AI predicts repeat heart attacks better than experts

AI continues to amaze us. It can detect tumors and bone fractures and: will soon be able to predict the risk of a heart attack. CARA Lab—a collaboration between Radboudumc, Amsterdam UMC, and Abbott—is currently investigating how AI can detect weak spots in the vessel wall; spots that can cause a new heart attack. “The model proved to be better than specialized laboratories,” says technical physician Jos Thannhauser. Editor of IO+, Elcke Vels, spoke with him about the newest developments in Radboudumc.

Every day, approximately 92 people in the Netherlands are admitted to hospital with a heart attack. A heart attack occurs when a coronary artery becomes blocked by a blood clot. This is often because the vessel wall is already narrowed due to atherosclerosis. The heart then receives too little oxygen. The standard treatment is angioplasty: a doctor inserts a small balloon into the blood vessel and stretches it. Usually, a small tube, called a stent, is then placed to keep the vessel open.

Unfortunately, about 15% of people have another heart attack within two years of their first one. Some of them need to undergo another angioplasty, and a few even die. Thannhauser and his team are therefore investigating whether they can better detect weak spots in the vessel wall – spots that can cause a new heart attack – by using a mini camera and an AI model. The model can also help with stent placement.

This is how it works: OCT (Optical Coherence Tomography) is a technique in which a mini camera is inserted into the blood vessel. The camera uses near-infrared light to produce highly detailed images of the inside of the coronary arteries. "When you look at a blood vessel under a microscope, you can see three layers: the intima (inner layer), the media (middle layer), and the adventitia (outer layer). Sometimes you see fatty substances in the vessel: plaque. This appears as dark areas on the images. You can also see calcium deposits. Thousands of images were manually annotated for the study." An AI model was trained based on this.

AI and stent placement

AI can offer advantages in several ways. First, in determining where the stent should be placed. OCT takes no fewer than 540 images at a time, making it difficult for doctors to fully assess the images. “AI can help analyze these images and thus determine the right place for the stent,” says Thannhauser.

In addition, AI can check whether the stent is positioned correctly. “Ideally, no tears should occur in the vessel wall during stent placement. AI can automatically recognize and analyze the stent framework,” explains Thannhauser.

A better predictor than humans

AI is not only used during stent placement. “We are also looking for areas in the blood vessel where there is no stent, but which are vulnerable,” continues Thannhauser. What did the CARA Lab study show? When an expert in a specialized laboratory found a weak spot in the blood vessel, approximately 11% of those patients experienced heart problems again within two years. When the AI performed the assessment, more weak spots were detected, and more than 12% of patients experienced another heart problem.

Even more striking: the AI was much better than humans at ruling out a new cardiac event — 98% versus 93%. “That's because the AI doesn't just look at the narrowing itself, but rapidly analyzes the entire blood vessel,” explains technical physician Thannhauser. “This allows you to make a much more accurate assessment of the risk.”

Implications for treatment

Being able to accurately predict risks is important for patients. It has significant implications for treatment. “If we know which patients have risky weak spots in their blood vessels and where they are located, we can adjust their medication. Or we can place a stent as a preventive measure.”

Next step: CE marking

The next step is to use the system in real procedures. “For that, we need CE marking. Such a process usually takes four to five years. Until then, we will continue to use the model in research and for training doctors.”