Artificial Intelligence, Hospitals

Mayo Clinic selects U.K startup’s AI tool to gain insights on Covid’s effect on the heart

A young UK startup is collaborating on a research study with Mayo Clinic to use an FDA-cleared automated echocardiogram assessment and quantification tool to unlock the mysteries of how Covid-19 attacks the heart.

Ultromic’s AI product, EchoGo Core, is an AI-powered tool to automatically quantify and assess cardiac echocardiograms

Artificial intelligence in healthcare cut its teeth in radiology with many companies developing algorithms to read MRI and CT scans and catch things that the human eye doesn’t see.

A U.K. startup — Ultromics, based in Oxford — is taking that effort into the world of echocardiography and last week won a major endorsement. The company’s AI software will now be used by Mayo Clinic in all its 10 echo labs in a research study, said Dr. Patricia Pellikka, a cardiologist, and clinical researcher at Mayo Clinic, to gain a better understanding of how Covid-19 affects the heart. Pellikka noted that it will be an international study and she has already seen interest from medical centers in other countries.

“There’s just still fairly limited information about cardiac manifestations of Covid. We know that it can invade the cardiac myocyte[muscle cells] and have a variety of deleterious effects, but we are going to gain a lot of understanding by this multicenter collaboration – just getting these echocardiograms and looking at them so we characterize and better understand exactly how the virus impacts the heart,” Pellikka said. “I think the results of our study will enable us to manage patients more expeditiously and improve the speed and quality and reproducibility of our assessments in the future, depending on what we find,” Pellikka said.

The multi-site study will look at echocardiograms (cardiac ultrasounds) of 500  Covid-19 positive men and women between 18 and 89. The retrospective study will examine the images of patients who have had a clinically-indicated cardiac ultrasound in a three-month period. The main goal is the assessment of automated cardiac measurements, ejection fraction and Global Longitudinal Strain, for the classification of Covid-19 patient outcomes. [Global Longitudinal Strain is emerging as a new metric among other measurements used to predict cardiovascular outcomes in patients.]

“There is information within the images that the human eye doesn’t appreciate even [for] the expert cardiologists reviewer,” she added. “There are patterns and details, which have meaning and that we still need to explore and that’s the exciting thing about the collaboration with Ultromics.”

There are other companies that automate the assessment of echocardiograms — for instance, San Francisco-based Bay Labs received FDA clearance for its EchoMD AutoEF software product. Specifically, the clearance was for its automated clip selection and calculation of left ventricular ejection fraction (EF) in June 2018.

But Pellikka nonetheless chose Ultromics and it EchoGo Core AI tool.

“There are several companies but I wanted to find the absolute best partner for Mayo Clinic and to utilize our resources here most effectively and I was impressed with that group and their dedication to science and our common mission in trying to improve patient outcomes,” she said.

Ultromics has been around roughly three years and it has raised £13 million (roughly $15.7 million in today’s exchange rates) from angels and other investors including lead investor Oxford Sciences Innovation, said its CEO, Ross Upton, in a recent Zoom interview. The company is in the midst of raising a Series B round of funding.

Upton, who co-founded the company in 2017 and was named to the Forbes 30 under 30 list in 2019, of course, claims that no other company is doing what his startup is. This possibly relates to the second product that Ultromics has developed and which has CE Mark but not FDA clearance so far. Called the EchoGo Pro, it can predict coronary artery heart disease by analyzing echocardiograms.

“Most AI devices to date generally don’t take it as far as we’ve taken it. They either highlight in an image where something might be or they help doctors do things more quickly for triage purposes,” Upton said in a recent Zoom interview. “To our knowledge, we are the only AI company to predict one-year patient outcomes and we are certainly the only company within echo[cardiography] to do that.”

The company’s data was trained in a trial that began in the John Radcliffe Hospital within Oxford University Hospitals but later the EVAREST study was expanded to 30 hospitals in the U.K. The study encompassed about 8,000 patients who generated 700,000 cardiac ultrasound images

“We followed every patient for a year after their echocardiograms to see if they’d had surgery, angiogram, a heart attack and we predicted that one-year event, the one-year outcome of that patient,” Ross said.

Typically, without the use of AI, cardiologists reads an echocardiogram of a patient and based their analysis recommends the patient for an angiogram, surgery or sent home.

“That decision historically probably isn’t as accurate as it could be and that’s why we created EchoGoPro to try and improve that decision process,” Upton said.

And validation data based on a trial in the U.S. showed that the EchoGo Pro had high accuracy when it came to making predictions.

“We had four different cardiologists, independent cardiologists read the echocardiogram without our device and then read it with our device to see how much we would improve their reads using our recommendation and broadly speaking without the device, cardiologists are 80 percent accurate in making that decision,” Upton said. “When they read it with our device, the sensitivity was 96% and specificity of 88%. Overall accuracy was about 93%.”

Upton said Ultromics’ software is not only more accurate than human eyes but is also more efficient saving 15 min of the cardiologist’s time per patient in reviewing the ECGs, which has real implications on hospital costs. But ultimately, the success of AI companies in healthcare, including that of Ultromics, will largely depend on reimbursement. And while the fee-for-service system that has shackled U.S. healthcare system may finally begin to ease, we are still far from actually paying for outcomes.

“AI companies are particularly challenged because a lot of AI companies – they want to bring their technologies in and they ultimately want to improve patient outcomes and unfortunately the RUC and the payer system in the U.S. isn’t set up to pay to improve patient outcomes,” Upton said. “They pay for work done by clinicians.”

RUC or the Relative Value Scale Update Committee is a volunteer group of 31 physicians — part of the American Medical Association — who historically have advised Medicare on how to value a physician’s work and are regarded to be very influential in determining what gets paid through public insurance.

Still, Upton is confident, pointing that a new CPT code that went into effect in January would be something physicians can use to get reimbursed by Medicare should they use the company’s AI echocardiography automatic quantification and assessment tool – the EchoGo Core.

“We made sure that our product delivered what the CPT code outlined, which has meant we must be one of the only AI companies out there that has FDA clearance, CE Mark and a category one CPT code,” he said. “So we are in a very fortunate position.”

Correction: A previous version misspelled Dr. Pellikka’s name. 

Photo: Ultromics