August 6, 2024 | GE HealthCare and Amazon Web Services have announced a strategic collaboration to develop purpose-built foundation models and generative artificial intelligence (AI) applications designed to help clinicians improve medical diagnostics and patient care.
“At GE HealthCare, deep learning is not new,” said Parminder Bhatia, Chief AI Officer at GE Healthcare. “We have been working on this for more than a decade.” This year GE HealthCare topped a U.S. Food and Drug Administration (FDA) list of AI-enabled device authorizations with 72.
The collaboration with AWS is also not new. In 2021, the two companies announced a strategic collaboration focused on AI and cloud-based imaging solutions, integrated data, and clinical and operational insights to hospitals and healthcare providers. The terms of the newest announcement focus on collaboration to develop purpose-built foundation models and generative artificial intelligence (gen AI) applications designed to help clinicians improve medical diagnostics and patient care.
GE HealthCare mentioned three Amazon services specifically that they plan to use as part of the new collaboration. GE HealthCare intends to use Amazon Bedrock, a fully managed service that provides secure access to the industry's leading foundation models, to create and deploy bespoke gen AI applications. GE HealthCare plans to build and scale its own proprietary gen AI applications for healthcare use cases with an aim to enhance efficiency, care delivery, and the patient experience.
GE HealthCare’s internal developers plan to use Amazon Q Developer, a generative AI–powered assistant to accelerate software development by generating real-time code suggestions, securely completing tasks, and more. The company also expects to use Amazon Q Business to explore the intersection of multi-modal clinical and operational data with an aim of reducing the cognitive burden on physicians, enabling personalized care, and increasing efficiency.
Finally, GE HealthCare plans to modernize its suite of applications with its own foundation models developed on Amazon SageMaker, a fully managed service to build, train, and deploy machine learning (ML) models. By developing its own foundation models specialized for medical use cases, GE HealthCare intends to accelerate the development and deployment of web-based medical imaging applications and integrating these foundation models to drive efficiency, interoperability, and improve user experiences across the company’s equipment and software solutions. Customers could use GE HealthCare's generative AI-powered applications, that will integrate with AWS HealthLake and AWS HealthImaging, to quickly and securely analyze various types of patient data, leading to improved clinical efficiency and better patient care.
Since the pandemic, Bhatia said, many of GE HealthCare’s customers have moved to the cloud. “Part of bringing some of these capabilities… into the cloud as part of our digital strategy [is] so that our customers can get these capabilities wherever they would prefer.” Plus, collaboration with Amazon will bring scalability, security, and reach to GE HealthCare’s efforts, Bhatia explained.
“We have built 72 FDA-approved app authorizations,” Bhatia said. “How do we make it easier and faster for us to build and scale our own proprietary generative AI applications that can actually speed up the development of innovative healthcare applications that can go across our devices and software solutions?” AWS’s generative AI tools make it easier to build and scale gen AI applications, he said “With Bedrock you have a lot of off-the-shelf large language models, foundation models, which you can select and customize for your specific use cases.”
Security and privacy are paramount to us, he added. “AWS enterprise-grade security and privacy really bring in a lot of value as we are building and scaling these things.”
Finally, Bhatia said Amazon’s size and global reach coupled with GE HealthCare’s own size amplifies the reach either company can have on its own. “How does one plus one become greater than two? Their size and global reach makes it well-suited to help realize that next phase of [our] generative AI journey.”
Generating Connections, Not Just Data
Generative and foundation models are helpful in understanding and synthesizing unstructured data, Bhatia explained. “I think there is a disconnect when people talk about generative AI. It’s not just about generating new content, but even in the existing content, how do you get to that right information?”
For example, patients amass clinical notes in different places, from different doctors. Here, generative AI can help build and train models and uncover connections within those disconnected data.
AI tools already improve healthcare outcomes in many ways. “We have capabilities in ultrasound where we added Caption AI. All it does is help a sonographer with 1 to 15 years’ experience come to the same level field. We give them guidance; a co-pilot kind of experience… ‘Move a little to the left; move a little right. Now you have a diagnostic-quality image,’” he said.
GE HealthCare’s AIR Recon DL product suite has been trained with input from 20 million patients. The deep-learning-based reconstruction algorithm enables radiologists to achieve pin-sharp images quicker and removes noise and ringing from raw images. Sonic DL is a deep learning technology that acquires high-quality magnetic resonance (MR) images up to 12 times faster than conventional methods, enabling cardiac imaging within a single heartbeat. “It’s reduced time for MR scans by 50%,” Bhatia said.
But building algorithms like these is time consuming. “It takes a lot of effort to build, train, scale these models,” Bhatia explained. “How do we actually accelerate the innovation cycle [so] that we have faster iterations?”
He’s betting gen AI is the answer. In the past ten years, he observed, machine learning models have been built for specific use cases—text or image. “Now you can start to combine those things in a meaningful way, which becomes a great complimenting component for healthcare,”—a distinctly multimodal discipline, he noted, with a lot of unstructured (and unused) data.
“One of the anticipated outcomes of this work is GE HealthCare’s ability to really use these capabilities to… actually combine data to help clinicians get the most out of the data,” he said, “making it easier for clinicians to provide more specific and personalized care.”
Bhatia foresees new models impacting CT/MR, ultrasound, and x-ray imaging across disease areas including oncology, cardiology, and neurology, “bringing in more AI-shine into these products,” he said, a term he used to refer to AI-first thinking to further AI capabilities.
“I think partnerships like these really help us accelerate and bring these things [to fruition] at a faster pace,” he said.