How do you mend a broken heart? There’s an algorithm for that, and B.C. innovators are on the case

It’s all about squiggles on a page. Even if you have never lain on an examination table while a technician stuck electrodes all over your chest—and then flicked on the switch!—you’re likely familiar with the image of an electrocardiogram (ECG): a page of graph paper covered in wavy lines. They’re unintelligible, or seem that way to the untrained eye.

For a cardiologist, however, those lines reveal the nature and path of the electrical impulses that activate the heart, showing everything from heart rate and rhythm to the size and nature of heart chambers. Experienced specialists, the ones who look at tens of thousands of scans a year, can also identify heart defects and past damage, the effect of heart drugs, even the performance of an implanted pacemaker. It’s kind of magical.

Five years ago, Peter Noseworthy, a Canadian physician-researcher working at the Mayo Clinic in Rochester, Minnesota, started to wonder if he could supercharge the magic by reviewing all ECG results with artificial intelligence. Instead of relying on bleary-eyed humans staring at a practical maximum of about 50,000 scans a year, Noseworthy wrote an algorithm—really, just a very specific set of instructions—that could guide a computer to sample millions of images, to see what it could see. The Mayo system, including dozens of small clinics, as well as two other major campuses in Florida and Arizona, conducts more than 400,000 ECGs a year and has more than four million scans on file; it’s a dataset that makes this kind of deep learning possible.

The results, according to B.C. cardiologist Brett Heilbron, were “mind-blowingly good.” Cross-trained with patient demographics, outcomes and cardiac imagery, Noseworthy’s algorithm was able to detect a whole range of factors that even the most diligent humans can’t see—everything from the age and gender of a subject to valve function, valve disease, inherited cardiac conditions and more. And there are uses outside of cardiology, too: Noseworthy explained in a recent virtual UBC Medicine seminar that there is even an identifiable signal  for cirrhosis (serious scarring of the liver).

Adding to the promise, ECGs are easy, non-invasive and cheap, especially compared to alternative diagnostic tests. Echocardiograms, for example, require more expensive equipment and greater expertise from an ultrasound technician, and take almost an hour. Heilbron says each test costs the system around $1,000, and you can wait more than a year to get one. ECGs, on the other hand, take much less training to administer and are remarkably quick; the actual scan lasts for as little as 10 seconds and a full appointment can be concluded in under 15 minutes. Add in the price of disposables, the time of the cardiologist to read the scan—even the rent, heat and light in the lab—and you can still book one by tomorrow and get out for a small charge to the system of around $35.

The problem is the chasm between conceiving of a medical innovation and implementing a transformative clinical practice. New diagnostics, like new drugs, need to be proven and approved by the Food and Drug Administration (FDA) in the U.S. and by Health Canada in this country—a process that is complicated, time consuming and expensive. B.C.’s medical system is also already overwhelmed. (Heilbron goes so far as to say “it’s broken.”) In a time when people are struggling to schedule, execute and pay for the treatments already in place, there is little capacity—or appetite—to fund and implement something new.

Heilbron is undeterred. He’s a serial innovator (he was instrumental in introducing the use of CT scans in cardiology), and while he explains this by saying, “I’m old, so I have time on my hands,” the comment is hard to take seriously. Even at 63 (not as old as it used to be), Heilbron gets up early three days a week to ride his bicycle 50 kilometres. After that, he gets to work, seeing patients, managing as co-director of the Advanced Cardiac Imaging Program at St. Paul’s Hospital in Vancouver, or teaching as a clinical associate professor for UBC. In his spare time, he’s also vice-president and head of cardiac testing at LifeLabs, Canada’s largest private-sector medical laboratory, so he knows both the public and private systems. Still, the challenge of overcoming the inertia of B.C.’s medical system would be an impossible uphill climb without, he says, the path-clearing assistance of Vikram Devdas.

