AI learns to detect atherosclerosis before it leads to blood clots

Shortness of breath. Tightness in the chest. Pain radiating down the arm. Or perhaps… nothing at all.

Some people experience warning signs before a blood clot strikes. Others do not. In both cases, atherosclerosis may have been developing silently for years without being detected in time.

Danish researchers hope to change this through the Nordic Collaboration on AI and Cardiac Imaging, which has now reached an important milestone. Around half of the planned 100,000 cardiac CT scans have been collected, allowing the researchers to begin training AI to identify and quantify atherosclerosis.

The aim is to develop a clinical decision-support tool that can help doctors identify patients at increased risk of serious heart disease earlier, while making assessments more consistent across clinicians and hospitals.

"We can already learn a great deal from a CT scan, but if AI can help us quantify precisely how much atherosclerosis a patient has, we have a much stronger basis for deciding who needs further examination or preventive treatment. At the same time, it could reduce some of the variation that inevitably occurs when people interpret the same images,” says Simon Winther, Consultant Cardiologist at Gødstrup Regional Hospital and Clinical Professor at Aarhus University.

AI is not here to replace clinicians

The researchers are building one of the world’s largest cardiac CT datasets, drawing on scans from Denmark, Norway and Sweden. The dataset ranges from healthy individuals to patients with extensive atherosclerosis, providing the AI models with examples across a broad spectrum of coronary health and disease.

In the longer term, the ambition is for AI not only to identify atherosclerosis, but also to quantify the extent of the disease and analyse blood flow through the coronary arteries. This could give clinicians a stronger basis for assessing which patients are at greatest risk of developing serious heart disease.

“A patient may not need treatment right now. But if we can see that their atherosclerosis is progressing in a way that increases the risk of a blood clot, we may be able to intervene earlier and potentially prevent the disease from progressing,” says Simon Winther.

The researchers also expect the technology to free up clinical time, allowing doctors to focus their attention on patients who require the most complex assessments.

“AI is not a competitor to clinicians,” says Monika Colombo, Assistant Professor at the Department of Mechanical and Production Engineering at Aarhus University and coordinator of the project.

“On the contrary, all our models build on the knowledge and experience that clinicians bring to the project. We use AI to make better use of clinical time and resources, allowing clinicians to focus on the patients who need the most thorough assessment.”

Turning images into knowledge

AI for medical imaging is developing rapidly, but the researchers believe the project has a particular strength in its combination of an unusually large dataset, advanced computing infrastructure and close collaboration between clinicians and engineers.

The project brings together expertise from across the Nordic countries. Monika Colombo from Aarhus University coordinates the project, while Simon Winther from Aarhus University and Gødstrup Regional Hospital and Samuel Emil Schmidt from Aalborg University are among the researchers leading the project.

Together, the researchers combine clinical expertise in cardiovascular disease with engineering, medical imaging and artificial intelligence.

Some of the advanced computations are carried out on Gefion, Denmark’s AI supercomputer. Its computing power enables the researchers to analyse very large volumes of imaging data and perform complex simulations of blood flow through the coronary arteries.

Another key objective is to make the results available to the wider research community. The AI models are being developed as open source, enabling other researchers to test them, build on them and compare them with other methods.

"We have a unique combination of one of the world’s largest datasets, state-of-the-art AI tools and strong interdisciplinary collaboration across the Nordic countries. At the same time, it is essential to us that the models are freely available to other researchers. That gives us the best possible foundation for developing solutions that could ultimately make a real difference for patients,” says Monika Colombo.

Contact

Associate Professor Monika Colombo
Department of Mechanical and Production Engineering, Aarhus University
Mail: mc@mpe.au.dk
Tel.: +4587151664

Jesper Bruun
Journalist, Aarhus University
Mail: bruun@au.dk
Tel.: +4542404140