You’ll find image recognition technology in almost every kind of consumer service you would use today — like Facebook, Google, or any number of other apps — but Yann Fleureau is hoping that he can build a whole business off of its increasing success in healthcare.
Fleureau hopes that his startup, Cardiologs, can use image recognition to analyze an EKG and detect potential problems going forward. Though he says his family members have dealt with cardiovascular problems over time, Fleureau says the motivation was a business opportunity to apply emerging technology that is getting more and more accurate every day in a slice of the healthcare universe that hasn’t seen a lot of movement yet. The company said it has raised $6.4 million in new financing today.
“At the end of the day, doctors don’t really care about AI,” Fleureau said. “They care about the patient, so the question is whether we move the needle for the patient. That’s what the discussion was about. For them, AI is kind of nice, but they want to see tangible results.”
Doctors and specialists are trained to recognize any kind of aberrations in an EKG, which allows them to quickly identify if there are any abnormalities in how someone’s heart is working. That can include something along the lines of arrhythmia — or improper beating of the heart — that could signal long-term complications. But like any engineer, Fleureau hopes to streamline the process of looking at hours of EKG data and make it easy to upload the information and get a quick analysis powered by a machine.
It’s the same kind of idea behind teaching a computer to detect a cat in a photo, but rather than just for typical consumer behavior (like searching for a new pair of shoes), that kind of image recognition is seeing a lot of new applications. While image recognition isn’t perfect yet — and that’s also wildly dependent on the data set that’s available — Fleureau said that the company is upfront on its success rate just like any other company that has to go through the FDA to get into the hands of doctors.[relatedd_articles”No one has 100% accuracy, it’s to have a match between what you claim and do,” Fleureau said. “But if you say you have 95% accuracy you’d better have the proof of claim. If something goes wrong and the FDA finds our you only have 90% accuracy, from that perspective the device framework enables this well.”
Fleureau says that the company has figured out a way to build a kind of defensibility based on its relationships with experts that can help pin down the best signals to improve their EKG analysis. Most machine learning startups — image recognition or otherwise — will live and die by their data sets and the ability to accurately create the rules that will define whether they are successful. The startup’s success will also be a factor of getting into the hands of doctors and getting an increasing pool of data. Cardiologs works with larger medical device companies and providers to gain adoption, Fleureau said.
To be fair, that has been movement in this area. Like any startup — especially in healthcare — Cardiologs will face steep competition from all fronts. The obvious competitor is iRhythm, which similarly looks to analyze potential problems with heart patterns. In the end, if a product can diagnose a potential problem and, as Fleureau says, move the needle on the patient, it’ll probably win in some way. Fleureau says he hopes by focusing on software and partnering with hardware manufacturers, Cardiologs will provide something easy to use for doctors that will hit widespread adoption.