Five years ago, Fabien Beckers left his home in Paris and arrived at Stanford University’s Graduate School of Business with a Ph.D. in physics from Cambridge in his pocket, and — like many of his classmates — dreams of launching a company. Located both physically and metaphorically at the heart of Silicon Valley, Stanford has educated many of the women and men who have been shaping our online lives — from the founders of Netflix and PayPal to Google’s Larry Page and Sergey Brin. Like them, Beckers was looking for an idea that would transform an entire industry. “I didn’t leave France for California to build an app,” he says. “I wanted to build a company that would have a meaningful impact on our lives.”
Today, Beckers is the CEO of Arterys, a company using deep learning and artificial intelligence to process data generated by medical imaging machines. Its cloud-based algorithms can show doctors blood flow details that were once impossible to see. “We want to enable data-driven medicine,” Beckers says. “The goal is to build an intelligent platform that helps physicians diagnose ailments and prescribe the most effective treatment. Cloud computing and artificial intelligence have this transformative power.”
The first concrete outlines of this vision are already emerging. Last fall, the U.S. Food and Drug Administration cleared the Arterys software that precisely measures and displays blood flow through the heart in 4D by processing magnetic resonance imaging (MRI). Arterys will be offering this initial product in partnership with GE Healthcare as ViosWorks. (GE Ventures is also an investor in the privately held company).
In 2012, Beckers started Arterys together with digitally savvy physicians Albert Hsiao and Shreyas Vasanawala and Stanford-educated engineer John Axerio-Cilies. They first targeted the heart to prove the power of their cloud AI approach. Conventional MRI works by taking static images of the body one slice after another and then assembling them together. With GE’s new 4D flow sequence, a full scan of the chest can be acquired at once. Arterys transfers those thousands of MRI images of the heart — each about 1 GB in size — to the cloud and then applies advanced algorithms to analyze the data, and quantify blood flow. “If the physician suspects that a patient has a heart defect, assessing blood flow is key,” Beckers says.
With conventional technology, it takes about an hour to obtain cardiac MRI images, Beckers says, and patients frequently have to hold their breath for up to 20 seconds during a scan. “This can be a major obstacle for imaging small children or patients with severe heart problems,” he says. But with advancements in GE’s MRI scanners, the scanning time can be less than 10 minutes and the patient can breathe normally, making MRI a quicker and a more comfortable process.
Within minutes of acquiring a 4D flow MRI scan, physicians can evaluate data in seven dimensions — three in space, one in time, and three in velocity direction — and see actual blood flow in the heart as a 3D image. “Arterys provides the most comprehensive view of blood flow and heart function,” Beckers says.
Cloud computing makes it possible to process even the largest files, and it allows for more robust and predictive analytics. Arterys’ technology also uses deep learning*. It can train algorithms to automate some of the more tedious radiology tasks and deliver practical insights for improving treatment decisions and patient care.
Arterys founders also developed a proprietary technology that strips imaging data of personal information prior to sending images to the Arterys cloud that complies with privacy regulations prescribed by HIPAA. “The ability to leverage the cloud computational power while protecting confidential patient information provides the physicians additional convenience and accessibility.” Beckers says. He says that doctors can access imaging data anywhere in the world through a standard Chrome browser.
Beckers and his colleagues now have set their sights on new clinical targets such as precision medicine and improving the radiologist’s workflow. “While the initial application is in the cardiac space, the company is currently exploring deep learning applications to further its analytical capabilities in other areas of the body,” he says.
* Deep Learning technology is 510(k) pending at FDA. Not available for sale in the United States and not yet CE marked. May not be marketed or placed into service until made to comply with CE marking. Not Commercially available in all regions.