![\"Physician](\"https:\/\/rtmedical.com.br\/wp-content\/uploads\/2026\/08\/ultrassom-mama-ia-deephealth.jpg\")
The pitch is not replacing the radiologist. It is standardizing the most fragile step of the exam: lesion description. Instead of each reader typing morphologic descriptors by hand, the system extracts measurements, classifies findings using the ACR BI-RADS lexicon and returns a structured draft report.<\/p>\n
The multi-reader study numbers<\/h2>\n
The data behind the clearance come from a multi-reader, multi-case (MRMC) study run with 16 U.S. board-certified radiologists at selected imaging centers and hospitals. Three results were reported by the company:<\/p>\n
- \n
- Lesion localization accuracy above 98%<\/strong> \u2014 when a lesion exists, the software points to the right place.<\/li>\n
- An 8% gain in sensitivity<\/strong> for breast cancer detection versus unassisted reading.<\/li>\n
- A 37% reduction in radiologist interpretation time.<\/strong><\/li>\n<\/ul>\n
MRMC is the gold-standard design for evaluating decision-support tools in imaging, precisely because it measures the same reader with and without the software, controlling for between-reader variability. Worth noting, though: these are manufacturer-reported figures submitted through the regulatory pathway. A peer-reviewed publication with full methodological detail is the natural next step.<\/p>\n
How the tool actually works<\/h2>\n
The workflow has three layers. First, automated lesion detection on standard breast ultrasound images. Second, characterization: the system classifies shape, orientation, margin, echo pattern and posterior features \u2014 exactly the descriptors in the American College of Radiology BI-RADS lexicon. Third, automated report generation, already populated with standardized measurements captured during acquisition.<\/p>\n
That last layer is what changes the sonographer’s day. Today a sizable share of scanning time goes to manual measurement and transcription. Automating standardized measurement extraction removes a classic source of typing error and of mismatch between what was measured and what was reported.<\/p>\n
Why breast ultrasound needed this<\/h2>\n
Among breast imaging methods, ultrasound is the most operator-dependent. The image only exists while the transducer is in position \u2014 there is no complete volumetric dataset to reinterpret later, as there is with tomosynthesis or MRI. If the lesion was not documented in the right plane, it simply is not in the study.<\/p>\n
Add to that the variability in applying the BI-RADS lexicon. The boundary between category 3 (probably benign, six-month follow-up) and 4a (low suspicion, biopsy indicated) rests on subtle morphologic judgment. That band is where avoidable biopsies and under-called follow-ups both concentrate. Standardizing descriptors attacks that directly.<\/p>\n
Volume matters too. Ultrasound remains the main supplemental exam in dense breasts and stays central to the supplemental screening debate \u2014 we have asked whether screening ultrasound is still needed in the DBT era<\/a>, and covered evidence that breast MRI outperforms mammography in dense breasts<\/a>. Under any scenario, breast ultrasound volume is not going down.<\/p>\n
Reimbursement, scale and RadNet’s role<\/h2>\n
A detail routinely missed in AI headlines: this product ships with a reimbursement path. DeepHealth says U.S. customers can pursue payment under an existing Category III CPT code for quantitative ultrasound tissue characterization \u2014 subject, as always, to individual payer policy.<\/p>\n
Deployment scale is just as notable. The solution is slated to roll out across RadNet’s network of centers by the end of 2026, covering an estimated 700,000-plus breast ultrasound studies per year that may be eligible. That is one of the largest real-world AI imaging deployments announced to date, and it turns RadNet itself into the product’s biggest field laboratory.<\/p>\n
“With DeepHealth’s solution, we can achieve greater standardization of workflows, improving consistency while saving time,” said Dr. Jason McKellop, RadNet California Women’s Imaging Medical Director.<\/p>\n
What changes for the service<\/h2>\n
For radiologists, the most immediate gain is consistency rather than diagnosis. A report that always carries the same five descriptors, in the same order, with traceable measurements, is a report that is comparable over time \u2014 an indispensable condition for judging nodule growth on follow-up.<\/p>\n
For managers, the math is throughput: a 37% cut in interpretation time, if it holds outside the study environment, changes schedule design. Implementation still demands attention to three practical points: integration with the existing PACS and reporting system, a standardized acquisition protocol across technologists, and an explicit review process \u2014 the AI output is a draft, not a signed report.<\/p>\n
Outside the U.S., commercialization depends on local regulatory registration, and the Category III CPT reimbursement model rarely has a direct equivalent. The right read is to treat this as trend signaling \u2014 the same direction as the recently cleared software that guides bedside DVT ultrasound<\/a>: AI is migrating from the static exam to the operator-dependent one.<\/p>\n
Limitations and what to watch<\/h2>\n
Three honest reservations. First, the performance figures are manufacturer-reported and come from a study with 16 readers at selected centers; performance in unselected populations tends to be lower. Second, the sensitivity gain has to be read alongside its effect on specificity, which was not published with equal prominence \u2014 any tool that increases detection risks increasing benign biopsies. Third, 510(k) clearance attests substantial equivalence to a predicate device, not clinical outcome benefit.<\/p>\n
What will be worth watching over the coming months is precisely the real-world data from the RadNet network: recall rate, biopsy positive predictive value and reporting time before and after deployment, across hundreds of thousands of exams. If those indicators move in the right direction at scale, this stops being a regulatory footnote and becomes evidence.<\/p>\n
Sources:<\/strong> AuntMinnie \u2014 DeepHealth lands FDA nod for breast ultrasound AI software<\/a> and the official DeepHealth announcement<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"
FDA clears DeepHealth breast ultrasound AI: 98% lesion localization accuracy and 37% less reading time. See the data and the caveats.<\/p>\n","protected":false},"author":1,"featured_media":18873,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"ngg_post_thumbnail":0,"_rt_cluster":"","fifu_image_url":"","fifu_image_alt":"","footnotes":""},"categories":[102,100],"tags":[],"class_list":["post-18887","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai","category-radiology"],"aioseo_notices":[],"rt_seo":{"title":"","description":"FDA clears DeepHealth breast ultrasound AI: 98% lesion localization accuracy and 37% less reading time. See the data and the caveats.","canonical":"","og_image":"","robots":"index,follow","schema_type":"Article","include_in_llms":true,"llms_label":"FDA clears DeepHealth breast ultrasound AI","llms_summary":"The FDA granted 510(k) clearance (K260303) to DeepHealth Breast Ultrasound on July 30, 2026; the software detects and characterizes lesions using the BI-RADS lexicon, with over 98% localization accuracy, an 8% sensitivity gain and a 37% cut in interpretation time.","faq_items":[],"video":[],"gtin":"","mpn":"","brand":"","aggregate_rating":[]},"_links":{"self":[{"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/posts\/18887\/"}],"collection":[{"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/posts\/"}],"about":[{"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/types\/post\/"}],"author":[{"embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/users\/1\/"}],"replies":[{"embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/comments\/?post=18887"}],"version-history":[{"count":1,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/posts\/18887\/revisions\/"}],"predecessor-version":[{"id":18889,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/posts\/18887\/revisions\/18889\/"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/media\/18873\/"}],"wp:attachment":[{"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/media\/?parent=18887"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/categories\/?post=18887"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/tags\/?post=18887"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
- An 8% gain in sensitivity<\/strong> for breast cancer detection versus unassisted reading.<\/li>\n