{"id":19206,"date":"2026-08-31T05:28:23","date_gmt":"2026-08-31T08:28:23","guid":{"rendered":"https:\/\/rtmedical.com.br\/tmp-en-1788164902681\/"},"modified":"2026-08-31T05:28:30","modified_gmt":"2026-08-31T08:28:30","slug":"breast-imaging-research-roundup-2026","status":"publish","type":"post","link":"https:\/\/rtmedical.com.br\/en\/breast-imaging-research-roundup-2026\/","title":{"rendered":"Breast Imaging: The Studies That Defined 2026"},"content":{"rendered":"

The breast imaging research roundup published by Diagnostic Imaging<\/em> gathers a set of studies that, read together, tell a rather different story from the “AI will find more cancer” narrative. The consolidated gains in this cycle sit in efficiency<\/strong> \u2014 less reading time, fewer images per exam, shorter MRI protocols \u2014 and in modality choice<\/strong>, with contrast-enhanced mammography moving out of its niche. Additional detection, where it appears, comes bundled with more recalls. Each front is worth unpacking.<\/p>\n

Tomosynthesis: the gain is now time, not detection<\/h2>\n

Digital breast tomosynthesis (DBT) is already standard across much of screening, and the problem it created is well known: it multiplied the number of images per exam and, with them, reading time. Recent research went straight at that point.<\/p>\n

\"Health
Tomosynthesis, abbreviated MRI, ultrasound and contrast-enhanced mammography all advanced in parallel this research cycle. Photo: Klaus Nielsen\/Pexels<\/figcaption><\/figure>\n

In multireader, multicase studies, concurrent AI support cut mean reading time from about 54.4 to 48.5 seconds per exam \u2014 small in absolute terms, large when multiplied across thousands of studies in a screening shift. Work on concurrently applied computer-aided detection reported reading time reductions as high as 29.2% while maintaining interpretation performance. Where the AI system was accurate, concurrent use improved AUC, sensitivity and specificity and lowered the recall rate.<\/p>\n

The most interesting line, though, is AI-based slab reconstruction. Rather than improving the read, it reduces the material to be read, grouping thin sections into thicker synthetic slices. A retrospective review of 119,662 screening DBT examinations<\/strong> published in Radiology<\/em> found no evidence of a difference in cancer detection rate before and after the technology was implemented. That is the kind of “null” result that matters a great deal: if detection does not change and image volume falls, the workflow gain is clean.<\/p>\n

Abbreviated breast MRI consolidates as viable screening<\/h2>\n

A study published in February in Radiology<\/em> showed that screening with an abbreviated MRI protocol sustains positive outcomes over time \u2014 the missing piece, since the standing criticism of short protocols was always that they might miss lesions on longer follow-up.<\/p>\n

The logistics numbers explain the fast adoption. Abbreviated protocols cut acquisition time from 25 to 35 minutes down to roughly 8 minutes, with mean reading time just over one minute. That changes the economics of the exam: instead of two patients per magnet hour, a service accommodates five or six. For a modality whose main barrier was always access and cost rather than sensitivity, shortening the protocol is the highest-leverage intervention available. The head-to-head comparison of abbreviated MRI, full MRI and contrast-enhanced mammography has already been published in Radiology<\/em> and is required reading for anyone designing a supplemental program.<\/p>\n

Contrast-enhanced mammography leaves the niche<\/h2>\n

Contrast-enhanced mammography (CEM) drew the most new evidence in this cycle, along two distinct lines.<\/p>\n

On the diagnostic front, a prospective study in dense breasts with multifocal and multicentric carcinoma found index lesion detection in 96.8% of cases with CEM, against 69.2% with tomosynthesis and 51.9% with digital mammography<\/strong>, along with the highest lesion conspicuity and reader confidence. That is a gap wide enough to change surgical planning, and this setting \u2014 local staging in the dense breast \u2014 is where CEM has its strongest argument.<\/p>\n

On the screening front, the interim analysis of the prospective TOCEM trial, also in Radiology<\/em>, evaluated adding CEM to tomosynthesis in women with a personal history of breast cancer and found increased cancer detection with a limited increase in recalls. “Limited” is the operative word: contrast in screening always risks trading detected cancer for unnecessary biopsy, and the trial suggests the balance is manageable in this specific population.<\/p>\n

One subtle, practical finding rounds it out: in CEM-detected lesions, adding tomosynthesis shifted BI-RADS classification toward the true pathology, and the effect was larger among less experienced readers<\/em>. In other words, the two modalities are not competitors \u2014 combined, they reduce interobserver variability, one of the persistent problems in breast imaging.<\/p>\n

