12th Edition | X-rays in orbit, MedSAM2, and lower cost imaging

Welcome to Radiology Access, your biweekly newsletter on the people, research, and technology transforming global imaging access. If you want to stay up-to-date with the latest in Radiology and AI, then click here to get RadAccess sent directly to your inbox!

Bitcoin launches portable X-ray to the moon! Well, space at least…

Last week the FRAM2 mission, a pioneering polar-orbit human spaceflight, was launched. Fram2 was privately funded and commanded by crypto billionaire Chun Wang and operated by SpaceX with a Crew Dragon spacecraft. In addition to over 20 other scientific studies, FRAM2 also achieved a significant medical imaging milestone by capturing the first human X-ray in space. This endeavor was facilitated through the collaboration of SpaceXray project, KA Imaging, and MinXray, whose portable x-ray technologies enabled this historic accomplishment.

KA Imaging's Reveal 35C X-ray detector, featuring SpectralDR technology, was selected for its ability to produce three distinct images—soft tissue, bones, and a traditional digital radiograph—from a single X-ray exposure. Complementing this, MinXray contributed its IMPACT system, a compact, battery-powered X-ray generator designed for remote environments, making it ideal for microgravity applications. Together, these technologies allowed the FRAM2 crew to perform diagnostic-quality radiographs in space, a first in human spaceflight history.

On April 2nd, Chun shared the first x-ray collected in space on X. The inaugural image captured during the mission featured a hand wearing a ring, mirroring the first-ever X-ray taken by Wilhelm Roentgen in 1895. This symbolic gesture not only paid homage to the X-ray’s origins, but it also demonstrated the feasibility of conducting medical imaging in microgravity, paving the way for monitoring astronaut health during extended space missions.

The successful capture of X-ray images in space has profound implications for expanding access to radiology on Earth—especially in remote, underserved, or adverse environments. The ability to collect high-quality diagnostic images without the need for bulky infrastructure or stable power sources opens the door to life-saving imaging in places that have traditionally lacked access to radiology. The same portable x-ray that can image astronauts can also expand imaging access to rural areas, regions without medical infrastructure, and disaster zones.

MedSAM2 slashes medical image segmentation costs by 85%

Over the weekend, Jun Ma and Bo Wang put out an arXiv article on MedSAM2—the latest foundation model for 3D medical image and video segmentation. This is the same team behind other recent impactful segmentation works, including the original MedSAM and the U-Mamba architecture. Their latest algorithm was developed by fine-tuning the Segment Anything Model 2 on an extensive dataset containing over half a million 3D medical images or videos with segmentation pairs. This broad training data, which included CT, PET, MRI, ultrasound, and endoscopy, enabled them to develop a highly generalizable model while maintaining impressive speed and accuracy.

MedSAM2 achieved state-of-the-art performance on 10+ segmentation benchmarks, outperforming existing methods across modalities and demonstrating strong generalization to unseen pathologies and organs. A key highlight of the work is its integration into a human-in-the-loop annotation system, which allowed the authors to create large, diverse datasets (e.g., 5,000 CT lesions, nearly 4,000 liver MRI lesions, and more than 250K echo video frames) while reducing manual annotation costs by over 85%! The model is promptable, fast, and can be deployed out of the box on platforms like 3D Slice and Google Colab—making it easy for researchers to experiment and adapt for their own purposes.

With its strong performance, generalizability, and ease of use, MedSAM2 could significantly streamline segmentation workflows and accelerate AI development in medical imaging. The algorithm has something for every medical imager. For researchers, it offers a robust new tool for building larger datasets and validating models. For industry, MedSAM could significantly reduce the cost of obtaining large, expert-labeled datasets and enable faster iteration. For the clinical community, it could help reduce annotation bottlenecks and bring high-quality AI tools closer to real-world deployment. Best of all, MedSAM2 is open source and has easy to use Google Colab and 3D Slicer deployments.

India Scheduled to Deploy Homegrown MRI in October

India is on the cusp of deploying the country’s first homegrown MRI, a major milestone in the country's effort to make advanced medical imaging more affordable and accessible. It was recently announced that the first prototype is scheduled to be installed and tested at AIIMS Delhi this October, with clinical trials to follow shortly after. The scanner was developed by SAMEER, which is a government run R&D institute. The project is called Indigenous Magnetic Resonance Imaging (iMRI) and aims to reduce India’s dependence on imported medical equipment. The new 1.5T MRI system is expected to cost around $300K, which is significantly lower than available options.

This project is part of a broader push by the Indian government to advance self-reliant healthcare technologies. SAMEER is also sponsoring the development of an Linear Accelerator (LINAC) for radiation therapy, which could drastically reduce treatment costs for cancer care. If successful, these innovations could not only transform healthcare delivery across India, but also potentially provide affordable devices for other countries facing similar challenges in radiology and oncology access.

EMVision launches clinical trial on portable brain scanner

On a similar note, The Australian company EMVision announced they have initiated a clinical trial for their portable, electromagnetic brain imaging device, called emu. For some context, electromagnetic imaging (EMI) is a portable, non-invasive brain imaging technique that uses low-power radiofrequency waves to detect differences in tissue properties. Unlike MRI, which provides high-resolution anatomical detail using strong magnetic fields, EMI offers faster, point-of-care imaging without the need for heavy infrastructure or shielding. While not as detailed as MRI, EMI has shown promise for rapid stroke detection and other urgent brain assessments, especially for field applications or in resource-limited settings.

EMVision’s is aiming for FDA De Novo clearance for emu and has commenced their clinical trial at The Royal Melbourne Hospital in Australia. They have also shipped a device to the University of Texas Health Science Center in Houston and are planning a site initiation visit. Additional sites in both countries are expected to join the trial. EMVision is aiming to enroll 300 suspected stroke participants across six sites over an estimated 6–12 months, with the primary objective of demonstrating hemorrhage detection sensitivity and specificity exceeding 80%. If successful, this trial could bring an entirely new diagnostic tool to the emergency medicine field.

Resource Highlight

MedSAM Public Imaging Datasets List

If you're on the hunt for segmentation datasets, don't miss the MedSAM team's curated list of public imaging data. This detailed spreadsheet catalogs over 300K images across ~20 modalities and 30 anatomical regions. Each entry includes the dataset name, modality, anatomy, segmentation targets, number of classes, and image count—plus best of all, they provide licensing info alongside links to the datasets and relevant papers. It's a must-bookmark resource for anyone working on medical image segmentation.

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.