DeepHealth Breast Ultrasound FDA Clearance
A RadNet subsidiary cleared the FDA with AI handling detection, characterization, and report generation in a single breast ultrasound workflow. 98% localization accuracy. 37% less radiologist time. This is a deployment signal.
382,640 women will be diagnosed with breast cancer in 2026. 42,140 will die from it. And for decades, the core problem has been the same: breast ultrasound is one of the most operator-dependent procedures in all of radiology, with significant variability in how images are acquired, how lesions are characterized, and how reports are written. Different sonographers. Different reads. Different outcomes for the same patient.
On July 30, 2026, DeepHealth - a wholly owned subsidiary of RadNet (NASDAQ: RDNT) - received FDA 510(k) clearance for an AI-powered breast ultrasound platform that handles lesion detection, characterization, and report generation in a single automated workflow. The clinical validation numbers are not incremental. They are structural.
This is not a flagging tool that highlights areas of concern and hands the rest back to a radiologist. This is end-to-end AI automation of an entire imaging modality. And it cleared the FDA. It has reimbursement codes. It is going live across a national radiology network before the end of this year.
1. The Problem This Solves - And Why It Has Been So Hard
Breast ultrasound is not like chest CT. There is no fixed anatomy, no standardized field of view, and no universal acquisition protocol that produces consistent results across operators. When a sonographer acquires images in a busy community clinic, the sweep angle, probe pressure, gain settings, and focal zone adjustments all vary. Significantly.
The result: two patients with identical pathology can receive radically different workups depending on who held the probe and who read the resulting images.
This variability is especially dangerous in dense breast tissue. Approximately 43% of women aged 40-74 in the United States have dense breast tissue. Dense tissue is radiologically similar to tumor tissue on mammography, which is why mammography sensitivity drops sharply as density increases. The numbers from a 2026 comparative study are stark: mammography sensitivity fell from 85.7% in the least dense group to 61.0% in the most dense group. Ultrasound, when performed well, offers a significant supplement - 85.3% sensitivity, 88.4% specificity.
But "when performed well" is exactly the problem. Ultrasound is only as good as the operator and reader behind it. That gap is what DeepHealth built to close.
📊 Mammography sensitivity in highest-density (BI-RADS D) breasts: 61.0%. Ultrasound sensitivity in the same group: 85.3%. The clinical gap is real. The AI addresses it at the workflow level.
2. What the Clearance Actually Says - The Clinical Validation Numbers
The FDA clearance for DeepHealth Breast Ultrasound was supported by a multi-reader multi-case clinical validation study involving 16 U.S. board-certified radiologists. Three headline numbers define what the platform does.
First: greater than 98% accuracy in localizing breast lesions. This is not a theoretical benchmark on a curated dataset. It is a prospective multi-reader validation under conditions that reflect real clinical practice. For a system that automates detection - the step that has historically required a trained radiologist's eye - this figure is the foundation everything else rests on.
Second: 8% improved sensitivity for breast cancer detection. In cancer detection terms, 8% is substantial. If 1,000 women with suspicious ultrasound findings go through a system that detects 8 more cancers per thousand than the baseline, those are 8 patients who receive early-stage diagnoses instead of late-stage ones. The Stage I breast cancer five-year survival rate is approximately 99%. Stage IV: 31%.
Third: 37% reduction in radiologist interpretation time. At scale, this is not efficiency - it is capacity. If a radiologist interpreting breast ultrasound reads 30 cases per session and AI cuts that interpretation burden by 37%, you have effectively created space for 11 additional cases in the same session. Multiply that across a national network and you have a material change in throughput without adding headcount.
3. Why Operator Dependence Is the Real Problem - And Why AI Is the Answer
Dr. Jason McKellop, Medical Director of Women's Imaging for RadNet California, said it directly in HIT Consultant's coverage of the clearance: significant variability in image acquisition, interpretation, and reporting is the core problem the system addresses.
