SRI’s AI platform is rewriting rural healthcare

---

No hospital? No problem. SRI’s data fix could save rural medicine. 

• Rural healthcare access is declining, and mobile care is a promising fix. 
• Interoperability is the central obstacle to healthcare innovation. 
• AI alone isn’t enough — verification is what makes it trustworthy. 

---

Last year, SRI announced its work on mobile healthcare to improve access and outcomes across rural America. But delivering quality mobile healthcare is about more than just equipping a truck with high-tech equipment and a healthcare professional. Today, healthcare is about data: electronic health records (EHRs), diagnostic data in numerous different formats, lab tests, and much more. This information needs to be shared in a way that’s accurate, secure, and also user-friendly for clinicians. To support the larger PARADIGM effort, SRI is building a healthcare data solution in support of ARPA-H’s mobile care delivery platform.  

Interoperability: The key to healthcare innovation 

When it comes to healthcare data, says SRI technical director Grit Denker, the big problem is interoperability. Today, most hospitals function within closed ecosystems created by large vendors. Incorporating new medical devices and technologies is costly and can require months of effort.  

“What if every clinician on a patient’s care team shared a single, real-time, complete health picture, so every patient receives the best possible care? POET makes that reality.” — Grit Denker, SRI technical director

Rural hospitals struggle with legacy IT systems that make it challenging to incorporate the latest medical advancements. If that’s the case for brick-and-mortar hospitals, how will rural healthcare systems possibly incorporate next-generation mobile care vans?  

They need a data solution that can translate between legacy IT systems and new innovations and is easy to use, affordable, and scalable. SRI has proposed the Platform for medical interoperability (POET).  

“What if every clinician on a patient’s care team shared a single, real-time, complete health picture, so every patient receives the best possible care?” asks Denker. “POET makes that reality.”  

POET data for health-tech integration  

POET takes advantage of large language models (LLMs) to automatically generate the software that can facilitate data flow between devices and hospital IT systems. POET seeks to reduce technology integration timelines from months to weeks or even days.  

The challenge: LLMs can hallucinate, and they lack built-in verification. To create a system that’s robust enough to handle healthcare data and diagnostic information, SRI’s system goes beyond out-of-the-box LLMs.  

POET pairs LLMs with formal verification methods in an iterative loop. SRI has been advancing formal verification capabilities since the 1970s and has developed new approaches to formal verification that can be used to mathematically prove the accuracy of LLM-generated healthcare integration software. 

The POET team is also developing a user interface to assist hospital staff in refining specifications, implementing a chat feature to enable staff to investigate potential failures, and creating digital twins for devices and hospital systems to test the code before it’s deployed.  

How POET can unlock mobile care (and more) 

POET connects medical devices to electronic health records (EHRs) and a task guidance system to provide real-time decision support for clinicians. POET will provide a comprehensive patient health profile, including electronic health record systems, external lab systems, imaging platforms, and more. 

“POET will help clinicians deliver expert care everywhere,” emphasizes Stephane Graham-Lengrand, director of SRI’s Computer Science Laboratory and the principal investigator on this work. “And it’s not just for mobile care. This same open-source approach can help hospitals solve the interoperability challenges that slow down innovation. In healthcare, efficient data interoperability equals improving and saving lives.”  

————————————————————————————————————————— 

This research was funded, in part, by the U.S. Government. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the U.S. Government.