Source: GlobeNewswire — DeepHealth, Inc. (RadNet) — 2026-07-30
Summary
DeepHealth, RadNet's AI subsidiary, received FDA 510(k) clearance for an AI tool that automates lesion detection, characterization, and reporting in breast ultrasound exams. A multi-reader, multi-case validation study involving 16 U.S. board-certified radiologists found the tool improved breast-cancer-detection sensitivity by 8%, achieved greater than 98% lesion-localization accuracy, and cut radiologist interpretation time by 37%. RadNet plans to roll the platform out across its national outpatient imaging network by the end of 2026, a deployment expected to touch an estimated 700,000+ annual breast ultrasound exams.
Key Takeaways
- This clearance comes with an actual reported effect size — 8% sensitivity improvement, 37% time reduction, >98% localization accuracy — which is rarer than it should be among AI-diagnostic FDA clearances, many of which clear on safety/equivalence grounds without a published performance delta.
- A 37% reduction in interpretation time, applied across an estimated 700,000+ annual exams once fully deployed, is a workflow-capacity story as much as an accuracy story — it effectively expands how many exams the same radiologist workforce can read.
- The validation design — 16 board-certified radiologists, multi-reader multi-case — is a credible methodology for this kind of claim, though it's still a pre-deployment study; the real test is whether the sensitivity gain holds up once it's running across RadNet's full outpatient network.
- Breast ultrasound is typically used as a supplemental tool alongside mammography, especially for dense-breast-tissue patients where mammography alone is less sensitive — so an accuracy gain here has outsized relevance for exactly the patient population mammography serves worst.
Reel Script
Hook (~16s, 36 words): The FDA just cleared an AI tool that made radiologists 37% faster at reading breast ultrasounds — and more accurate at catching cancer while doing it. That combination is unusually hard to pull off.
Core Concept (~55s, 125 words): Breast ultrasound is often used alongside mammography, particularly for patients with dense breast tissue where mammograms alone miss more. Reading one is a manual, time-consuming process: a radiologist scans the images looking for lesions, characterizes what they find, then writes up a report — three separate cognitive tasks stacked together. DeepHealth's tool automates all three: it detects lesions in the scan, characterizes what it's looking at, and drafts the report, with the radiologist reviewing and signing off rather than doing the raw detection work from scratch. That's the mechanism behind the time savings — the AI does the first-pass scan, the human does the judgment call.
Hands-On (~50s, 115 words): The numbers, straight from the validation study: 16 board-certified U.S. radiologists tested it in a multi-reader, multi-case design — a standard methodology for this kind of claim — and found greater than 98% lesion-localization accuracy, an 8% improvement in cancer-detection sensitivity over the unaided baseline, and a 37% cut in interpretation time. Worth putting all three on screen as a simple before/after bar comparison: baseline sensitivity versus AI-assisted sensitivity, and baseline reading time versus AI-assisted reading time. RadNet's rollout plan — full national outpatient deployment by end of 2026, an estimated 700,000+ annual exams — is the scale context that makes the time number matter.
Takeaway (~22s, 48 words): An 8% sensitivity gain sounds modest until you remember it's layered on top of a 37% speed gain — this isn't a tradeoff between accuracy and throughput, it's a rare case of getting both. Worth watching whether that combination holds once it's running at RadNet's full scale.