That framing matters. The FDA clearance is not a story about AI being faster. It is a story about AI being more consistent. Consistency at scale is worth more than speed in a vacuum, because variability in radiology is where errors hide.
Consider what variability actually costs. A sonographer at an FQHC serving a largely uninsured population may have less specialized breast imaging training than one at an academic medical center. The AI does not care. It applies the same detection, characterization, and reporting logic regardless of who acquired the images or where the patient is being seen.
This matters enormously for health equity. Dense breast tissue is more prevalent in younger women and in Asian and Hispanic women. These are not populations who are disproportionately represented at specialized breast imaging centers. They are far more likely to show up at community health centers, safety-net clinics, and radiology networks serving Medicaid and uninsured populations.
📊 Approximately 40% of all women will undergo breast ultrasound at some point in their lifetime. RadNet projects 700,000+ annual breast ultrasound studies are eligible for AI-assisted reimbursement under the existing Category III CPT code.
4. The Reimbursement Piece Most Analysts Are Missing
Every AI company that gets a new FDA clearance faces the same question from health system CFOs: does it bill? The answer for DeepHealth Breast Ultrasound is yes - immediately.
Category III CPT codes 0689T and 0690T for quantitative ultrasound tissue characterization are already active. These are the same codes used by Koios' DS Breast Ultrasound AI software, which has demonstrated what clinical deployment of these codes looks like financially. At one Koios customer facility, these codes produced a greater than 30% increase in professional fees and a greater than 10% increase in technical fees. That is real revenue from AI-generated reimbursement.
Category III codes are temporary codes for emerging technologies. The established path is that a functioning Category III code becomes a permanent Category I code within approximately five years. CPT 0721T - the lung AI code at the center of Oatmeal Health's business - followed exactly this trajectory. DeepHealth Breast Ultrasound enters a reimbursement environment that is already primed, not one it has to build.
At RadNet's scale - 700,000+ eligible breast ultrasound studies annually - the revenue potential is significant. RadNet's Digital Health segment already reported Q1 2026 revenue of $29 million, more than 50% growth year over year, with annual recurring revenue approaching $97 million against a target of more than $140 million by year-end. The breast ultrasound clearance is the next piece of that ARR engine.
5. The Competitive Context - Where This Fits in the Radiology AI Market
DeepHealth's breast ultrasound clearance arrives into a radiology AI market that has been bifurcating clearly. Companies building point solutions - a single detection flag, a single finding classification - are struggling to generate durable revenue. Companies building end-to-end workflow platforms with reimbursement are pulling ahead.
The automated breast ultrasound systems market is valued at $1.03 billion in 2026 and is projected to reach $2.17 billion by 2033 at an 11.1% CAGR. AI in breast imaging specifically is growing from $320 million in 2025 to $441 million by 2031. These are the numbers before the practical deployment of multi-function AI platforms like DeepHealth's.
The competitive moat here is not just the technology. It is the RadNet network. RadNet operates more than 400 outpatient radiology centers across the United States. When DeepHealth deploys breast ultrasound AI across that network, it has instant real-world evidence generation at scale, instant distribution, and instant revenue conversion. No health system partnership negotiation. No pilot site procurement process. The distribution channel is already owned.
GE HealthCare has expanded its mammography collaboration with DeepHealth, extending global access to DeepHealth's AI-powered breast cancer screening solutions. The platform's reach is not limited to the domestic RadNet network.
6. The Five Questions Every Radiology Leader Must Ask Before Deployment
The clinical validation numbers for DeepHealth Breast Ultrasound are strong. That does not mean procurement decisions should skip due diligence. End-to-end AI workflow automation introduces a different set of risk considerations than traditional AI flagging tools.
Here are the five questions any radiology department head, CMO, or imaging AI procurement team should demand answers to before signing a deployment agreement.
