The healthcare tools of the future often start as research projects in academic medical centers, but they may not break out into widespread clinical practice until some sort of logjam gives way to make them workable and affordable.
GE Healthcare, a unit of Boston-based industrial giant General Electric, recognized some limits on MRI scanners like its own, and started collaborating with Stanford University physicians more than five years ago on next-generation MRI technology to better diagnose people with heart disease. The new scans resulting from this work show a patient’s beating heart like a 3D animated movie—and the image can be rotated and viewed from any angle. Up until now, these scans have been used only at Stanford and nine other medical research centers.
Chicago-based GE Healthcare is now about to go to market with the new magnetic resonance imaging technique. But it took several extra layers of technology, and a collaboration with San Francisco startup Arterys, to make that possible. That’s because GE Healthcare, by tweaking its traditional MRI scanners, was capturing so much raw data on the heart that on-site hospital scanners and computers couldn’t assemble it into images that doctors could intuitively interpret.
To solve that problem, Arterys tapped into technologies used to develop three-dimensional video games like “Super Mario 3D World.” Arterys feeds the gigabytes of MRI information into its Web-based software application, which takes advantage of the outside computing power now broadly available from the network of servers-for-hire we call “the cloud.” Those servers rely on specialized logic chips called graphic processing units or GPUs, which are commonly used to produce 3D animations.
Less than five years ago, those GPUs weren’t available from commercial cloud services, says Arterys co-founder and chief technology officer John Axerio-Cilies.
“In 2012, when we started, Amazon had just released the GPUs into the public cloud,” Axerio-Cilies says. That opened a door for GE Healthcare’s next-gen cardiac MRI. Arterys uses Amazon’s Web servers to analyze data from the scans. “A nexus of events in just a couple of years allowed this to happen,” he says.
The commercial debut of the new GE Healthcare system will mark an industry milestone, partners in the collaboration say.
“This is one of the first applications of cloud computing in a clinical application,” says Arterys CEO and co-founder Fabien Beckers.
A product version of the new MRI system, dubbed ViosWorks, will be placed in ten pilot sites before the end of the year. Plans are to make it commercially available in 2017.
GE Healthcare will be the first MRI vendor to shift its data processing to the cloud, says Eric Stahre, CEO of GE Healthcare’s MR unit.
“It’s the first example of where we’re going as a company,” Stahre says. “We see this as the first of many.”
In addition to its work on MRI, GE Healthcare has looked to enhance other medical imaging technologies through computing power. In one project, the GE division’s IT team has written software to speed up the processing of data from CT scans of the brain so that doctors can make quicker treatment decisions for suspected stroke victims.
GE Healthcare’s CEO John Flannery has predicted a 50-fold increase in the flow of data from healthcare devices by 2020. To handle that flow, the company late last year announced the creation of GE Health Cloud, a Web-based data processing and treatment collaboration platform that it says can connect to more than half a million of its imaging devices. The company is inviting 3D medical imaging developers to work within the platform.
The data processing for ViosWorks scans will eventually be shifted from Amazon’s servers to GE Health Cloud. The “4D Flow” system captures as much as five gigabytes of data, compared to about 100 megabytes in a conventional MRI scan, Beckers says.
The U.S. Food and Drug Administration has cleared both the Arterys and GE Healthcare elements of the system for the U.S. market, the companies say. Stahre says GE Healthcare is to begin its marketing rollout of ViosWorks next month.
The next hurdle will be to persuade doctors and medical insurance companies that the more elaborate scans will improve diagnosis enough to lead to better patient outcomes.
“You still have to prove your worth,” Stahre says.
Here’s the problem GE Healthcare and its research partners were trying to solve when they turned the company’s conventional MRI scans into multi-dimensional displays:
Say a patient suffers from symptoms of heart disease (but not an immediate crisis condition such as a heart attack.) A cardiologist suspects that one of the patient’s heart valves is worn out or malformed. As the heart beats, driving blood from one chamber to the next, the valves are supposed to close up fully and prevent blood from leaking back into the heart chamber it just left. The valves then reopen to let the next surge of blood come through.
Current imaging tools such as echocardiogram/ultrasound and MRI can help doctors peer into the structure of the heart and assess the condition of the valves. But Beckers says it can still be hard to tell just how leaky a valve is. That means a doctor has to make a difficult call: either send the patient for invasive heart surgery, with its significant risks, or try to manage the condition with drugs, which may not be enough if the valve problem turns out to be worse than expected.
