively before any patient contact, and after deployment, we maintain constant monitoring."
"We don't update models while devices are in use without rigorous testing," Wolkotte notes. "It's a very controlled process."
Technical development often isn't the biggest hurdle. "The organisational aspects - changing workflows, getting healthcare professionals to adapt their practices - these are usually more challenging than the technical parts," van Buchem explains.
Projects should address genuine healthcare challenges, not implement AI for the sake of it. "It needs to have real value," Hekster emphasises. "And for that, you don't always need complex AI."
Van Buchem reinforces this: "One big point that we've learned is that it's really important to start from a need that is felt by healthcare professionals. That also helps with collaboration, because then healthcare professionals we work with are usually very motivated [when involved from the start]."
Vink adds a practical perspective: "Sometimes startups come to us with technologies, but it doesn't always fulfill a need or solve a problem. For example, for a nurse, it sometimes takes too much time because now she has to plug in more data to get a reliable output."
Regular interaction between technical teams and healthcare professionals is crucial. "Drink coffee with doctors," Hekster suggests. "They know exactly what features matter in medical images or patient data. You need their domain knowledge."
Vink emphasises the importance of co-creation: "Things could be solved more easily when developed in co-creation. Work with doctors, nurses and other staff, and focus on real problems they have. When things are developed in co-creation with neurologists and radiologists, then you’ll get a toolbox which people are actually going to use."
Wolkotte describes how Vitestro implements this: "We have a clinical team within our company, consisting of medical experts who maintain close contact with hospitals to understand their needs. Our team has direct contact with medical experts and AI specialists and tere's almost daily collaboration between the two groups."
UMC Utrecht follows a structured approach with healthcare professionals leading at every step. "We use an innovation funnel with go/no-go gates between different phases - idea and exploration, developing, piloting, testing and moving forward to implementation and evaluation," Ilse explains. "For every project, we make sure that we have at least two medical domain experts that join our project group, and we collaborate with them in every phase."
The medical departments maintain ownership throughout. "There's always a medical department lead involved - that's actually the person we report to and who gives us the assignment," Ilse notes. "Often there's a medical professional that comes to us with the idea, or we hear the idea from another hospital and pitch it to the healthcare professionals."
"A lot of us in healthcare started working on very difficult topics with AI, sort of believing that 'okay, this is a very difficult topic, AI will solve it for me,'" reflects Ilse from UMC Utrecht. "We're becoming better at estimating whether AI is actually going to be a solution for a problem or not."
This evolution in thinking has led to more pragmatic project selection. "This solution [the no-show prediction] is actually more of a low-hanging fruit solution," Ilse explains. "There are, of course, much more fancy things we could do with AI... but this is actually one that made a quick and big impact."
The success of 'simpler' projects demonstrates that impact often comes from addressing manageable challenges rather than trying to solve the most complex problems first - a hallmark of mature AI implementation in healthcare.
Technical capabilities must align with real-world resources. Hekster shares a cautionary tale: His team at IBM developed an AI system that found three times more potential heart failure patients than manual review. However, when they presented it, they found the hospital lacked the capacity to treat any of these additional patients.
"The data science team needs to be diverse and inclusive," Hekster advises. "You need people with different perspectives - ethics specialists, various medical professionals and people with different technical backgrounds. A data scientist working alone or with one doctor isn't enough."
"Large language models are the big thing," van Buchem shares. "Because of ChatGPT, suddenly everyone knows about them. We're getting a huge amount of questions from all different departments wanting to use language models."
While there's a lot of excitement around LLMs, our experts advocate for careful consideration. Hekster urges caution with rapid adoption: "In my view, it's too early to rely on these language models, because they're changing by the week. If you keep changing the application, how can you be sure that it runs in the same way as it used to?"
"We're currently testing auto-scribing tools that can transcribe and summarise medical conversations," van Buchem continues. "The potential is huge, but we need more evidence about error rates and information loss."
Vink sees AI playing a crucial role in advancing personalised medicine, particularly in women's health. "AI can make more precise prescriptions based on your DNA and other features. You get exactly the right medication and dose you need. That would be a big win because many people get too much or the wrong medication."
She emphasised opportunities in women's health: "There are a lot of problems that are not solved satisfactorily. Take birth control - we've had the pill since the '60s and there hasn't been much innovation. The younger generation is less willing to take it because of the risks and what it does to their hormones. AI could help develop new solutions, such as personalised contraceptives.”
A major opportunity lies in enabling more home-based care. "We want people, especially children, to go home as fast as possible after hospital treatment. Because it can be devastating psychologically," Vink explains. "You can send them home with smart devices to measure vital functions, but currently most of the data isn't reliable."
AI could transform home-based care through two key applications. First, making home monitoring data more reliable.
"Instead of investing endlessly in new smart devices, could AI specialists build an interface between the data gathered on existing devices like smartphones and Fitbits - which much of the population already uses - and transform that into reliable medical data," Vink says.
Second, determining which patients can safely recover at home. explains Ilse. "We're working on prediction models to indicate whether a patient will need help or remain stable," explains Ilse. "This helps prioritise who needs monitoring, since we can't monitor everyone remotely."
Van Buchem highlights the promise of models that can combine different types of data: "Combining text, images, audio and genetic data could better mirror how healthcare professionals actually diagnose patients. This area is new, and I think it will be quite some time before you see impact in actual clinical practice, but that's a really interesting technology."
"What would be helpful is the predictive maintenance of different medical devices where AI can help," notes Wolkotte. "And that's more than just medical devices."
"A large part of the world doesn't have access to quality healthcare. Using AI to make quality diagnostics and treatments more accessible—that's where we could have a massive impact."
Hekster shares an example: "In a country like India, with so many patients and doctors perhaps not all at the level of developed Western countries, AI systems could capture all the possible knowledge around different types of oncology. And help many doctors to improve their treatments.”
Our experts had a clear message for AI professionals: healthcare needs you.
"There's a huge need for very smart technical people," van Buchem emphasises. "There are lots of challenges in terms of accessibility and affordability in healthcare, and technical innovation is the way to solve them. There are more and more AI roles within hospitals, so this is a great time to transition into healthcare."
Ilse is emphatic on this point: “You are needed! Especially if you have already worked in a different sector with AI - that knowledge is particularly valuable in healthcare. We recently hired someone from a totally different industry, but they had extensive experience implementing AI. We're so happy they joined because they can teach us things. We need people who have knowledge on how to implement and how to monitor."
For those interested in making the switch, our experts offer several recommendations:
The opportunities are there, and the field needs technical talent. As Hekster puts it: "We have so much capacity. Kickstart AI also shows that there's a lot of talent here in the Netherlands. We need to maintain that talent, keep it within Europe at least, so that we can all benefit."
Note: All the experts emphasised that no amount of reading can replace direct interaction with healthcare professionals to understand the medical context and challenges.
AI in healthcare stands at an exciting juncture. While there are challenges, the bigger hurdles are often in implementation and organisational change within medical institutions. So success requires more than technical skills. It needs strong collaboration with healthcare professionals and a commitment to addressing real needs, not AI for its own sake.
The future of AI in healthcare holds a lot of promise, particularly in making quality healthcare more accessible and reducing administrative burden on healthcare professionals.
The Netherlands, with its strong healthcare system and growing AI ecosystem, is perfectly positioned to lead this transformation - but only if we can attract and retain the brightest minds in AI to solve these crucial challenges.
