18th Edition | 1% is Better Than Nothing!

Campbell Arnold
Jul 1
6 min read

“Sustainable use of MRI in remote communities will only be possible if the MRIs can be easily built and maintained autonomously in the community.”

— Sarty et al., Frontiers in Neuroimaging 2025

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

In this issue, we cover:

MRI at the Edge: From Remote Canada to Rural Ethiopia
A Little Dose Goes a Long Way: 1% Dose Improves PET Synthesis
The Future of MRI Is Sealed: Siemens Launches Helium-Efficient MRI System
Signs of Life in the Radiology AI Market

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

MRI at the Edge: From Remote Canada to Rural Ethiopia

Special issue explores how to bring MRI to the 5 billion people without access.

*Sarty et al., Frontiers in Neuroimaging 2025*

Despite its critical role in diagnosing and managing neurological conditions, MRI remains inaccessible to over two-thirds of the global population—over 5 billion people. High-field MRI systems are costly, power-intensive, and dependent on highly trained personnel—barriers that disproportionately affect lower income countries and remote regions. The latest special issue in Frontiers in Neuroimaging focuses on a promising solution: autonomous low-field MRI systems with simplified operations, improved image quality, and minimal reliance on human expertise.

Highlights from the first five articles include:

Community-built MRI in Remote Canada: A student team used aircraft maintenance skills to build a prototype low-field MRI scanner called Owl MRI in northern Canada, demonstrating how remote communities could one day construct and maintain their own imaging infrastructure.
First Post-Contrast Brain Tumor Imaging with Hyperfine: In a single-case study, researchers showed that post-contrast imaging of brain tumors is feasible using a 64mT low-field MRI—expanding more diagnostic tools to these portable devices.
Native Noise Modeling for Low-Field MRI: A new AI technique improved signal-to-noise ratios by up to 32% in 0.05–0.3T images by leveraging native edge noise and simulated data—a highly generalizable denoising approach that works with minimal training data.
AI Aligns Low-Field and High-Field Brain Volumes: Deep learning models like SynthSR and HiLoResGAN reduced volumetric discrepancies between 64mT and 3T MRI, a key step in standardizing data for clinical and research use.
MRI and EEG Feasibility in Rural Ethiopia: Interviews revealed that community acceptance of neuroimaging was high when paired with educational videos and local engagement, highlighting key social considerations during rural deployments.

Together, these contributions point to a new future for brain imaging—one in which AI-augmented low-field MRI could deliver high-quality, point-of-care diagnostics to areas where traditional MRI has never reached. Submission for this issue are still open if you have cutting edge low-field research.

Bottom line: To bring MRI to the 5 billion people who lack access, we’ll need innovative new ways to build, maintain, and operate devices—these articles highlight early steps toward sustainable, community-driven neuroimaging.

A Little Dose Goes a Long Way

How AI could bring PET imaging to a broader group of patients.

While PET and CT offer tremendous clinical value, they also expose patients to ionizing radiation—fueling long-standing efforts to reduce dose and, more recently, to synthesize PET images using non-ionizing modalities like MRI. In a recent arXiv study, researchers evaluated whether diffusion models could generate high-quality, full-dose PET images from either MRI alone or MRI combined with just 1% of the typical PET dose. The work offers some interesting insights into the role of data and models in medical image synthesis.

Here are some key highlights:

Researchers trained two score-based diffusion models and a Transformer‑UNet using either MRI data alone or MRI + 1% dose PET data from 52 epilepsy patients.
Performance was evaluated using quantitative measures of metabolic asymmetry—an important metric in PET studies.
For MRI-only PET synthesis, diffusion models outperformed the Transformer‑UNet, but also exhibited inconsistencies across slices.
Notably, models trained with MRI plus 1% PET data achieved significantly higher quantitative performance—nearly indistinguishable from full-dose PET and far better than MRI alone.
Interestingly, after adding PET data the performance gap disappeared between diffusion models and Transformer-UNet.

The findings highlight a crucial takeaway: good data will always be more important than your model architecture. Even a small amount of actual PET signal dramatically boosted model performance. In this case, 1% dose proved sufficient to stabilize image quality and preserve key clinical features. Though not the "zero-dose" holy grail, ultra-low-dose PET could significantly reduce radiation exposure, simplify logistics (especially for short half-life tracers), and expand access to advanced metabolic imaging in a broader range of clinical settings.

Bottom line: Even a tiny amount of underlying PET signal—just 1% of the standard dose—can dramatically improve the image quality and consistency of synthetic full-dose PET images.

The Future of MRI Is Sealed

Siemens joins Philips in launching helium-efficient MRI systems.

Siemens Healthineers has secured FDA clearance for its Magnetom Flow.Ace, marking the company’s first helium-efficient 1.5 T MRI scanner, featuring a closed-circuit system that uses only 0.7 L of helium—over 99 % less than conventional systems. The design also eliminates the need for a quench pipe and features an Eco Power Mode to reduce energy usage by roughly 30%.

Philips led the way with an earlier approval of their BlueSeal technology, the first FDA-cleared “helium-free” 1.5 T systems featuring a sealed-magnet design. Similarly, these systems do not require a quench pipe, which makes them lighter weight and allows for more flexible installations, including in mobile units. GE is also developing a reduced helium system called Freelium, though it has not yet received FDA clearance.

These helium-efficient systems represent a major shift in imaging infrastructure—dramatically lowering helium consumption and installation barriers. As helium supply concerns grow, and clinics seek more sustainable, lower-maintenance, and accessible MRI solutions, this marks a pivotal moment in modernizing and future-proofing MRI.

Bottom line: Helium-efficient MRI systems have reached 1.5T, which could lower the barriers to install and maintain high-field systems worldwide.

Signs of Life in the Radiology AI Market

A flurry of deals suggests the sector is heating back up in 2025.

After a quiet spell in health tech funding, the radiology AI sector is showing signs of momentum once again. Several recent deals suggest that investors—and companies—are regaining confidence in AI’s role in the imaging landscape. In just the past few weeks, we’ve seen a string of funding rounds and acquisitions.

Here are a few notable moves:

Hoppr raised $31.5M in a Series A to scale its AI imaging infrastructure platform, aimed at helping developers build and deploy foundation models across imaging workflows.
Rad AI extended its Series C to $68M, adding investment from four major US health systems. The company’s radiology report automation tools are among the most widely deployed algorithms in the field.
ThinkSono secured $6M in new funding to accelerate development of its ultrasound guidance system, with plans to expand their European distribution and seek FDA clearance.
Samsung is rumored to invest $100M in Exo, the handheld ultrasound company that has been rapidly expanding into AI-based imaging tools. The deal would signal serious confidence in the future of point-of-care imaging.
On the M&A front, Function Health acquired Ezra, the direct-to-consumer full-body MRI company, while Gleamer picked up both Pixyl and Caerus Medical earlier this year to strengthen its position across modalities and anatomies.

Taken together, these moves signal renewed vigor in the radiology AI sector. With large health systems backing enterprise deployment, companies landing solid eight figure Series A rounds, and large players like Samsung stepping into the space, 2025 is shaping up to be a pivotal year for the industry.

Bottom line: After a quiet stretch, investment in radiology AI may be picking up pace, signaling growing confidence in the field’s clinical and commercial value.

Resource Highlight: The Imaging Wire

For this week's resource highlight, I’m recommending you subscribe to The Imaging Wire. It’s one of my favorite ways to stay updated on industry trends. This publication posts twice a week, offering an in-depth exploration of a single topic while also providing a brief overview of other notable developments.

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.