First: what were the characteristics of the study population? Multi-reader multi-case studies are the gold standard, but they can still be run in environments optimized for success. Sixteen board-certified radiologists is a robust panel. The question is whether the patient cases reflected the full range of breast density, lesion size, and clinical presentation that a community practice would encounter.
Second: what is the false negative rate, not just the localization accuracy? Greater than 98% localization accuracy is impressive. But the clinical cost of the remaining missed lesions depends heavily on what those lesions were - small, early cancers or obviously benign findings. The FDA clearance process evaluates the data that was submitted. Operators should ask for the full receiver operating characteristic curve, not just the headline sensitivity number.
Third: how does the system handle edge cases? Dense tissue, small lesions at less than 5mm, atypical presentations in post-surgical breasts, and implant interference are the places where AI systems fail in ways that aggregate statistics do not capture. Any deployment should include a documented failure mode analysis.
Fourth: does the CPT reimbursement pathway work for your payer mix? Category III codes 0689T and 0690T are billable. Whether your specific contracted payers will pay on those codes is a separate question. Payer coverage for Category III codes is uneven. A commercial strategy that depends on reimbursement before payer contracting is confirmed is a cash flow risk.
Fifth: where does liability sit when the AI generates the report and the radiologist signs off? AI-generated radiology reports are an area of active regulatory and legal development. The radiologist who signs the report carries primary malpractice liability under current AJR guidance, even when the AI drafted it. That liability framework needs to be explicit before clinical deployment.
Deep Dive - The End-to-End AI Model Is Now Real
The conventional wisdom in radiology AI for the past decade has been that AI assists but does not replace. Flag the suspicious lesion. Surface the finding. Let the radiologist make the call. This model produced a lot of useful tools and almost no durable revenue, because assistance without workflow integration does not generate billable events.
DeepHealth Breast Ultrasound breaks that model explicitly. The platform does not flag and defer. It detects, characterizes, and generates a report. The radiologist reviews and signs. The workflow is fundamentally different: AI produces the primary document; the physician applies clinical judgment to validate it.
This is not a radiology AI question anymore. This is a practice management question. If AI generates reports that radiologists verify, then:
Radiologist throughput scales with AI capacity, not headcount
Revenue per radiologist hour increases because interpretation is pre-structured
Sonographer documentation burden drops because measurement extraction is automated
Report standardization across sites becomes achievable without manual protocol enforcement
The math on RadNet's 700,000 eligible studies is instructive. At a conservative Category III code reimbursement of $50 per study - and Koios data suggests the real number is higher - that is $35 million in potential annual AI-generated revenue from a single modality at one network. Before any expansion, before any additional clearances, and before any Category I code upgrade that would increase reimbursement further.
What the GE HealthCare Partnership Signals
GE HealthCare does not expand collaborations without a commercial thesis. The expansion of the mammography partnership to extend global access to DeepHealth's AI-powered breast cancer screening solutions is a signal that GE HealthCare is positioning the DeepHealth platform as infrastructure for its global installed base.
GE HealthCare has the largest global installed base of mammography equipment. Embedding DeepHealth AI into that installed base means the platform can reach markets where independent AI company distribution would take years to establish. This is the enterprise radiology AI playbook: partner with the hardware manufacturer who already owns the relationship with the radiology department.
What This Means For You
FQHC executives and CHC leaders: Breast ultrasound AI with existing CPT reimbursement codes means this technology does not require a de novo reimbursement fight. If your radiology partners are deploying DeepHealth or equivalent platforms, you can bill today. The equity argument is also direct: dense breast tissue is more common in the patient populations FQHCs serve. Standardized AI detection at community clinics closes a detection gap that currently favors patients who can access academic medical centers.
Health system administrators and CMOs: The 37% reduction in interpretation time is not a pilot result from an optimized research site. It is a multi-reader validation number from 16 board-certified radiologists. Model what that number means for your breast imaging volume. If you are running 50,000 breast ultrasounds per year, AI-assisted interpretation at that efficiency level frees the equivalent of several full-time radiologist slots. That is real capacity, not hypothetical headcount savings.
