Impressions | 23rd Edition

Campbell Arnold
Sep 15
3 min read

“[This] is really about giving MR scanners a memory. One day, the more we see you, the faster we will be able to scan you.”

— Dan Sodickson, Chief of Innovation, NYU Radiology Department

Welcome to RadAccess: Impressions, your quick-read companion to the main RadAccess newsletter. Here we only deliver the essential information. For more details, you can always turn to the full RadAccess newsletter.

In this issue, we cover:

OmniMRI: One Model to Rule Them All?
Trust, but Verify: Smarter MR Reconstruction
AI-Guided, On-the-Fly Protocol Adaptation

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

OmniMRI: One Model to Rule Them All?

A new arXiv paper introduces OmniMRI, an ambitious vision-language foundation model trained on 220,000 MRI volumes to tackle a broad range of radiology tasks, including reconstruction, segmentation, pathology detection, diagnostic suggestion, and report generation. The authors detail a flexible multi-stage training paradigm capable of integrating any public dataset through an Instruction-Response framework. While early qualitative results look promising, the work is still preliminary:

The code and benchmarks have yet to be released.
Text outputs show some inconsistencies with associated imaging.
Generalization beyond trained tasks is unproven.

Still, OmniMRI may represent a bold step toward a general-purpose radiology foundation model, one worth following with cautious optimism.

Trust, but Verify: Smarter MR Reconstruction

An IEEE-TMI study introduces the Trust-Guided Variational Network (TGVN), a deep learning MR reconstruction framework that can accelerate scan times by leveraging side information, such as prior scans, complementary contrasts, or even other modalities. Unlike typical generative approaches that risk hallucinating structures, TGVN defaults back to measured k-space data when inputs are degraded or misaligned, helping to ensure robust and accurate reconstructions. The authors demonstrated their algorithms for knee and brain MRI with up to 20× undersampling. TGVN achieved higher quantitative metrics , better preservation of anatomy, and even improved segmentation. By enabling context-aware imaging that safely leverages prior exams, TGVN could both shorten scan times and expand access through enhanced low-field reconstructions.

AI-Guided, On-the-Fly Protocol Adaptation

A recent article in European Journal of Radiology evaluated the ability of Cerebriu’s Apollo to adapt brain MRI protocols in real time by identifying potential infarcts, hemorrhages, or tumors while the patient is still in the scanner and recommending additional sequences. The approach was tested on 414 exams, Apollo achieved 94% sensitivity for infarcts, 82% for hemorrhages, and 74% for tumors, though at the cost of more false positives. While it didn’t significantly boost overall radiologist performance, the system did highlight missed lesions and outperformed technologists for tumor detection. The study shows both the promise and challenges of AI-guided, adaptive workflows. Scanners may be getting smarter and dynamically guided by algorithms, but radiologist oversight is still crucial for safe deployment.

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.