Stroke is the second leading cause of death worldwide. One common type is called ischemic stroke, which is often caused by a build-up of fatty plaques in blood vessels—a condition known as atherosclerosis.
To diagnose it properly, doctors need to see the plaques and vessel walls clearly using medical images. This usually involves a time-consuming manual process that depends heavily on the skill of the operator.
Although some computer tools exist to help, they still aren’t accurate enough for everyday use in hospitals.
Now, a research team led by Dr. ZHANG Na at the Shenzhen Institutes of Advanced Technology in China has developed a much better method using artificial intelligence (AI).
Their new tool can quickly and accurately detect and measure the plaques and vessel walls in high-resolution MRI scans. This can help doctors assess the risk of stroke more effectively and make better treatment decisions.
The new method works in two main steps. First, the researchers created a deep learning model called Vessel-SegNet, which is trained to find and outline the vessel wall and the space inside it (called the lumen).
Then, the tool uses what it has learned about the shape and structure of vessel walls to better find and measure the plaques—this step uses special techniques called priors, based on either expert input or mathematical tools like the Tversky loss function.
The research was based on MRI scans from 193 patients with plaque build-up in their blood vessels. The scans were taken from five different hospitals. Of these patients, 107 were used to train and refine the AI, 39 were used to test it internally, and 47 were used to test how well it works in new settings.
The results were very promising. The AI was able to match expert-level performance in outlining the vessel walls and lumen, with an accuracy score (called Dice similarity coefficient, or DSC) of over 90%.
When it came to detecting plaques, adding information about the vessel wall helped improve the accuracy by more than 10%. With advanced techniques, the DSC for plaque detection reached over 88%.
Even better, the tool can complete a full scan analysis for one patient in under 3 seconds—something that could take a human expert much longer. This makes it fast enough to be used in real hospital settings.
“Our goal is to use AI to produce clear, reliable results that help doctors make better decisions about stroke care,” said Dr. ZHANG. “In the future, we plan to test this method in more hospitals with different types of machines and patient groups to make sure it works everywhere.”
If you care about stroke, please read studies about Half of people with heart rhythm diseases dying of heart attack, stroke and findings of New method reduces heart attacks and strokes over five years.
For more about stroke, please read studies about Blood thinner drug prevents strokes in hidden heart issues and findings of Intensive blood pressure treatments could prevent stroke in older adults.
The study is published in European Radiology.
Copyright © 2025 Knowridge Science Report. All rights reserved.