With the technology now in use, physicians do their best to guide an MRI technician to capture 2D cross-section images of the planes of the heart where the suspected faulty valve resides. The process, which can take as long as 90 minutes, can be stressful for sick patients, who may be asked to stay still and hold their breath repeatedly, Beckers says.
In addition to spending time guiding the MRI scan, doctors must interpret the images—literally drawing on the scans to outline the contours of the heart muscle, GE Healthcare cardiac MRI leader Anja Brau says. They then use software to calculate blood flow through the heart chambers, valves, and vessels.
GE Healthcare wanted to provide doctors with more complete, three-dimensional data, not only of the whole cardiovascular structure, but also on the flow of blood over time. This kind of imaging is called 4D flow. The data is collected for three spatial dimensions, along with a fourth dimension—time. The company also wanted to speed up the data capture, to save time for the doctor, the technician, and the patient, as well as minimizing time on the scanner. It reduced the exam time to about 10 minutes, and patients don’t need to hold their breath.
Not to get too technical, but GE Healthcare made these advances by re-programming the sequence of radiofrequency pulses aimed at the chest during an MRI scan. This is a common method of tailoring the performance of medical imaging equipment for specific diagnostic purposes.
What GE Healthcare’s efforts yielded was a stack of many 2D slices (or planes) of the full chest, including the heart. But physicians would still have to pore through all those layers to find the suspected heart defect. Then Albert Hsiao, a Stanford clinician with software experience, worked up a prototype graphics program to assemble all those 2D slices into a 3D volume that could show the blood flow in the beating heart. Hsiao demonstrated it to Brau in a basement lab at Stanford’s hospital.
“That was the beginning,” Brau says.
The idea led to the founding of Arterys in 2011 by Hsiao, Beckers, Axerio-Cilies, and Stanford radiology professor Shreyas Vasanawala, co-director of MRI at Stanford University. Hsiao is now Associate Director of Cardiovascular Diseases for the Center for Translational Imaging and Personalized Medicine at UC San Diego.
In addition to its work constructing 3D images from MRI data, Arterys also developed software to protect patients’ privacy by stripping their personal identifying information from the scanner files before they are uploaded to the cloud.
In March, GE’s corporate venture capital arm GE Ventures was one of the investors joining in Arterys’s $12 million Series A financing round. The round was led by Emergent Medical Partners, with participation from the Stanford-StartX Fund, Norwich Ventures, and prior investors Asset Management Ventures, AME Cloud Ventures, and Morado Ventures. The new capital followed a $2 million seed funding round.
Arterys stands to receive licensing revenue from GE Healthcare, which is acting as the startup’s distributor for the use of its technology in the ViosWorks cardiac MRI system.
German industrial giant Siemens, a major competitor with GE Healthcare in MRI, does not have a 4D flow MRI system on the market, a Siemens spokeswoman says. But such a system is in the works, she says.
Stahre wasn’t specific about the pricetag for adopting the new MRI system, besides describing it as “cost-effective.”
“We’re trying to make it available to any patient around the world,” Stahre says.
Insurers reimburse hospitals for conventional MRI scans at a fixed rate, even though some scans take longer than others, Brau says. GE Healthcare doesn’t say whether it would seek a higher payment amount for ViosWorks scans. But the company makes a case that hospitals would save money by using the company’s faster scans and its automated analysis.
With ViosWorks exams taking as little as 30 minutes for the set-up time and scan, rather than as much as 90 minutes for conventional MRIs, medical centers could examine more patients per day with the same scanner, the company maintains. Doctors wouldn’t need to spend time guiding an MRI technician or calculating blood flows—those figures would be generated by the Web-based software within minutes.
And hospitals would still be able to use their existing MRI machines—GE Healthcare can modify its own scanners that are as old as 15 years, Stahre says. A cardiologist can look at the patient’s 3D scan outside the hospital using the browser on a Web-connected computer.
Over time, GE Healthcare and Arterys expect to use machine learning to extract insights from the aggregate MRI data analyzed from large numbers of scans taken from different (anonymous) individuals. The Arterys analytics software would search for image features that can be used to predict the advance of a heart condition, and the likelihood that a certain treatment will be successful.
Both Arterys and GE Healthcare plan to extend the use of their technologies to illnesses beyond heart disease.
Stahre says customizing the radiofrequency pulse sequences of MRI scanners could yield valuable images of the brain and musculoskeletal structure, and diagnostic insights in the treatment of diseases including cancer and multiple sclerosis.
“We’re just scratching the surface of the potential here,” Stahre says.