Radiologists and breast imagers: The report generation capability is the piece that changes your workflow most directly. Your value shifts from generating the primary document to validating and signing a pre-structured report. That is a different cognitive task. Prepare your department for what it means to review AI-generated reports at volume, including establishing local protocols for when and how you override AI characterizations.
Healthcare investors and founders: RadNet's Digital Health ARR trajectory - $97 million heading toward $140 million - is what vertical integration of AI into owned distribution looks like financially. The platform model beats the point solution model because it generates multiple billable events per patient encounter, not one. Watch how DeepHealth prices the breast ultrasound module relative to its mammography suite.
Policy advocates: Dense breast notification laws already exist in 38 states, requiring providers to tell patients when their mammography is limited by tissue density and to discuss supplemental screening. AI-standardized breast ultrasound is the supplemental screening option those laws were written to enable. This clearance has direct policy relevance for any state still debating dense breast notification requirements.
Closing
End-to-end AI workflow automation in radiology is no longer a research thesis. DeepHealth Breast Ultrasound cleared the FDA, has reimbursement codes, and is deploying across a national network of 400+ imaging centers before the end of 2026. The model is real.
The question for every imaging leader is not whether this technology exists. It exists. The question is whether your procurement governance, your payer contracting, and your liability framework are ready to absorb it at scale.
The operators who answer those questions now will be billing on AI-generated reports while their competitors are still in pilot mode.
One more thing worth noting: the diagnostic gap this platform addresses - inconsistent detection in dense breast tissue at community-level practice settings - is structurally identical to the gap Oatmeal Health addresses in lung cancer screening. The modality is different. The problem is the same. AI that standardizes performance regardless of operator experience or practice setting is not a nice-to-have in communities that cannot afford to lose patients to late-stage diagnoses.
Reply to this newsletter if you are deploying breast ultrasound AI or evaluating it. I read every response.
About the Author
Jonathan Govette is the Co-Founder and CEO of Oatmeal Health, an AI lung cancer diagnostic company catching cancers earlier in the communities that need it most. Oatmeal uses AI to identify unscreened high-risk patients, navigate them to care, and score every lung CT for malignancy risk - billed under CPT 0721T. Stage I survival is 77%. Stage IV is 9%. We work in FQHCs because that gap is largest there.
Jonathan writes daily about radiology, pulmonology, AI diagnostics, health policy, hospital operations, and healthcare startups.
Subscribe to stay ahead of healthcare's most important shifts. Weekly deep-dives on AI, radiology, health policy, FQHCs, and the business of healthcare - written for operators, clinicians, and investors who want the signal, not the noise. Subscribe at oatmealhealthjonathangovette.substack.com
Key References
DeepHealth FDA 510(k) Clearance Press Release, GlobeNewswire, July 30, 2026 - https://www.globenewswire.com/news-release/2026/07/30/3336456/0/en/deephealth-receives-fda-clearance-for-ai-powered-breast-ultrasound.html
HIT Consultant coverage of DeepHealth FDA clearance, August 3, 2026 - https://hitconsultant.net/2026/08/03/deephealth-receives-fda-clearance-ai-breast-ultrasound-radnet/
National Breast Cancer Foundation 2026 statistics - https://www.nationalbreastcancer.org/breast-cancer-facts/
Automated Breast Ultrasound Systems Market Analysis, Coherent Market Insights, 2026 - https://www.coherentmarketinsights.com/market-insight/automated-breast-ultrasound-systems-market-78
AI in Breast Imaging Market Report, GlobeNewswire, January 2026 - https://www.globenewswire.com/news-release/2026/01/26/3225347/0/en/AI-in-Breast-Imaging-Market-Report-2026.html