AI algorithm from the Heart Institute interprets ECGs better than a human

An artificial intelligence algorithm developed by a team at the Montreal Heart Institute is as good as, or even better than, humans when it comes to interpreting electrocardiograms.

The tool is even able to detect disorders invisible to the human eye and whose first symptoms might not emerge until months later, said one of the project’s leaders.

“We can detect (heart) failure (…) or predict the risk of arrhythmia in people who do not have it at the moment, but who could have it or develop it in the coming years, or detect genetic diseases,” explained Dr. Robert Avram, who is a cardiologist at the ICM.

“These are diseases that cannot traditionally be detected from an ECG, but thanks to our model, we are able to detect them with very good accuracy.”

Developed in particular by Alexis Nolin-Lapalme and Achille Sowa, the DeepECG model was trained on more than one million ECG results, then validated retrospectively in eleven centers internationally.

DeepECG is the world’s first fully open-source “foundational” ECG model, Dr. Avram stated. This means that hospitals and research centers worldwide can adapt it to their local population to develop new cardiac biomarkers, even with limited data.

“From an electrocardiogram, a cardiologist can detect several diseases,” said Dr. Avram. “But by training a foundational model, we are also able to teach the algorithm to detect diseases that are invisible to the naked eye and undetectable by the conventional means we have in medicine.”

DeepECG is able to detect conditions that even the most expert eyes might miss, such as reduced heart force; structural heart diseases, such as aortic stenosis or dilation of the heart muscle; or electrical signatures associated with hereditary heart disease.

The tool offered “robust” performance even in the presence of less frequent problems, the researchers say, but was slightly less effective in the face of very rare conditions, a situation they attribute to the fact that the data needed to train it was available in limited quantities.

“We could screen in advance for diseases that are sometimes invisible to the naked eye, and then initiate treatments before a complication such as hospitalization or an acute episode of heart failure develops,” Dr. Avram explained.

He cites, for example, the case of a patient who presented at the ICM for a cardiac evaluation before knee surgery and showed no signs of heart disease, but in whom a DeepECG detected heart weakness. The condition was subsequently confirmed with an echocardiogram.

“Without artificial intelligence, the doctor would probably have said it was okay to perform the surgery,” said Dr. Avram. “But this led to further tests to understand why his heart was weakened. And with heart disease, the earlier medication is started, the better the chances of recovery, and even of returning to normal.”

Artificial intelligence, he asserts, does not take “anyone’s job away,” quite the contrary. Rather, it allows existing staff to do more and do better with the often limited resources at their disposal, Dr. Avram believes.

In this regard, he points out, one currently has to wait about twelve months in Quebec to benefit from a cardiac ultrasound, and in half of the cases, the examination does not detect anything abnormal.

“It’s reassuring for the patient, but when you think about patients who wait up to a year for the test and who may have health problems that could be confirmed with these tests, well, it’s not necessarily optimal resource management,” said Dr. Avram.

DeepECG could allow for prioritization of care and offer “more precise and earlier personalized care to people identified as high risk,” he added, while keeping in mind “that hospitalization costs much more than a consultation in the doctor’s office.”

DeepECG has been undergoing a phase II clinical trial since October, in order to measure its concrete impact on the access, speed and quality of cardiovascular diagnoses in a real context.

Details of this innovation were recently revealed in the European Heart Journal.

–This report by La Presse Canadienne was translated by CityNews