New AI Can Predict Heart Attacks Years in Advance
Recent studies have discovered new technology that can identify those with a large risk of heart attack around five years before the potentially fatal event takes place, as stated by research funded by the British Heart Foundation. This new scientific discovery uses artificial intelligence to pin-point areas of blood vessels that may cause problems in the future.
Working from the University of Oxford, group of researchers have designed and implemented a new ‘biomarker’ named the Fat Radiomic Profile (FRP) which uses cutting edge machine learning techniques. It identifies problematic parts of the lining of blood vessels that supply blood to the heart (aka coronary arteries) including scarring, swelling and abnormal changes to the usual shape/direction of the passages.
Heart attacks can occur when a coronary artery is blocked. This causes the muscle of the heart to become damaged or even destroyed. Most of the time the causes of blockages are build-ups of substances such as fat or cholesterol, meaning that the causes can be mitigated if caught early. This makes technology like the FRP so important, as the simple information it finds can easily save many people’s lives.
In hospitals, when a patient is suspected of having the potential for a heart attack, a common response is to have a Coronary CT Angiogram (CCTA) done. This analyses the coronary arteries to check if any have been reduced or obstructed. Around 75% of CTTA scans show the patient as being free or narrowing of the arteries, however some of them will still have a coronary heart attack in the future. Unfortunately, for now there are no fail-safe ways for doctors to check if a patient will or will not have a heart attack in the future.
In order to train the AI, the team used fat biopsies from 167 people with matching CCTA scans. The artificial intelligence algorithms analysed the genes associated and context for each example. They also used the CCTA scans of nearly 5500 people who had a heart attack under five years of having the scan. The algorithm used machine learning to understand the links between the CCTAs and heart attacks.
To test the AI, the researchers gave it 1,575 CCTAs to analyse and the team claims that the FRP had an excellent correct prediction rate. In a few years’ time, the team hopes to make the software publicly available, so it can be used as standard practise along with CCTA scans.
Thumbnail source: Harvard Health