An AI tool for chest X-rays built on a foundation model has received clearance from the US Food and Drug Administration. DeepHealth, a subsidiary of the imaging company RadNet, announced the 510(k) clearance, the most common FDA route for medical devices, on September 16.
Most cleared radiology AI tools are trained for one narrow task. A foundation model is trained broadly first, here on large numbers of chest X-rays, and specific tasks are built on top of it. DeepHealth says that lets it add new findings faster without starting over each time.
Four findings for now
The cleared tool, called Chest XRay, marks and locates four kinds of findings on a chest X-ray: nodules, consolidation (areas of lung filled with fluid or other material), abnormalities in the middle of the chest and problems in the space around the lungs, such as a collapsed lung.
The software comes from Gleamer, a French company, and was earlier sold as ChestView. A version with the European CE mark is already in use. The company says that version reads more than 2.8 million exams a year, based on its own 2026 deployment data.
What the company has not shown
The announcement includes no accuracy figures. Nothing in it shows that a foundation model makes the tool better than narrower systems. The claim so far is about speed of development.
"Foundation models are important building blocks," said Sham Sokka, DeepHealth's chief operating and technology officer. The FDA clearance covers the four findings, not the platform.
