DEEPVESSEL Plaque FDA clearance: AI plaque scoring for CCTA
Keya Medical Technology received DEEPVESSEL Plaque FDA clearance on June 4, 2026, a 510(k) decision under number K260497 for AI software that measures coronary plaque and stenosis from CCTA scans. In a multicenter US study, the company reports a correlation of r = 0.976 with expert readers for total plaque volume. For imaging programs, it adds plaque quantification to scans they already take, with trained analysts checking the AI.
What problem coronary CT plaque analysis solves
A coronary CT angiography scan, or CCTA, shows more than whether an artery is narrowed. It also shows the plaque building up in the vessel wall, but turning that image into numbers a clinician can use takes careful measurement. Keya Medical Technology built DEEPVESSEL Plaque to help qualified clinicians assess the presence and extent of coronary atherosclerosis from CCTA through quantitative plaque measurements, stenosis categories and visual reports.
What the FDA cleared on June 4, 2026
The FDA cleared DEEPVESSEL Plaque under 510(k) number K260497, with a decision of substantially equivalent. The agency received the submission on February 13, 2026 and issued its decision on June 4, 2026.
The device is classified as automated radiological image processing software, under regulation number 892.2050 and product code QIH, and was reviewed by the radiology panel. The applicant is Keya Medical Technology Co., Ltd., based in Beijing, with the law firm Hogan Lovells in Philadelphia as correspondent. The record shows a traditional summary, no third-party review and no predetermined change control plan.
What the software measures
For each scan, DEEPVESSEL Plaque reports a set of measurements:
- total plaque volume
- non-calcified plaque volume
- calcified plaque volume
- low-density plaque volume
- plaque burden
- the CAD-RADS stenosis category
The results arrive in two reports. The CardiSight Overview Report gives a visual overview of coronary anatomy, plaque location and the CAD-RADS category. The Plaque Analysis Report gives vessel- and patient-level plaque volumes, burden, composition and the stenosis assessment.
How the AI-assisted analysis works day to day
The workflow has three steps, and a person checks the AI along the way. First, previously acquired CCTA images in DICOM format are sent through a secure remote workflow. Second, deep-learning segmentation of the arteries is reviewed and, when needed, edited by trained analysts. Third, the quantitative and visual results go back to the qualified clinician for review.
DEEPVESSEL Plaque is separate from the company's DEEPVESSEL FFR, which has its own FDA clearance, workflow and reports. The company says the two together can support a more comprehensive CCTA assessment.
How accurate the DEEPVESSEL Plaque results were in testing
The company's evidence comes from COMPOSE-CT, a multicenter US study that compared DEEPVESSEL Plaque with the consensus of three Level III expert readers. Keya reports a Pearson correlation of r = 0.976 for total plaque volume, r = 0.994 for calcified plaque volume and r = 0.945 for non-calcified plaque volume. For stenosis categories, it reports a CAD-RADS weighted kappa of 0.810.
These are the company's published figures. The FDA record lists the decision and classification, not the study results.
What the DEEPVESSEL Plaque FDA clearance changes for imaging programs
For a hospital or imaging center that already runs CCTA, the clearance means plaque quantification can be added to scans it has already acquired, with the segmentation done by software and reviewed by trained analysts. The task the product handles is the measurement and reporting; the clinical decision stays with the clinician.
Keya has not published prices, turnaround times or cost savings for the service. A buyer weighing it would compare the analyst-reviewed reports against how its own team measures plaque today.
Questions and answers
What is DEEPVESSEL Plaque?
It is AI-assisted software from Keya Medical Technology that measures coronary plaque volumes, plaque burden and the CAD-RADS stenosis category from CCTA scans, with trained analysts reviewing the segmentation.
When did DEEPVESSEL Plaque get FDA clearance?
The FDA issued its 510(k) decision on June 4, 2026, under number K260497, after receiving the submission on February 13, 2026.
How was DEEPVESSEL Plaque validated?
In the multicenter US COMPOSE-CT study, the company reports a correlation of r = 0.976 with three Level III expert readers for total plaque volume.
MOZGII view
How we would build it
Medicine benefits most from AI and robotics where the work is repetitive, precise and slow for a person: tracing vessels on a scan, counting cells, holding an instrument steady. Coronary plaque measurement fits that pattern. A trained reader can do it, but it takes time per scan, and the result depends on who does it.
If we were building this kind of service, we would treat the analyst review as part of the product, not a stopgap. The model segments the arteries, an analyst corrects the edges, and every correction goes back into the training set with the scanner type and image quality attached. The clinician gets numbers and a picture, and the report says which parts a person edited.
Where it breaks is input quality. Older scanners, heavy calcium and motion blur all make segmentation harder, and the error shows up as a confident wrong volume. We would flag low-quality scans before analysis instead of after. The one metric we would watch is the share of cases where analysts make major edits, broken down by site and scanner. Next we would build follow-up comparison, so the same patient's plaque can be measured the same way on scans taken a year apart.
Sources
Where the facts in this story come from.