黑料福利网

Imagine this: you develop skin cancer on your face. The tumor has to be removed. That leaves a wound that needs reconstruction. For some patients, achieving the best possible cosmetic result is most important. For others, a fast recovery or minimizing additional surgeries matters most.

This is exactly where plastic surgeon wants to apply AI in the future. Not to replace the physician鈥檚 decision, but to support the conversation between patient and doctor. Which treatment makes medical sense? And which outcome aligns with the patient鈥檚 desired quality of life?

Making your own choices

For , this is an opportunity to apply his knowledge in practice. 鈥淚f we collect enough data, we can show future patients鈥攖hrough different models鈥攚hat their face might look like after surgery. That way, they can help decide on their treatment. It won鈥檛 be a black box for them anymore.鈥�

The ultimate goal is to present patients with two types of predictions: first, a prediction of satisfaction for a given treatment choice, based on demographic characteristics, patient preferences, wound location, etc. 鈥淲e call this the satisfaction model,鈥� d鈥橦ondt explains. In addition, there will be a visual prediction of the face, generated by an AI model trained on hundreds of photos of previous surgeries and their outcomes.

Patients currently undergoing treatment won鈥檛 yet see a ready-made AI tool on the desk鈥攁nd that鈥檚 exactly the point. 鈥淩ight now, patients mainly notice that they are providing data,鈥� Hoogbergen says. 鈥淭hey fill out questionnaires, have photos taken, and give consent. We are focused on collecting the right data. That鈥檚 necessary to make reliable predictions later.鈥�

Laying the groundwork

AI often sounds like something magical: a smart system that produces answers in seconds. But before such a system can safely and reliably assist in healthcare, a lot of less glamorous work has to be done. At the moment, Catharina Hospital is essentially laying the groundwork: collecting data, organizing processes, and ensuring patient information is securely and usefully recorded.

Over the past year, d鈥橦ondt worked with Industrial Design master鈥檚 student Blom van der Toom to fully map out the care process for patients with facial skin cancer, identifying where and how data can be collected and used. When does the patient receive information? When is consent requested? When are photos taken? When do they complete questionnaires? And how do we ensure all that information ends up in the right place?

鈥淵ou have to get that right,鈥� Hoogbergen says. 鈥淵ou don鈥檛 want healthcare providers constantly figuring out which form is needed or which step comes next. The process must be set up so that the patient is well informed and we can collect reliable data in a safe way.鈥�

This data includes medical images. Patients are photographed at fixed points in time: before the procedure, after reconstruction, and during follow-up. They also complete the FACE-Q Skin Cancer questionnaire, which measures how patients experience their appearance, recovery, and quality of life.

The patient鈥檚 voice

For Hoogbergen, that is the core of the project鈥攏ot the technology itself, but the patient鈥檚 voice. 鈥淎s a physician, you鈥檙e always influenced by your training, your experience, and your own ideas of what a good result is,鈥� he says. 鈥淏ut what I consider a good outcome may not be what the patient finds most important.鈥�

This is also a crucial element for d鈥橦ondt. 鈥淵ou want patients to stand behind their own treatment process.鈥� He describes the project as 鈥渁 great combination of rigorous problem modeling and providing patients with useful information.鈥� He attended multiple conversations between doctors and patients, gaining insight into how decisions are currently made, what information is missing, and how better data collection could improve that.

Reducing anxiety

鈥淵ou want to take away people鈥檚 fears and reassure them that things will turn out well,鈥� d鈥橦ondt explains. Hoogbergen adds his own perspective, recalling patients who, by his technical standards, had excellent results but were still dissatisfied. And others for whom he was critical of the outcome, while they were actually very happy because they could return to what mattered most to them.

鈥淭hat鈥檚 when I thought: wait a minute. How can my assessment differ so much from the patient鈥檚 experience? I need to do something with that.鈥�

Since then, Hoogbergen has focused on outcome measurements in plastic surgery. This goes beyond medical results to include what the treatment means for a patient鈥檚 daily life. Can they socialize again? Do they feel comfortable with their appearance? Was the burden of treatment worth it?

Learning from previous patients

The next step is for AI to learn from previous patients by combining photos, medical data, and questionnaire outcomes. In the future, this could allow the system to predict which treatment offers each patient the highest likelihood of achieving an outcome aligned with their preferences.

But the project isn鈥檛 there yet. 鈥淲e鈥檙e really at the beginning,鈥� Hoogbergen says. 鈥淔irst, you need a strong AI system that can predict what the result of a reconstruction might look like. Then you have to link that prediction to questionnaire outcomes. Only then can you truly determine which treatment gives each patient the best chance at their desired quality of life.鈥�

Testing with eyebrows

To build such an AI system, it must first be tested with data similar to the still-limited patient data available today. d鈥橦ondt is currently working on this exploratory research with master鈥檚 student Rozhan Moosavi. 鈥淪he鈥檚 testing the system using many different images of eyebrows. This allows us to train a model to generate eyebrows with specific characteristics鈥攕uch as thickness or color鈥攖hat we define in advance. Just like the model will later need to generate scar tissue based on the treatment method applied.鈥�

So it鈥檚 not simply a matter of building an app and being done. It involves years of work: collecting data, training models, verifying predictions, and ensuring everything is done safely and carefully.

Building something exciting

The more data, the better. d鈥橦ondt hopes that Catharina Hospital can persuade partner hospitals to collect data as well. 鈥淭hat way, we can build something truly exciting much faster.鈥� And he can reassure people concerned about patient privacy: 鈥淭he photos are securely stored under strict digital safeguards. We design the model so that it does not leak sensitive data, such as images.鈥�

The project aligns with the mission of the Center for Safe AI, a 黑料福利网 initiative currently in development. Within this center, both technical and socio-technical challenges associated with the development and application of AI are addressed, with particular focus on effectiveness, robustness, fairness, accountability, and human鈥揂I collaboration.

Keeping the focus

In this project, AI remains a tool to support the conversation. The physician is still essential. The patient remains at the center. The technology helps provide a stronger foundation for decision-making. This also reflects Catharina Hospital's broader approach: AI is not an end in itself. It should help make care smarter, more personalized, and affordable.

Within the clinic, d鈥橦ondt observes that there are some doubts about the system being tested. 鈥淭he physicians who participate and invest their time aren鈥檛 seeing much in return yet. Right now, we鈥檙e mainly focused on collecting data. They have to trust that it will pay off.鈥� It鈥檚 up to d鈥橦ondt and the master鈥檚 students supporting him to keep medical specialists engaged and interested.
 

The base of this article was written by as part of their monthly series 鈥榁ooruit met de zorg鈥� (Moving Healthcare Forward). In this series, the hospital shows how innovation, such as AI, contributes to better, more patient-centered care.