An AI system that can predict what a patient’s knee X-ray will look like a year in the future could transform how millions of people with osteoarthritis understand and manage their condition, according to research by the University of Surrey.
A new study, published at the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2025), details how the Surrey-developed technology uses advanced machine learning to generate a realistic “future” X-ray alongside a risk score for disease progression in the knee. These two outputs combined can give doctors and patients a clear, visual forecast of how the condition may develop over time.
Osteoarthritis is a degenerative joint disease that affects more than 500 million people worldwide and is the leading cause of disability in older adults. Trained on nearly 50,000 knee X-rays from almost 5,000 patients — one of the largest osteoarthritis datasets in the world — the system outperforms comparable AI tools for predicting osteoarthritis progression, doing so around nine times faster and in a much more compact form. Researchers say this blend of speed, scale, and accuracy could accelerate adoption in real clinical settings.
Using an advanced generative model, called a diffusion model, the Surrey system generates a “future” version of a patient’s knee X-ray and highlights 16 key points in the joint. These points add transparency by showing which areas the AI is monitoring for changes, making the system easier for clinicians to understand and trust.
In the future, this research could pave the way for similar tools in other chronic conditions for example, predicting lung damage in smokers or tracking heart disease progression, giving doctors and patients the same kind of visual insight and opportunity to act early. The team is also seeking partnerships to bring the technology into real-world clinical settings.
Here is an exclusive Tech Briefs interview, edited for length and clarity, with Professor Gustavo Carneiro, AI and Machine Learning, and Postgraduate Research Student David Butler.
Tech Briefs: What was the biggest technical challenge you faced while developing this AI system?
Carneiro: The challenge is more on the perception of the tool much more than on the technical challenges, because that implies the new way of seeing the tool as something that can help doctors and patients to follow a treatment, see if the treatment will work, visualize how the treatment will perform one year from today. So, it's more on the behavior, let's say, of doctors and patients than the technical challenges.
Butler: Most of the tools were already there, but trying to design them in a way that would lead them to be more accepted in the clinic is really more important.
Tech Briefs: Can you explain in simple terms how it works, please?
Carneiro: Imagine that you go to a doctor; maybe you already have some issues in your knee — you're developing arthritis, but the knee is pretty healthy at that stage. Then the doctor requests an X-ray of the knee, you take the exam, and then you show that to the doctor. The doctor looks at it and then assesses the risk of the condition — will it worsen in one year's time? In that case, the doctor needs to work on some treatment. At this point it's the opinion of the doctor, right? The current systems that are available would basically take that image and then just estimate a risk of the condition worsening in one year.
What our system does on top of that is it shows how the knee is likely to appear in one year. We also show a few of the biomarkers in the knee to represent more quantitatively how the condition will progress.
Butler: These biomarkers will help doctors measure with a bit of a certainty that the condition will actually worsen. At that point, the doctor will have his or her own opinion about it. They will have a risk estimate from the system. They will have an image of how the knee will look in one year and will have these biomarkers, right? This means that the doctor will have a lot more things to show and to work with.
Tech Briefs: Do you have any set plans or further research, work, etc.? If not, what are your next steps?
Carneiro: Our immediate next step — which we are already doing, we are submitting a paper very soon about it — is to have better predictions and better mappings of the knee. Right now, we have six biomarkersand we are replacing that with a map of how the regions of the knee are likely to get worse in one year. That's one thing: how we show the results. Also, we want to improve in terms of risk estimation and so on, and also the quality of the images. This is what we are submitting very soon, but then after that, we will take more information into account (patient information, demographics, more of the previous images). That's the idea: Try to have a better tool with more information from the patient.
From there, predict images more into the future instead of just one year, maybe two years, three years. But then the degree of uncertainty will increase with that. So, we need to communicate this uncertainty to the doctors really well.
