this The interview analysis is sponsored by Microsoft and Nvidia, and has been written, edited and published in an alignment with us. Content Guidelines sponsored by Emerj. Learn more about our thought leadership and content creation services Emerj Media Services Page.
Hospitals in the US face an unprecedented digital infrastructure crunch. After pushing for the digitalisation of health records nationwide, 96% of hospitals use a now-certified EHR system, according to the U.S. Department of Health and Human Services. 2022 JAMA Network Open Study With 43% of US adults using telehealth in 2022, we found that demand for healthcare computing power is rising rapidly.
However, many health systems feel their IT backbone is thinning. Published 2024 Peer Review Study Health Problem Scientist Using data from the American Hospital Association, only about a fifth of hospitals that had deployed some form of AI solution by 2022 has been revealed.
Meanwhile, 35% of health leaders cited in a 2023 survey conducted by healthcare IT leaders That limited budget and resources were the number one barrier to adopting AI tools in your organization.
Core clinical platforms can suffer under these constraints. Clinicians encounter poor EHR performance and downtime that their servers and networks cannot keep up.
At the same time, the Cloud Security Alliance points out that hospitals are cautious about expanding cloud capacity. Healthcare holds just 47% of sensitive data in the cloud (61% in other industries), relying heavily on aging on-premises data centers.
These tensions over capacity and security have been at the highest in C-Suite in the post-pandemic era.
Emerj Editor-in-Chief Matthew Demello recently hosted a conversation with Lyndi Wu, senior director of ecosystem business development for Nvidia's healthcare and life sciences, and Will Guyman, leading group product manager for the “AI in Business” podcast at Microsoft's Healthcare AI Models, explore how healthcare leaders can navigate these bottlenecks and future data infrastructure.
Their discussions spanned a wide range of topics, including the development of unified AI and the deployment of healthcare. Both Will and Lyndi highlighted the need for seamless integration of AI agents and the use of scalable infrastructure to accelerate the adoption and impact of AI technologies in improving patient care and healthcare operations.
The overall talk point highlighted the potential for transformation of agent AI in healthcare, and the need for the role of scalable GPU infrastructure in interdisciplinary collaboration, efficient data management, and optimizing AI system performance.
This article explains two important insights from the CX reader conversation.
- Work with teams to optimize your AI agent deployment. Working together between clinicians, developers, and data scientists to identify issues and create customized AI agents that address specific healthcare challenges.
- Save money with smart clouds. Run AI in the cloud to accurately match your GPU power to your workload needs, avoid wasted expenses on oversized, unused on-premises hardware, and pay only for the performance you need.
Listen to the complete episode below:
guest: Lyndi Wu, Senior Director of Ecosystem Business Development in Healthcare and Life Sciences, NVIDIA
Expertise: Artificial Intelligence, Business Strategy, Partner Relationship Management
Simple recognition: Lyndi is a creative, analytical senior executive with a successful track record of building high-performance teams. Prior to NVIDIA, she worked for Google for over 15 years. There, he played a variety of leadership roles, including leading business development for Google's Healthcare and Life Sciences Research Team and leading business development for Google Cloud Platform's Healthcare and Life Sciences Partnerships team. She holds a BS in Electrical Engineering from Princeton University and an MBA from Wharton School.
guest: Will Guyman, Principal Group Product Manager for Microsoft's Healthcare AI Models
Expertise: Artificial Intelligence, Healthcare, Computer Vision
Simple recognition: Will has been with Microsoft for over a decade and is working on computer vision for the Azure AI platform. He holds a bachelor's degree in mathematics and calculation design from Stanford University.
Working together between teams to optimize AI agent deployment
Over the course of the conversation, Will begins to explain that AI agents not only simply answer questions like chatbots, but also represent one of the most promising developments in healthcare.
Instead, they directly contribute to successful patient outcomes with human agents by processing management tasks, integrating and processing multiple types of healthcare data, such as clinical notes, and providing direct insights to physicians prior to interactions with patients.
For example, agents can prepare a timeline summary of patient history and alert care teams about new research findings and clinical trial opportunities related to a particular case.
It also emphasizes that this is important given the scale of healthcare data exponentially larger than what is found in industries such as streaming, making it impossible for human teams to manage effectively. He points out that the development of agent workflows relies on close collaboration between clinicians, developers and data scientists.