Risk AI and the automation bias problem<\/h2>\n

The roundup also covers AI models that flag cancer risk years before diagnosis and, in the same breath, a cautionary study on automation bias. The juxtaposition is not accidental: they are two faces of the same system.<\/p>\n

On the promising side, image-based risk models have already shown they can stratify risk from an apparently normal mammogram, and patients accept that stratification well \u2014 as shown by the study we covered on acceptance of AI-based breast risk assessment<\/a>. On the cautionary side, the accumulated evidence on automation bias is consistent: when the algorithm marks, the reader tends to confirm; when it does not, the reader tends to relax. We detailed the mechanism in our analysis of what protects a radiologist from erring alongside the model<\/a> \u2014 expertise and context, not trust in the output.<\/p>\n

Ultrasound remains a complement, not a substitute<\/h2>\n

Ultrasound held the role the evidence assigns it: a supplemental modality in the dense breast and a characterization tool, not primary screening. The practical question the tomosynthesis era opened \u2014 how much ultrasound is still needed once DBT has already cut the recall rate \u2014 remains open, and we reviewed the available data in screening ultrasound in the DBT era<\/a>. What is new is automation: AI systems cleared for breast ultrasound, as in the FDA clearance granted to DeepHealth<\/a>, shift the bottleneck from the operator to the interpreter.<\/p>\n

What to do with this in your department<\/h2>\n

Four decisions fall out of this body of evidence. First: if the service runs DBT and reading time is the bottleneck, AI slab reconstruction offers the best evidence-to-risk ratio in the package \u2014 it cuts image volume with no detection cost documented in a large series.<\/p>\n

Second: if there is demand for supplemental screening in dense or high-risk breasts and the magnet is the constraint, an abbreviated protocol is the highest-impact change, and it is a protocol change rather than a purchase. Third: CEM deserves deployment where local staging of the dense breast happens and where MRI is unavailable \u2014 but it requires an iodinated contrast protocol, with all the burden of renal function screening and reaction management.<\/p>\n

Fourth, on governance: any risk or detection AI tool must arrive with local audit of recall rate and positive predictive value before and after. Without that measurement, a department cannot tell whether it gained detection or merely biopsies.<\/p>\n

Limits of the roundup<\/h2>\n

A roundup has structural limits that need stating. It aggregates studies of very different designs \u2014 large retrospective series, small prospective cohorts, multireader work in a laboratory setting \u2014 and reader performance in a controlled study usually beats performance in routine practice. Interim analyses, such as TOCEM’s, can shift with complete follow-up. And reading-time gains measured in seconds per exam only translate into real capacity if the service has a queue to fill the freed time.<\/p>\n

Even so, the direction is consistent: the next frontier in breast imaging is not detecting more, it is detecting the same with less time, fewer images and a better modality choice per risk profile. For services with a backlog and a shortage of readers \u2014 which is most of them \u2014 that is the frontier that counts.<\/p>\n

Source:<\/strong> Diagnostic Imaging \u2014 Emerging Research in Mammography, Breast MRI, Ultrasound and AI<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

AI in tomosynthesis, abbreviated MRI and contrast-enhanced mammography: the breast imaging research roundup and what to apply.<\/p>\n","protected":false},"author":1,"featured_media":19174,"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-19206","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai","category-radiology"],"aioseo_notices":[],"rt_seo":{"title":"","description":"Breast imaging research roundup: AI in tomosynthesis, 8-minute abbreviated MRI and contrast-enhanced mammography. What changes in practice.","canonical":"","og_image":"","robots":"index,follow","schema_type":"Article","include_in_llms":true,"llms_label":"Breast imaging research roundup 2026","llms_summary":"Roundup of recent breast imaging studies: AI cuts DBT reading time (54.4 to 48.5 s) and slab reconstruction leaves detection unchanged across 119,662 exams; abbreviated MRI drops to ~8 minutes of acquisition; contrast-enhanced mammography detects the index lesion in 96.8% of dense-breast cases versus 69.2% for DBT.","faq_items":[],"video":[],"gtin":"","mpn":"","brand":"","aggregate_rating":[]},"_links":{"self":[{"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/posts\/19206\/"}],"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=19206"}],"version-history":[{"count":1,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/posts\/19206\/revisions\/"}],"predecessor-version":[{"id":19208,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/posts\/19206\/revisions\/19208\/"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/media\/19174\/"}],"wp:attachment":[{"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/media\/?parent=19206"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/categories\/?post=19206"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/rtmedical.com.br\/en\/wp-json\/wp\/v2\/tags\/?post=19206"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}