Typically heart strength is quantified using ejection fraction, or the proportion of blood expelled from the heart with each beat. But this percentage may not show high muscle strain, and interpretations can vary depending on who peforms the exam. (Getty Images)
It’s been a running pitch for years that artificial intelligence could help deliver more accurate and more consistent healthcare outcomes, especially in diagnoses that may rely on a subjective eye. But the COVID-19 pandemic offered a chance for trial by fire.
And in this case, at least, the machine wins.
In a human vs. AI study, researchers found that the technology was much more effective at predicting which coronavirus patients would suffer severe complications and face a higher risk of dying, based on analyses of cardiac ultrasound exams.
By parsing recorded videos of echocardiograms, algorithms developed by the Oxford University spinout Ultromics were able to assess the beating heart muscle and spot signs of damage caused by COVID-19—linked to higher mortality rates both inside the hospital and in the months following treatment.
Typically the strength of the heart is quantified using the left ventricle’s ejection fraction, or the proportion of blood expelled from the heart with each beat. However, this percentage may remain at normal levels despite higher muscle strain, or be affected by changes in blood pressure, and its interpretation can vary depending on who performs the exam.
Instead, mapping the structure of the heart with a grid of points, and tracking how the ventricles move with each twist, squeeze and contraction, allowed the AI to spot patterns of underlying muscle weakness.
The late-breaking study, presented at the annual scientific sessions of the American College of Cardiology, showed that changes in the shape of the left and right ventricles—as well as the patient’s age and history of lung disease—could independently predict in-hospital death rates, while ejection fraction could not.
The analysis covered 870 patients hospitalized with COVID-19 in nine countries. Nearly half were admitted to intensive care units, and 27% were placed on a ventilator; 188 patients died in the hospital, while 50 died within the three-to-six months to follow.
The AI was more accurate in predicting severe cases than a panel of experts trained in echocardiograms, while the study also showed that the use of AI led to less variability among scans and readers.
Ultromics received an FDA clearance for its AI service in early January, to help support clinicians in diagnosing coronary artery disease. Its EchoGo Pro cloud-based offering automatically determines cardiac function and analyzes strain on the heart muscle’s walls.
“COVID-19 has placed an even greater pressure on cardiac care and looks likely to have lasting implications in terms of its impact on the heart,” Ultromics founder and CEO Ross Upton said in a statement at the time. “The healthcare industry needs to quickly pivot towards AI-powered automation to reduce the time to diagnosis and improve patient care.”
More recently, the company garnered a CE mark for its EchoGo Core module in late April, following its deployment within the England’s National Health Service. The algorithm selects the clearest echocardiogram images and highlights important features, while delivering a report with strain and ejection fraction analysis.
In the early stages of the pandemic last year, the FDA gave the green light to a slew of cardiac devices from Philips, Eko and Caption Health to help detect heart damage from COVID-19.
Philips received a full 510(k) go-ahead for several portable and point-of-care ultrasound systems, while the agency expedited clearance for a new version of Caption Health’s AI-powered ultrasound coach, which guides clinicians through capturing an echocardiogram.
The FDA also delivered an emergency authorization to Eko for its electrocardiogram-based algorithm that screens COVID-19 patients for low ejection fraction.