AI-based PET, MRI Scans Bring Life-Saving Technology to More Patients via the AWS Cloud

Subtle Medical uses technology from AWS and NVIDIA, an APN Advanced Technology Partner, to reduce scan times up to 75%, making the technology accessible to more patients

A Remarkable Response to Rising Medical Costs

It's no secret that medical costs are under pressure; insurers, the government, and employers who typically pay the premiums all want the expenses of high-priced medical technologies contained. It's a pressure perhaps only rivaled by the equally forceful demands of consumers for greater access to those technologies and their life-saving benefits. The two could not be in greater conflict. Or so you might think.

One of the innovative exceptions is a new, FDA-approved medical imaging technology deployed at multiple facilities, from Hoag Memorial Hospital in Southern California to Connecticut’s Middlesex Health network. SubtlePET from Subtle Medical vastly reduces the duration of scans, so it's more comfortable and tolerable for more patients. And because the scans are faster, providers can serve additional patients more efficiently, reducing overhead costs per scan and further increasing revenue and access.

Because the technology is software-based, healthcare providers can scale out the technology to the imaging equipment they already have without investing in costly new imaging systems until they’re ready to do so. That’s a benefit not just to large hospitals and medical systems, but also to smaller providers, particularly in underserved areas, that might have no other practical way to bring this level of imaging access to their patients.

All this is a big contrast to traditional medical imaging. While detailed imaging of a patient’s organs and internal structures can give doctors life-saving insights to diagnose and treat disease, full-body PET (Positron Emission Tomography) scans typically take up to 45 minutes, and MRIs (Magnetic Resonance Imaging) can take one to three hours. That's countless hours and billions of dollars a year—for those patients who can tolerate the process.

But many others can’t, such as children and the elderly who may not be comfortable lying still for an hour or more in an imaging chamber. Then there are potentially harmful chemicals, such as gadolinium, used during “contrast” scans; the more such chemicals are needed, the greater the risk to patient safety.

Using AI for Better Image Acquisition

Researchers have long sought ways to address these challenges. While other imaging solutions focus on image analysis, Subtle Medical’s suite of software solutions focuses on the acquisition phase of imaging workflow. It uses artificial intelligence algorithms developed through training on robust paired datasets covering a wide range of clinical indications, patient cohorts, vendors, and scanner models. The Subtle Medical solution is compatible with every brand of PET machine. It requires no change in the radiology tech's or radiologist’s workflow, so healthcare providers can readily adopt it without the need for retraining their staff members.

By using AI in the image-acquisition phase, Subtle Medical enhances the image quality from “noisy,” faster scans, making those faster scans as effective as traditional, slower scans. The company says that PET scans with SubtlePET can be done up to four times faster. That’s the case at Middlesex Health, which cut average scan time per patient from 24 minutes to 6 minutes.

And while faster and more efficient, the technology provides “a superior patient experience [with] equal image quality, not otherwise impacting workflow,” according to Michael Brant-Zawadzki MD, FACR, Senior Physician Executive at Hoag Memorial Hospital Presbyterian in Newport Beach, California, which uses SubtlePET.

The Cloud Makes It Real

Making better imaging technology safer and available to more people isn’t a side benefit for Subtle Medical. It’s what the company was founded to do.

“I was inspired by my research at Stanford University to make medical imaging more accessible,” says CEO Enhao Gong, who founded Subtle Medical along with Dr. Greg Zaharchuk, Neuroradiologist at Stanford University. “We were doing tremendous research, but I saw that it could take medical device manufacturers 10 years to get innovations cleared for new products. That’s what got us thinking about finding ways to make existing imaging equipment smarter, more effective, and more productive.”

To turn that vision into reality, Gong, Zaharchuk, and their colleagues turned to Amazon Web Services (AWS) and NVIDIA, an AWS Partner Network (APN) Advanced Technology Partner. Subtle Medical uses Amazon Elastic Compute Cloud (Amazon EC2) P3 instances with NVIDIA GPUs and CUDA-X architecture for deep learning, which enables it to maintain image quality with briefer scans. Subtle Medical researchers use TensorFlow for their inference framework and also use Keras and PyTorch for research and development projects.

“We needed to make a huge difference in inference time, and NVIDIA enables us to make that difference," says Gong. "We wouldn't have a viable product without it, one that delivers better patient care, quality, and flexibility."

The other crucial part of the Subtle Medical solution is making the technology accessible, reliable, secure, and cost-effective for healthcare providers of all sizes throughout the US. That's a tall order. And that's where AWS comes in.

“Hospitals and imaging centers want to adopt this solution without burdening their IT departments to acquire GPU expertise and build and maintain costly data centers or mini-clouds. They want to be successful with their deployments with the least amount of effort and investment… AWS makes this possible.”

- Enhao Gong, Founder and CEO, Subtle Medical

"The biggest issue we have is scalability, and AWS solves that issue for us," says Gong. “Hospitals and imaging centers want to adopt this solution without burdening their IT departments to acquire GPU expertise and build and maintain costly data centers or mini-clouds. They want to be successful with their deployments with the least amount of effort and investment. They want to start simply, with one or two scanners and roll out the solution more broadly when and as they're ready. AWS makes this possible."

For example, the solution uses AWS IoT Core to securely manage private data with just a web app on the front end, avoiding costly and unreliable VPNs. And it isn't only hospitals and other providers using the Subtle Medical solution that benefit from AWS; so does Subtle Medical. The company uses AWS managed services to support its solution, including SOC 2 and HIPAA compliance, with a lean IT team.

“We use AWS to vastly reduce the personnel and expense of managing the solution and its compliance,” says Gong. “We use AWS to avoid the nightmare of obtaining FDA approval for hardware and then maintaining a hardware-based business, with all the implications that brings for taxes, finances, inventory and warehousing, distribution, and more. We're a software company, and we want to keep it that way. AWS does what it does best, so we can do what we do best.”

About Subtle Medical

Subtle Medical is a healthcare technology company working to improve medical imaging efficiency and patient experience with innovative deep-learning solutions. Its team is made up of renowned imaging scientists, radiologists, and AI experts from Stanford, MIT, MD Anderson, and more.


To bring the benefits of medical imaging to more people, especially underserved populations, Subtle Medical needed to make the process faster and safer and to deliver its solution in a way that healthcare providers can readily adopt and use with minimal investment, retraining, and overhead.


For the AI-based deep learning technology necessary for faster and fuller image acquisition, Subtle Medical chose NVIDIA. And for the platform to scale this solution to any healthcare provider of any size, reliably, cost-effectively, flexibly, and securely, it chose AWS.


Subtle Medical’s SubtlePET solution scans up to four times faster than traditional scans with no loss of image integrity, making the technology accessible to patients who can’t tolerate longer scans. Shorter scans mean less delay in scheduling scans. And those scans require less potentially risky "contrast" chemicals. The solution is as cost-effective for Subtle Medical to develop and deliver as it is for healthcare providers to deploy and use.


NVIDIA is an accelerated computing company. In 2006, the creation of its CUDA programming model and Tesla GPU platform brought parallel processing to general-purpose computing. A powerful new approach to computing was born. From the world’s largest supercomputers to the vast data centers that power the cloud, this new computing model is helping to answer complex questions, discover new science, and bring amazing capabilities to mobile devices.

Published April 2020