New AI tool can predict heart failure at least five years before it develops
A new AI tool analyzes texture changes in the fat around the heart on routine CT scans to predict a patient’s risk of heart failure up to five years before symptoms appear. Trained on over 72,000 scans from nine NHS Trusts, it reached 86% accuracy and identified a highest-risk group with a 20-fold greater chance of developing heart failure.
- AI analyzes fat texture around the heart on routine CT scans to spot hidden inflammation and predict heart failure risk 5+ years in advance
- Trained on 59,000 scans and validated on 13,424 more across nine NHS Trusts, hitting 86% accuracy
- Highest-risk patients were 20 times more likely to develop heart failure than lowest-risk ones, with 1 in 4 actually developing it
- Researchers plan to expand the tool to any chest CT and integrate it into NHS radiology workflows pending regulatory approval
A new AI tool from Professor Charalambos Antoniades’s team at Oxford can forecast an individual’s chance of heart failure at least five years before symptoms show up. By analyzing routine cardiac CT scans—about 350,000 of which are done annually in UK NHS hospitals—the program detects subtle textural shifts in the fat surrounding the heart. These changes signal underlying inflammation in the heart muscle, something radiologists can’t see by eye.
In tests involving more than 70,000 patients across nine NHS Trusts, people flagged in the highest risk group were 20 times more likely to develop heart failure within five years compared to the lowest-risk group. Specifically, one in four of those high-risk patients went on to develop heart failure. The model hit 86 percent accuracy when validated on an extra 13,424 scans, after being trained on 59,000 anonymized cases from England.
Currently focused on cardiac CTs, the researchers aim to extend the method to any chest CT—lung scans included—so that more patients undergoing imaging for various reasons can be assessed automatically. With regulatory approval pending, the plan is to integrate the risk scoring into standard radiology workflows across the NHS. Doctors would then use each patient’s risk number to decide on monitoring intensity and preventive treatments.
Funded by the British Heart Foundation and supported by NIHR’s Oxford Biomedical Research Centre, the project leverages advances in bioscience and computing to catch heart disease earlier. By turning existing scans into predictive tools, it offers a path to more targeted care and could ease hospital pressure by preventing severe heart failure down the line.