The Hospital Ultrasound Revolution Has a Governance Problem
Device fleets are manageable. Everything that determines whether clinicians can use them safely at scale is still being figured out
At HIMSS 2026, GE HealthCare is taking the stage with Oregon Health and Science University to present what it calls an "enterprise framework for scaling handheld ultrasound." It is part of a wider industry effort to position AI-assisted point-of-care ultrasound, known as Pocus, as ready for system-wide deployment.
FDA submissions for AI-enabled medical devices have more than doubled in four years, climbing from 114 in 2020 to 235 in 2024, according to data presented at a November 2025 expert panel hosted by the Global Ultrasound Institute. GE's Vscan Air SL, which includes Caption AI for real-time cardiac guidance and ejection fraction estimation, is among the devices driving those numbers.
Deploying these devices across a health system requires more than hardware and a fleet management portal. It requires credentialing infrastructure, image archiving workflows, and institutional governance, and on all three fronts, the field is still catching up.
GE's Tools Manage Devices. They Don't Manage Governance
GE's enterprise offering includes MyDeviceHub, a web-based portal that gives IT administrators centralized control over all Vscan Air devices, grouping them by department, managing user access, configuring settings, and tracking usage. On the IT and biomedical engineering side, that capability is real.
The image archiving layer is a different story. GE's own support documentation states MyImageCloud is designed for individual users rather than enterprises or large clinics, and caps storage at 500 exams per year.
Hospitals that opt into the Fleet Solution tier, the enterprise track, lose access to MyImageCloud entirely. GE's FAQ states that "for Fleet-enabled probes, MyImageCloud is not available."
In practice, it means the archiving question is left to each institution to solve independently, with no documented integration pathway from GE for non-Centricity systems. For a hospital like OHSU, which runs Epic and a Philips-based picture archiving and communication system infrastructure, that is an open integration task.
Credentialing is similarly unsettled at an industry level. An October 2025 study published in Jama Network Open, based on interviews with 20 Pocus leaders at U.S. health systems, found no defined national standards currently exist to guide institutions in regulating Pocus use or clinician best practices. Policies varied widely in scope, requirements, and enforcement.
Even at institutions with finalized policies in place, the gap between policy and practice was notable. Only 35 percent of surveyed institutions reported that more than 75 percent of diagnostic Pocus exams were performed by credentialed clinicians.
One participant described the workflow dependency: "Just having a lot of machines doesn't mean people are going to do it. But if you have folks work too hard to find devices, they're probably not going to. The second piece of that is having a system in place for image management, so that if you're doing these exams, we track your portfolio and then have a process for providing quality assurance."
The same study captured how device partnerships can expose the governance gap rather than close it. A participant noted: "What drove the policy was that our institution entered a partnership with [device] in 2019, with the goal of expanding Pocus to our regional health network. And so through that partnership we recognized a big gap in compliance and credentialing policy when it comes to Pocus."
AI Raises the Stakes on a Problem That Isn't Solved Yet
Standard Pocus credentialing frameworks were designed around operator skill, whether a clinician can acquire and interpret an image. AI-assisted devices introduce a layer those frameworks were not built to address: what happens when a clinician acts on an AI-generated output, and who bears responsibility if that output is wrong.
No standard answer exists. The Compass-AI survey, published in July 2025 in the Ultrasound Journal, surveyed 1,154 healthcare professionals across 20 countries, with roughly 20 percent from North America, and found 77.4 percent had never used an AI or machine learning tool in patient care.
When asked to identify the single greatest barrier to adoption, 27.1 percent cited training and education. A further 80.5 percent agreed that a lack of clear institutional or professional guidelines on AI use in Pocus contributed to their hesitation.
Dr. Renee Dversdal, a professor at OHSU and chief medical officer at VAVE Health, addressed the transparency problem directly at the GUSI expert panel: "Clinical integration requires transparency. Clinicians need to understand not just what the AI is telling them, but how it arrived at that conclusion and what confidence level we should have in that interpretation."
AI-assisted Pocus tools are reaching hospitals faster than the training programs, governance policies, and documentation standards needed to use them safely at scale.
Whether any vendor's toolset is equipped to solve that problem, rather than the narrower device management problem the current generation of tools was built for, remains an open question.