Researchers in the US have developed a rapid, artificial intelligence (AI)-based test that can identify patients with an abnormal heart rhythm, even when it appears normal. This 10 second test for atrial fibrillation could be a significant improvement over current test procedures that can take weeks or even years (Lancet 10.1016/S0140-6736(19)31721-0).
Atrial fibrillation is a common cardiac condition that is estimated to affect between three and six million people in the US alone. The condition is associated with an increased risk of stroke, heart failure and mortality – but it is underdiagnosed. This is because it can be asymptomatic and the patient’s heart can go in and out of the arrhythmia, making diagnosis tricky. It is sometimes caught on an electrocardiograph (ECG), but often detection requires the use of implantable or wearable monitors to capture infrequent atrial fibrillation episodes over time.
“Atrial fibrillation is an arrhythmia where the atrium, or top chamber of the heart, loses its coordinated contractual activity and instead quivers, because the electrical impulse is changed in the way it courses through the atrium,” explains Peter Noseworthy of the Mayo Clinic. “So, the top chamber beats irregularly and it causes the bottom chamber, the ventricle, usually to beat fast and irregularly, which can be bothersome, but most importantly it predisposes people to risk of stroke.”
He adds that atrial fibrillation can be caused by many conditions, such as valvular heart disease, sleep apnoea and hypertension, but often doesn’t have a clear underlying cause.
Noseworthy tells Physics World that researchers suspect that atrial fibrillation usually happens in a heart that is somehow abnormal. The team hoped that this would leave a signature that could be detected on an ECG during normal heart rhythm.
The researchers trained the AI model using data from 70% of the patients, validated the model using data from 10% of the patients and tested it on data from the remaining 20%.
They found that the AI was able to detect differences in the ECGs of patients with atrial fibrillation. From a single 10 s ECG scan that appeared to a medical professional to show a normal heart rhythm, the AI was able to identify patients with atrial fibrillation with an accuracy of 79%. When multiple ECGs for the same patient were tested, the accuracy improved to 83%.
“When we look at the ECGs, they don’t have a single pattern to them, but the AI is able to identify many different, more subtle patterns that we may as cardiologists just recognise as mild abnormalities and do not really draw any attention to,” Noseworthy explains.
According to Noseworthy, the test has two main potential applications: pre-screening the general population to identify those who could benefit from long-term screening for atrial fibrillation; and testing patients who have had a stroke to identify those who could benefit from an anticoagulant or longer term screening with an implantable device.
These possibilities need to be explored further and the AI tested on other datasets and the general population, Noseworthy says. “Everybody in our dataset came to medical care and had an ECG, so it is going to be enriched with people who had some sort of cardiac issue or are concerned about a cardiac issue,” he explains.
In a linked comment, Jeroen Hendriks of the University of Adelaide in Australia says that this approach could lead to a paradigm shift in recording normal rhythm rather than atrial fibrillation on an ECG, with a “specific focus on identifying structural changes”. He cautions, however, that the AI network “has been tested to retrospectively identify atrial fibrillation rather than predicting atrial fibrillation”, stating that it needs further validation.
Hendriks concludes that the researchers “are to be congratulated for their innovative approach and the thorough development and local validation of the AI enabled ECG. Given that AI algorithms have recently reached cardiologist level in diagnostic performance, this AI-ECG interpretation is ground-breaking in creating an algorithm to reveal the likelihood of atrial fibrillation in ECGs showing [normal] rhythm”.