He clarifies how clinicians explain their problems, and developers build tailored solutions and create a bridge between the frontlines of care and the technical team working on advanced AI models. Agents then suggest that they stand out as hero scenarios that could transform healthcare operations and significantly improve efficiency and outcomes.
Armed with a strong background in infrastructure development, Lyndi builds on these ideas by expanding the way in which agent AI systems work together not only individual agents, but often multiple agents.
She explains how agents frequently feed all inputs to orchestration or coordination agents before delivering the required answers or performing the intended task.
Layered, interconnected setups of agent systems involve considerable complexity, especially compared to those previously supported the most advanced Generated AI (genai) use cases, so that everything runs efficiently, with workstreams and computational processes in parallel.
However, she emphasizes that GPUs alone are not enough to provide the performance and ease that health care providers need. Nvidia's software stack, including optimized containers like NIMS, allows GPU acceleration to be usable, scalable and secure.
Throughout the podcast, Lyndi highlights these points by focusing on the power of partnerships with Microsoft. This comes from combining NVIDIA's software with GPU stacks with Azure scalability and security.
“At Nvidia, what we're trying to do is to build a full-stack software solution that seamlessly integrates with the broader infrastructure and computing needs of healthcare organizations, whether running locally on devices for digital or robotic surgical purposes or scaling the entire cloud environment.
We don't do this alone. One of the greatest strengths of our partnership with Microsoft is its ability to support Edge-to-Cloud deployments with enterprise-grade reliability. ”
– Lyndi Wu, Senior Director of Healthcare and Life Sciences Ecosystem Business Development, Nvidia
Save money using Smart Cloud
Additionally, Lyndi offers an example of Nvidia's collaboration with Microsoft. Here we will make sure that your Azure software stack is set up and ready to go out of the box, and significantly reduce the time until your first inference. That means someone can get up and run with use cases and workflows in just a few minutes, but this is not an exaggeration, she says.
She also highlights the scalability benefits of using cloud systems. Unlike on-premises, where a company is limited to a single type of GPU and even if it is more powerful and expensive than necessary, it needs to maximize its utility.
She explains the advantage of running in the cloud is that the company can carefully balance the cost and performance by choosing the right size and optimization, choosing the right GPU type and instance according to the workload.
“As an example, we employ radiological imaging. These workloads are not so important, and analysis can occur overnight, making them a major factor in real-time applications like digital humans interacting with patients. Like these scenarios, lathams need to be measured in microseconds to ensure a seamless, human-like experience.”
– Lyndi Wu, Senior Director of Healthcare and Life Sciences Ecosystem Business Development, Nvidia
Pulling from Lyndi's point of cloud efficiency, we explain that various AI use cases have very different performance and infrastructure requirements.
He cites examples of administrative tasks focused on text processing, focusing on text processing, as the cost of text model inference has declined sharply in recent years. However, as healthcare teams move towards image-rich tasks, such as using complex medical image data, they need to carefully plan and determine scenarios to optimize the required capabilities.
He advises healthcare providers to start by identifying their final ROI. That might be:
- Save time for radiologists, clinicians, or administrators
- Improving patient outcomes
- Reduce the risk of missing something important
- Reduces turnaround time
Once you have defined the target ROI, you can determine which AI model to use, estimate the data volume, and determine a quality threshold that is sufficient to begin the deployment, even for evaluation purposes. He quotes a specific success story:
“The University of Wisconsin has done an incredible job of AI for medical imaging. We have identified chest x-rays as one of the most common imaging procedures and most are normal. If AI gets the first pass by deploying the foundation model from AI concrete and customizing it for reliability. Not only did it reduce the workload of radiologists, it was also able to measure and optimize large-scale quality.”
-WilGuiman, Principal Group Product Manager for Microsoft's Healthcare AI Models
The above use cases achieved 99% accuracy in identifying abnormal X-rays, reduced radiologist workload by 42%, and significantly improved turnaround time.
Separating his points for a healthcare leader presents a strong end-to-end approach.
- Set the appropriate AI models, data volumes, and scaling strategies.
- Define the use case
- Determine the required quality threshold
- Plan how to measure ROI
