21st Edition | Commercial Radiology Foundation Models Are Here

Campbell Arnold
Aug 11
5 min read

“They're trying to move the decision-making point as far forward as possible because they don't have enough doctors.”

— David Specht, CEO MAUI Imaging

Welcome to Radiology Access! Your biweekly newsletter on the people, research, and technology transforming global imaging access.

In this issue, we cover:

HOPPR Releases First Commercial Product—Aims to Make AI Development Easier
MAUI Imaging Secures $14M to Push Ultrasound’s Boundaries
1,247 FDA-Authorized AI Medical Devices, and Counting!
RSNA Launches Multimodal Aneurysm Detection Challenge

If you want to stay up-to-date with the latest in Radiology and AI, then don't forget to subscribe!

HOPPR Releases First Commercial Product—Aims to Make AI Development Easier

A bellwether for how foundation-models-as-a-service will impact the radiology industry.

Fresh off a $31.5M Series A and a hiring spree, HOPPR has officially launched its first commercial product: the HOPPR Marie Curie Chest Radiography Foundation Model. Now available via HOPPR’s API, the release provides an early look at “foundation-models-as-a-service” (FMaaS) in radiology—complete with built-in tools for fine-tuning, deployment, and regulatory support.

I sat down with members of the HOPPR team for a demo of their initial platform rollout. The foundation model is a vision transformer that’s intended to be integrated into a customer's medical devices or clinical workflow—designed to be fine-tuned for specific clinical needs and use cases. While the current release only supports binary classification tasks, the company’s roadmap includes multiclass, localization, and segmentation functionality. Support for loading prior weights and expanded hyperparameter tuning is also planned in future releases.

Even in this early version, the HOPPR Curie model posted impressive internal results, performing comparably to Microsoft’s MedImageInsight and Google’s MedGemma across 28 classification benchmarks. Even more impressive, the model is already having a real-world impact. HOPPR’s partner DeepHealth reported they were able to fine-tune and deploy models within weeks, stating that the foundation model allowed them “to move quickly and significantly reduce our development costs and improve our operational quality and effectiveness.”

HOPPR’s secure AI development platform gives downstream developers the tools and infrastructure to build clinical solutions. The development environment has built-in infrastructure for data handling and experiment tracking. It’s also connected to HOPPR’s quality management system, which ensures traceability and supports teams developing products for medical use. This reflects HOPPR’s current strategy: providing infrastructure for building medical devices. The company sees three major use cases for the platform:

Lowering model development barriers through secure, validated infrastructure
Augmenting clinical workflows to reduce burnout and reading time
Accelerating development timelines with strong performance and quality systems

What's next? HOPPR is actively expanding its platform with plans to introduce in-house data services, broaden support across imaging modalities and workflows, and roll out new model capabilities. The team has already previewed a promptable vision-language model at recent conferences, hinting at what’s to come in future releases.

HOPPR team members shared their enthusiasm for the pace of progress. “I joined less than a year ago, and we’re already releasing a commercial model,” said Kalina Slavkova. Akash Pattnaik added that he’s excited about the fine-tuning tools they’re releasing alongside the model, which “open the door for small clinical teams to build models—even without deep learning expertise.”

Bottom line: HOPPR’s release is not just a new foundation model, but a whole new business model. How will radiology companies receive foundation models as a service?—stay fine-tuned.

MAUI Imaging Secures $14M to Push Ultrasound’s Boundaries

How this device could allow you to see around and through barriers.

Photo Credit: Cheronis et al., JACEP Open 2025

MAUI Imaging has secured $14 million in Series D funding to advance its ultrasound system designed to see through barriers that block traditional devices, such as bone and gas. The technology, called Multiple Aperture Ultrasound Insonification, uses advanced signal processing to recover lost data and reveal anatomy normally hidden from view.

A recent JACEP Open article co-authored by MAUI and collaborators showed the system could deliver high-quality images in regions typically inaccessible to ultrasound, with potential to replace CT or MRI scans in trauma situations. Originally developed under a $4M DoD contract, the portable device is built for use by medics with minimal training—bringing rapid diagnosis closer to the point of injury.

With FDA clearance for standard ultrasound uses already in hand, MAUI will leverage the funding to seek expanded claims for “see-through” imaging. The round was led by Acertara (now their exclusive distributor) and brings MAUI’s total funds raised to $40M. If successful, MAUI could bridge the gap between conventional ultrasound and high-cost, fixed imaging systems—giving providers a deeper look at the point-of-care.

Bottom line: MAUI Imaging raised $14M to push for ultrasound that can bypass traditional imaging barriers.

1,247 FDA-Authorized AI Medical Devices, and Counting!

Who is leading the charge for imaging devices?

The FDA’s latest update reveals a total of 1,247 AI-enabled medical devices cleared or approved as of May 31, 2025 — an impressive jump of 297 new authorizations since August 2024. With radiology dominating the field (76.7% of all devices), it’s clear that imaging remains at the forefront of AI’s clinical impact.

The top five companies — GE, Siemens, Philips, Canon, and United Imaging — are all major OEMs with broad portfolios spanning multiple modalities. GE holds the crown for the fourth straight year with 100 FDA-cleared AI devices, and an ambitious target to double that by 2028. Siemens follows in second with 80 devices.

Aidoc leads the pack among AI-first medical imaging companies with 30 FDA-authorized tools. Two of my personal favorites also made the Top 20 list: Subtle Medical and Hyperfine. The report also notes the FDA’s plans to flag devices using foundation models in the future — a move that could bring more transparency around device development practices.

Bottom line: The FDA’s AI device list grew by over 30% since August, underscoring how AI in imaging is scaling rapidly!

RSNA Launches Multimodal Aneurysm Detection Challenge

Are you ready to put your algorithm to the test?

The Radiological Society of North America (RSNA), in partnership with European Society of Neuroradiology (ESNR) and the Society of Neurointerventional Surgery (SNIS), has announced the 2025 Intracranial Aneurysm Detection AI Challenge. This pioneering competition focuses on developing AI tools to detect and localize intracranial aneurysms across multiple imaging modalities—CT angiography, MR angiography, and conventional MRI—mirroring real-world clinical scenarios.

Intracranial aneurysms affect over 3% of the global population and often go undiagnosed until rupture, which carries high risk of severe complications and death. Early and accurate detection is critical to guide timely treatment and prevent catastrophic outcomes.

The challenge features a large, diverse dataset with over 6,500 imaging studies and 3,500 annotated aneurysms from 18 sites worldwide, curated by expert neuroradiologists and neurointerventionalists. Participants will develop AI models to detect aneurysms in 13 specific intracranial locations, promoting opportunistic screening on routine brain scans done for other indications.

The challenge is hosted on Kaggle and open to all researchers through October 14th, with $50,000 in prizes awarded to winners at RSNA 2025 in Chicago.

Bottom line: RSNA launches an aneurysm detection challenge.

Feedback

We’re eager to hear your thoughts as we continue to refine and improve RadAccess. Is there an article you expected to see but didn’t? Have suggestions for making the newsletter even better? Let us know! Reach out via email, LinkedIn, or X—we’d love to hear from you.

References

Disclaimer: There are no paid sponsors of this content. The opinions expressed are solely those of the newsletter authors, and do not necessarily reflect those of referenced works or companies.