Devdas is director of medical technologies at Providence Health Care Ventures. PHC Ventures uses what Devdas describes as “a small fund,” including, for example, the $1 million that St. Paul’s makes on its parking lot every year, to leverage the output of the many researchers and clinicians in the Providence system. As an innovation hub and business accelerator, PHC Ventures works to help health professionals develop new drugs, devices, AI-driven tools and other innovations to improve health care directly and/or to make money to augment the income that Providence gets from the provincial government. “We help incubate companies,” says Devdas. “We connect the dots. We get the right team together and raise the funding.”

It’s work for which he is well trained. Devdas is one of those people who reminds you why it’s a good idea to welcome brilliant students into the Canadian post-secondary system. He came from India in 1989, at age 17, and graduated from UBC five years later with a degree in computer engineering, joining what was then a 30-person high-tech startup called Sierra Systems. In five more years, by which time Devdas had also completed a master’s degree in electrical and electronics engineering at UBC, PMC Sierra was the second-largest semi-conductor company in the world, behind Intel—a run-up that had made Devdas so wealthy that he decided to retire. But not for long. He has since spent his career enriching his education (at Harvard, Stanford and Cornell), starting four companies and acting as an angel investor, specializing in life sciences and bioscience. So when PHC Ventures recruited him in November last year, he says, it seemed like the perfect opportunity to give something back.

Devdas also notes that the ECG AI algorithm offers an excellent project for PHC Ventures. Although a U.S. company called Anumana has commercialized Noseworthy’s innovation and won FDA approval for its first application—identifying a heart complication called a low ejection fraction (LEF)—the AI still has to be proved in Canada. So, with PHC Ventures’ support, Heilbron and his colleagues are setting up a pilot project: a small study to prove the concept and generate the data necessary to win Health Canada’s approval for a  complete rollout.

The results could be transformative, Devdas says. LEF describes a kind of heart failure in which the left ventricle fails to pump out enough oxygenated blood. A normal ECG can give a hint that it’s happening, but then a patient will require a referral for an echocardiogram to confirm the diagnosis, sometimes waiting a year to 18 months. If, instead, someone at St. Paul’s could pop the ECG image into the cloud for a $5 review by Anumana’s ECG-AI LEF algorithm, Devdas says, “you get a diagnosis in hours instead of months, and you haven’t wasted tens of thousands of dollars. You also leave echocardiogram machines and staff available for people who really need them. You use resources more efficiently.”

There are also a host of other potential ECG AI applications, including one that has already been developed by a UBC cardiology resident named River Jiang, whom his UBC supervisor, Heilbron, describes as  “a superstar.”

Jiang is something of an AI hobbyist. He says that even in high school in Kitchener, Ontario, he enjoyed programming and web development: “It was a toss-up whether I would study comp sci or biology in university.” Biology won, but through his undergrad, his MD and his residencies in internal medicine and cardiology, he’s watched the development of AI, shadowing post-docs in the UBC electrical and computer engineer lab and teaching himself enough coding that he could create and validate an AI model. (“There’s so much information available online,” he says.)

Hearing about Jiang’s expanding expertise, Andrew Krahn, a professor and head of cardiology at UBC, challenged Jiang to do something with long QT syndrome, a hereditary condition that causes tachycardia—accelerated and uneven ventricular contractions. It often strikes people in their 30s and, under normal circumstances, the first presenting symptom can be sudden death. So Jiang wrote a new ECG algorithm on his laptop, accessing a dataset through the cloud to train his model. It’s now being validated with information from clinics across the country associated with the Hearts in Rhythm Organization (HIRO), which Krahn founded in 2016 and which specializes in hereditary conditions.

The bottom line, says Heilbron, is that AI is “so powerful.” Using it to review ECGs alone, “it will promote early identification, better diagnoses, better access to therapy, less inappropriate testing, less waiting and frustration,” he maintains. And across the system in general, Heilbron adds, “AI  will change everything in medicine, in a good way.”