- Campbell Arnold
- 4 days ago
- 3 min read
“It not just about image quality—it’s about preserving clinically relevant features that doctors rely on.”
— Long Wang, Subtle Medical ML Engineer
Welcome to RadAccess: Impressions—your quick-read companion to the main RadAccess newsletter. Like a radiology report's impressions section, we only deliver the essential information to respect your time. For more details, you can always turn to the full RadAccess newsletter.
In this issue, we cover:
Vendors showcase MRI of the future at ISMRM
Super-Resolution Boosts Low-Field MRI Stroke Sensitivity
ViT-Fuser: Leveraging Prior Scans to Accelerate Low-Field MRI
The New DEAL for Large Language Models
Want to stay up-to-date with the latest in Radiology and AI? Don't forget to subscribe!
Vendors showcase MRI of the future at ISMRM
At ISMRM 2025 in Honolulu, leading MRI vendors—including GE, Siemens, Philips, Canon, Hyperfine, and United Imaging—presented their roadmaps for the future of MRI. Upcoming innovations centered on AI-integration, streamlined operations, and more advanced hardware. Here are a few top trends coming in the next generation of scanners:
Deep learning reconstruction is the new standard-of-care with each OEM showcasing AI-based reconstruction for faster, sharper imaging.
Helium-free and portable systems to reduce overall siting costs and expand MRI access to new care settings.
An autonomous scanning process with innovations in auto-positioning and workflow automation.
Higher-performance hardware including powerful gradients that enable more specialized imaging.
More variety in field strengths both at the lower and higher end of the spectrum to meet varied user needs.
ViT-Fuser: Leveraging Prior Scans to Accelerate Low-Field MRI
A new arXiv article proposes a breakthrough in low-field MRI by using patients’ prior high-field scans to significantly improve image quality and reduce scan times. The technique, called ViT-Fuser, employs a vision transformer model to fuse feature-level information from previous high-field scans with current low-field acquisitions—regardless of vendor, sequence type, or field strength. Validated on both simulated and real-world low-field datasets, ViT-Fuser demonstrated improved signal-to-noise ratio, enhanced tissue contrast, and up to 4× scan acceleration, outperforming other state-of-the-art reconstruction models. This innovation could make low-field MRI a more viable option for longitudinal monitoring, especially in conditions which require frequent repeat imaging, such as multiple sclerosis.
Super-Resolution Boosts Low-Field MRI Stroke Sensitivity
A new study in Stroke showcases how deep learning can elevate the diagnostic potential of low-field scanners. Researchers developed SCUNet, a hybrid convolution-transformer model, to enhance spatial resolution and reduce noise in 0.23T MRI scans, significantly improving their performance on ischemic stroke detection. The algorithm was trained on open-source data and fine-tuned on paired scans from 282 stroke patients. SCUNet achieved notable gains in sensitivity (89% vs. 77%) and specificity (91% vs. 71%) compared to native low-field images. It also closely matched 3T MRI in key quantitative metrics, like lesion volume and ADC values. By preserving critical diagnostic features and narrowing the performance gap, SCUNet could help make high-quality stroke imaging more accessible in underserved settings.
The New DEAL for Large Language Models
As large language models (LLMs) become increasingly integrated into radiology workflows there is a growing need for more stringent scientific oversight. A new technical report in NEJM AI introduces the DEAL checklist to help improve transparency, consistency, and clinical relevance in LLM research. With many LLM applications currently operating outside FDA oversight, the DEAL framework aims to establish best practices for reporting model development and usage, helping to prepare for eventual regulatory pathways. It includes two tailored tracks: DEAL-A for advanced development work and DEAL-B for applied use of pretrained models. By standardizing documentation and encouraging reproducibility, DEAL offers a critical tool for researchers and developers aiming to transition LLMs from promising prototypes to clinically validated tools.
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.
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
Oved, Tal, et al. "Deep learning of personalized priors from past MRI scans enables fast, quality-enhanced point-of-care MRI with low-cost systems." arXiv preprint arXiv:2505.02470 (2025).
Bian, Yueyan, et al. "Quantitative Ischemic Lesions of Portable Low–Field Strength MRI Using Deep Learning–Based Super-Resolution." Stroke (2024).
Tripathi, Satvik, et al. "Development, Evaluation, and Assessment of Large Language Models (DEAL) Checklist: A Technical Report." NEJM AI (2025): AIp2401106.
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.
