A wave of FDA clearances and European milestones is turning an imaging-center operator into a software company, but clearance is a starting line, not a finish line
On June 25, RadNet‘s wholly owned subsidiary DeepHealth announced two new FDA 510(k) clearances for its breast-imaging software: a tool that flags breast arterial calcification on routine mammograms as a possible early marker of cardiovascular disease, and a feature that lets its diagnostic mammography product, now marketed as Mammo Dx, automatically pull in a patient’s prior exams to track lesions over time.
The clearances landed in a year RadNet had forecast would bring at least four approvals, and they cap a remarkable run: prostate and neurological AI cleared in 2025, remote-scanning software earlier in 2026, and a string of acquisitions that have widened the portfolio. For a company long known as the largest operator of freestanding imaging centers in the United States, the message is increasingly that RadNet wants to be judged as a software business too.
DeepHealth’s place in RadNet’s strategy
DeepHealth is the umbrella brand for everything in RadNet’s Digital Health segment, and that segment has become the company’s growth story. In the first quarter of 2026, Digital Health revenue rose more than 50% year over year to roughly $29 million, with annual recurring revenue nearly doubling to about $97 million and a stated target above $140 million by year-end.
Those are small numbers against RadNet’s total revenue of more than $575 million for the quarter, but they grow far faster than the core imaging business and carry the higher margins and recurring economics that investors prize in software. RadNet has assembled the portfolio aggressively, folding in companies such as Quantib, iCAD, See-Mode, CIMAR and, in early 2026, the French radiology-AI firm Gleamer, whose tools are already used across dozens of countries.
The strategic logic is twofold. RadNet can deploy these tools inside its own roughly 435 centers to speed workflows and ease staffing pressure, then sell the same software to outside hospitals and imaging providers, a base the company puts at more than 2,700 Digital Health customers. A new joint venture with Trinity Health’s Saint Alphonsus system, where multiple DeepHealth products will run across imaging sites, is being pitched as a blueprint for similar health-system partnerships.
Why AI imaging matters now
AI has become central to diagnostic imaging for reasons that are as much operational as clinical. Imaging volumes keep rising as populations age and screening expands, while the supply of radiologists has not kept pace, leaving many departments stretched. Software that can triage worklists, draft portions of reports, compare a scan against earlier studies, and flag suspicious findings promises to move cases through faster and let scarce specialists focus where they are most needed.
In practical terms, DeepHealth’s tools span cancer detection in mammography, prostate MRI workflows that automatically locate lesions and assist with standardized reporting, brain analysis aimed at neurodegenerative disease, and prioritization software that pushes urgent cases up the queue. The appeal for oncology networks is direct: earlier, more consistent detection and smoother handoffs across a fragmented care journey.
What a clearance actually means
It is here that careful language matters most. An FDA 510(k) clearance allows a device to be marketed for a defined intended use, generally by showing it is substantially equivalent to a product already on the market. A European CE mark, which several DeepHealth tools also carry, performs a similar gatekeeping function for that market. Neither is the same as proof that a tool improves patient outcomes, lowers the total cost of care, or can substitute for a radiologist.
Clearance speaks to safety and intended performance for a narrow task; it does not establish real-world clinical benefit, and it does not settle how a tool behaves once it meets the messy variety of scanners, populations, and workflows outside the submission data. Vendor figures such as the 90% sensitivity and 88% specificity reported for the calcification tool describe controlled testing, not a promise of everyday results. The honest framing is that these products are decision aids meant to support clinicians, not replace their judgment or their liability.
The competitive field
RadNet is not alone in chasing this opportunity, and its rivals are formidable. Large imaging manufacturers such as GE HealthCare, Siemens Healthineers and Philips embed AI directly into scanners and enterprise software and reach hospitals worldwide; indeed, RadNet’s platform push invites comparison with these players rather than only with other outpatient operators. A crowded field of specialist AI developers competes on individual applications, hospitals build or buy their own tools, and the pace of clearances across the industry means differentiation can erode quickly.
RadNet’s distinctive wager is integration: a single platform, marketed as DeepHealth OS, spanning many modalities and clinical areas, sold to a provider’s own centers and to outside customers alike. Whether breadth wins out over best-in-class point solutions is an open commercial question.
Risks and unanswered questions
The hurdles are substantial and mostly lie beyond the regulatory filing. Integrating AI into existing hospital IT and radiology systems is notoriously difficult, and reimbursement remains uncertain, since payers do not consistently pay extra for AI-assisted reads, leaving the business case resting on efficiency gains rather than new revenue. Model performance depends on data quality and can degrade across different equipment and patient groups, which is why ongoing clinical validation and post-market monitoring are essential rather than optional.
Clinician adoption is not guaranteed; tools that add clicks or generate false alarms can be ignored or quietly switched off. And the move toward data-driven imaging sharpens concerns about patient privacy, cybersecurity, and liability when an algorithm contributes to a missed or mistaken finding. RadNet itself, despite roughly $2 billion in 2025 revenue, posted a net loss for the year, a reminder that scaling this strategy is costly. The clearances are real and the momentum genuine, but the evidence that AI imaging improves outcomes and economics at scale is still being built, and that, more than any regulatory milestone, will determine how the bet pays off.
Read more biotech insights on OncoDaily Biotech and OncoDaily Radiotherapy.
Written by: Semiramida Nina Markosyan, Editor, OncoDaily Canada