On July 30, 2026, DeepHealth, a Massachusetts-based imaging informatics subsidiary of RadNet, announced it had received FDA 510(k) clearance for DeepHealth Breast Ultrasound, an artificial intelligence system designed to detect, characterize and report on lesions found during breast ultrasound exams. The clearance moves the tool from validation study into commercial deployment across a network that performs hundreds of thousands of these scans every year, and it marks one of the more concrete signs yet that AI is being trusted not just to flag suspicious images but to help write the clinical report a radiologist signs off on.
What the Software Actually Does
Breast ultrasound is typically used as a follow-up to mammography, especially in women with dense breast tissue where masses can be harder to see on an X-ray. DeepHealth’s software sits inside that workflow, automatically detecting lesions during the live scan, characterizing features like shape and margin that radiologists use to judge whether a mass looks benign or suspicious, and then drafting structured report language a sonographer and radiologist can review rather than write from scratch. The goal, according to the company, is to standardize how findings are documented and to compress the time between the scan and a discrete, actionable report.
The Numbers Behind the Clearance
DeepHealth’s FDA submission was backed by a multi-reader, multi-case clinical validation study involving 16 U.S. board-certified radiologists. The company reported more than 98 percent lesion localization accuracy, an 8 percent increase in cancer detection sensitivity when radiologists used the AI as an aid, and a 37 percent reduction in interpretation time per case. RadNet estimates roughly 700,000 breast ultrasound studies are performed annually across its outpatient imaging network, giving a sense of the scale at which the tool could eventually operate if adopted system-wide.
Reimbursement and Rollout
The software is now commercially available for sale to U.S. customers, and DeepHealth says providers can pursue reimbursement through an existing Category III CPT code covering quantitative ultrasound tissue characterization, a category created for emerging technologies still building a track record of clinical evidence. RadNet has said it intends to deploy the platform across its national outpatient imaging network by the end of 2026, which would make it one of the largest real-world rollouts of an AI ultrasound reporting tool in the country to date.
Where This Fits Among Recent AI Imaging Clearances
The clearance lands amid a wave of FDA authorizations for AI-assisted imaging tools in 2026, part of a broader regulatory trend in which the agency has increasingly greenlit software that assists, rather than replaces, a radiologist’s judgment. Breast imaging has been a particularly active category because dense-tissue callbacks and false positives are a well-documented source of patient anxiety and unnecessary biopsies, and vendors have pitched AI reporting tools as a way to reduce both variability between readers and the administrative burden of documentation.
Two Ways to Read the Rollout
Radiology groups adopting the tool are likely to frame it as addressing a workforce bottleneck: sonographers and radiologists face heavy caseloads, and a system that pre-populates a defensible structured report while flagging lesions with high accuracy can free up time for more complex reads. Skeptics of AI-generated clinical documentation counter that a report drafted by software, even one reviewed and signed by a radiologist, can create liability and workflow questions if reviewers grow accustomed to accepting AI-suggested language rather than independently characterizing every finding. The 37 percent time reduction cited in DeepHealth’s validation study is exactly the kind of efficiency gain that raises that tension: faster reads are valuable only if oversight doesn’t erode as speed increases.
What Comes Next
With commercial availability already in place, the next phase for DeepHealth Breast Ultrasound is real-world uptake data: how radiology practices outside the validation study perform with it, whether the Category III CPT reimbursement pathway proves durable, and whether cancer detection gains persist once the tool is used across a broader and more varied patient population than the study cohort. RadNet’s planned network-wide deployment by the end of 2026 will be one of the first large-scale tests of whether an AI reporting assistant can hold up its promised accuracy and time savings at scale.
