ECG-Trained AI can spot heart disease in under 2 seconds
One of the most common tests in medicine, used for more than a century, may be getting a major AI upgrade. Researchers have developed an AI tool that can identify signs of heart failure and heart valve disease from a routine electrocardiogram (ECG) in less than two seconds. The system has been trained on millions of patient records and is designed to spot patterns from ECG data that would be challenging for a clinician to see.
“It could be a solution to help fast-track the patients who are most likely to have a heart abnormality. When it comes to the heart, earlier diagnosis and treatment saves lives and improves lives”
Dr Sonya Babu-Narayan, Clinical Director of the British Heart Foundation
ECGs are already one of the most common tests in medicine, with about a billion performed worldwide per year. Traditionally, they have been very reliable for assessing heart rhythm and identifying problems like heart attacks, but cannot typically definitively diagnose structural heart diseases. Patients suspected of heart failure or valve disease are usually referred to an echocardiogram, which can have significant waiting times.
In a US trial with 67,000 patients, the AI identified 81% of people with heart failure and 90% of those with heart valve disease. Of course, this system is not intended to replace echocardiography or make a final diagnosis. Instead, it could act as a rapid triage tool by identifying higher-risk patients and prioritising them for further examination.
Rather than replacing the existing diagnostic pathway, this AI can add a layer of intelligence to a test that is already routinely performed. Someone undergoing an ECG for a completely unrelated reason could be flagged for a previously unsuspected heart disease before symptoms become severe.
The approach could be particularly useful in healthcare systems struggling with diagnostic backlogs. AI does not need to make the final clinical decision, simply identifying who should be investigated first would be enough to make a meaningful difference by shortening the time between disease developing and treatment beginning.
Of course, there are limitations. The system is not 100% accurate and therefore will not identify every patient with heart disease, and a high-risk result still requires confirmation with traditional testing and medical review. However, I believe AI is most useful not as an autonomous diagnostician, but as an additional layer of intelligence that reveals new insights to help clinicians understand and find the disease earlier.
