- Campbell Arnold
- Oct 7
- 4 min read
"This is the first-ever designation for AI with such broad coverage of medical conditions under one solution.”
— Aidoc Press Release on FDA Breakthrough Designation
Welcome to Radiology Access! your biweekly newsletter on the people, research, and technology transforming global imaging access.
In this issue, we cover:
SuperSynth: The Everything Model for Brain MRI
Aidoc Receives Breakthrough Designation for Multi-Indication Foundation Model
Hyperfine Launches PULSE, A Subscription Based Platform Aimed at Researchers
If you want to stay up-to-date with the latest in Radiology and AI, then don't forget to subscribe!
SuperSynth: The Everything Model for Brain MRI
Could a single algorithm replace multi-step neuroimaging pipelines?
The MGH team behind SynthSeg and SynthSR just unveiled their latest innovation: SuperSynth, a multi-task 3D U-Net built to handle virtually any MRI input. Like its predecessors, SuperSynth is trained on synthetic data, enabling it to generalize across a wide-range of contrasts, resolutions, and scanners, without retraining. This flexibility allows it to process everything from low-field to ex vivo scans.
SuperSynth performs an entire neuroimaging post-processing pipeline from a single input. The algorithm delivers super-resolution (to 1 mm isotropic), synthesizes T1, T2, and FLAIR contrasts, registers data to the MNI atlas, and performs segmentation of brain regions and white matter hyperintensities, along with cortical ribbon estimation. Essentially, an all-in-one tool for what are typically mutli-step neuroimaging pipelines.
While no dedicated publication has been released yet, SuperSynth is already available through FreeSurfer and is conceptually related to the BrainFM framework recently posted on arXiv. Though the BrainFM work also tackles other tasks, including CT synthesis and bias-field estimation.
By unifying super-resolution, synthesis, and segmentation in a single framework, SuperSynth has the potential to streamline post-processing across diverse scanners and protocols. Its ability to generate standardized, high-quality outputs from challenging data makes it especially promising for low-field MRI and global imaging research. SuperSynth is a powerful example of how synthetic data can produce robust, generalizable foundation models for medical imaging.
Bottom line: SuperSynth pushes toward a future where one adaptable model can process any MRI, regardless of contrast, resolution, or field strength.
Aidoc Receives Breakthrough Designation for Multi-Indication Foundation Model
Could this foreshadow a new FDA regulatory pathway for foundation models?
Aidoc has received FDA Breakthrough Device Designation for its new multi-triage CT foundation model. The system spans more than ten indications within a single submission. This marks a major regulatory milestone. While previous foundation models have been cleared for one indication at a time, Aidoc is aiming to bundle broad diagnostic coverage under a single unified pathway. The breakthrough designation accelerates review for technologies that address unmet clinical needs and signals that the FDA may be opening up to foundation models capable of multiple clinical applications.
The model under review is built on Aidoc’s CARE foundation model, designed to identify and prioritize high-risk findings across a wide range of pathologies. According to the company, this multi-condition approach could allow radiology teams to streamline workflows and eliminate the need for multiple, siloed AI tools. Aidoc CEO Elad Walach noted that this represents the first FDA designation for an AI solution covering double-digit conditions under one umbrella.
If successful, the implications could be far-reaching. Instead of pursuing separate submissions for each pathology, developers could soon target multi-pathology, generalist models through a single FDA process, significantly reducing overhead. For hospitals and health systems, this could translate to simpler integration and a more unified AI ecosystem, with less fragmentation across multiple AI devices and vendors.
Bottom line: Aidoc’s breakthrough designation could mark a turning point for foundation models with multiple indications rather than siloed, single-purpose tools.
Hyperfine Launches PULSE, A Subscription Based Platform Aimed at Researchers
Has Hyperfine found a way to monetize its growing research community?
Hyperfine, the maker of the world’s first portable MRI system, just unveiled PULSE (Portable Ultra-Low-Field Scientific Exchange), a new subscription-based research platform. While the company promotes Swoop as an easy-to-use device, with scan protocols selected from preset playlists, researchers have long sought greater control and customization. PULSE changes that by giving users deeper access to the system’s capabilities, allowing them to:
Use Hyperfine’s proprietary research sequences to explore new imaging applications.
Import, modify, and share community-developed sequences.
Access raw imaging data for algorithm development and analysis.
Thus far, low-field MRI has been more strongly embraced by the MR research community than by hospital systems, creating a large global network of scanners used primarily for research. With PULSE, Hyperfine may have found a way to generate additional revenue from these research devices, while simultaneously providing a platform for collaboration and innovation among users. The company also hinted that it plans to open the platform to third-party developers, following a path similar to Butterfly Network’s app ecosystem for portable ultrasound.
Bottom line: With PULSE, Hyperfine aims to transform its research network into both a revenue stream and an innovation engine.
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.
