27th Edition | Democratizing MR and AI

Campbell Arnold
Nov 4
5 min read

“We must move beyond a “donate-and-forget” culture that, while often well-intentioned, delivers limited and short-lived impact.”

— Derek K Jones et al., 2025

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

In this issue, we cover:

Small, but Mighty: Why Lower Strengths May be the Key to Stronger Imaging Access
Low-Field Brain Volumes Show High Reliability and Correspondence
The End of "One-Size-Fits-All" AI

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

Small, but Mighty Scanners

Why lower magnetic strength might be the key to stronger global imaging access.

MRI is a cornerstone of modern medicine, provided you’re in a well-resourced hospital in a high-income country. However, for billions of people worldwide this transformative technology remains out of reach. High device costs, complex infrastructure, and the need for specialized personnel have made MRI one of the most inequitable technologies in global health. In a recent BJR Open review, a multidisciplinary team of clinicians, physicists, and global health experts make the case for how low-field MRI could change that.

Low field systems operating below 100 mT are mobile, lower-cost, and increasingly powerful, thanks to rapid advances in hardware, acquisition physics, and AI-driven reconstruction. The authors challenge the long-held assumption that “low field means low quality,” urging the field to see low-field MRI not as a compromise, but “an additional frontier—a complementary domain for innovation.” Rather than replicating high-field performance, their goal is to reframe MRI as a technology that can be adaptive and globally deployed.

Key points from the review include:

Clinical and Research Value: Low-field MRI can generate clinically meaningful images, but also rich research data. Lower costs and increased mobility can enable new large-scale studies in areas like neurodevelopment, aging, and neurogenetics, especially in regions that were previously unexplored.
Open and Sustainable Technology: The authors advocate for open-source, upgradeable, and co-developed systems that empower local teams to maintain, adapt, and scale the technology, moving beyond the traditional “donate-and-forget” model.
Capacity Building: Real-world deployment requires robust data infrastructure, training pathways, and international collaboration frameworks, not just affordable scanners.

Ultimately, the authors call for a global reframing of MRI from an elite hospital technology to a widely accessible, sustainable diagnostic tool. By investing in open and collaborative low-field solutions, the imaging community can drive local innovation, expand research capacity, and deliver transformative health impact where it’s needed most.

Bottom line: Low-field MRI isn't about stepping down in quality, it's about stepping up to expand medical imaging access and high-impact research across the entire globe.

Low-Field Brain Volumes Show High Reliability and Correspondence

How low-field scanners could open up new avenues for global neuroimaging studies.

Our previous article highlighted the incredible promise of low-field MRI to democratize global health, but for this revolution to take hold, that promise needs to be backed by hard evidence. A new research paper in Imaging Neuroscience provides exactly that, tackling the critical question: Are low-field brain volume measurements reliable and robust? Or just a poor substitute for the multi-million dollar "gold standard" machines?

To answer this question, the researchers scanned a group of 23 healthy adults on Hyperfine’s 64 mT systems and a conventional 3T high-field scanner, directly comparing the results. Their findings showed:

High Reliability: The low-field scanners produced consistent and repeatable measurements of brain structures when scanning the same person twice.
Strong Correspondence: Brain morphometry measurements from the 64 mT scanners showed high correspondence to those from 3T scanners.
Clinical Relevance: The results showcase the potential of low-field for applications like mapping brain changes during development/aging and monitoring neurological disorders.
Open Science: In a crucial move for transparency and future innovation, the authors have made their image dataset and code publicly available.

This paper provides the critical validation needed to move low-field from a promising idea to a trusted clinical and research tool. By demonstrating robust reliability and making their data open, this paper helps pave the way to finally close the global gap in neuroimaging.

Bottom line: This study provides evidence that low-field MRI delivers reliable and robust brain measurements, and offers a valuable, open-access dataset of paired 64 mT and 3T scans.

The End of "One-Size-Fits-All" AI

How companies like HOPPR offer the ability to develop in-house foundation models.

In last week’s newsletter, we explored how site-specific foundation models can outperform large open-weight ones. However, building an in-house foundation model from scratch is a monumental task, to many it’s as daunting as a marathon. But what if you could start from mile 26?

HOPPR is leaning into that vision by providing everything most institutions can't build themselves: strong initial weights and training tools so users can fine-tune their own site-specific algorithms. HOPPR just announced their second foundation model, a 2D mammography model, following their chest X-ray model release earlier this year.

Instead of a locked, one-size-fits-all product, HOPPR is providing a "powerful starting point" that serves as the essential building blocks for developing in-house breast AI algorithms. This approach empowers developers and institutions to:

Accelerate Innovation: Quickly creating and fine-tuning models for a range of tasks, from cancer detection to breast density classification.
Build Tailored Solutions: The model is designed to be easily fine-tuned on an institution's own local or proprietary datasets, enabling more "tailored, high-performing, and robust solutions."
Lower the Barrier to Entry: This strategy is designed to drastically reduce the necessary cost and expertise required, allowing more researchers and companies to participate.

According to HOPPR CEO Dr. Khan Siddiqui, "foundation models are changing the pace of innovation in imaging AI, but only if they’re accessible, adaptable, and built with real-world deployment in mind." By releasing these foundational tools, HOPPR is sending a clear message to developers and healthcare providers: stop reinventing the wheel and start building site-specific, high-impact solutions tailored to your patients’ needs.

Bottom line: HOPPR latest offering is a 2D mammography foundation model that gives hospitals the building blocks to fine-tune their own high performing, site-specific algorithms.

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.