A retrospective study presented at the Society of Breast Imaging Symposium in Seattle, held April 16 to 19, 2026, found that AI detection software identified signs of breast cancer on mammograms that radiologists had already reviewed and cleared as normal, in some cases up to three years before the cancer was formally diagnosed. Researchers examined 341 women, with an average age of 66, who had screening-detected breast cancer along with at least three prior negative screening exams using digital breast tomosynthesis, a 3D mammography technique.
What the AI Found in Retrospect
Using Genius AI Detection 2.0, a commercial AI system made by Hologic, researchers reran the women’s earlier “normal” scans and found the software flagged signs consistent with breast cancer on the single most recent prior exam in 26.8% of patients, according to findings reported by Diagnostic Imaging. Looking back across all three prior negative exams, the AI flagged suspicious findings in 11% of patients, suggesting that in a meaningful subset of cases, subtle imaging signs were present for years before the cancer was caught through standard screening.
Why This Matters for Screening Intervals
Breast cancer screening guidelines in the United States generally recommend mammograms every one to two years for average-risk women starting in their 40s, a schedule built around balancing early detection against the costs and harms of over-screening, including false positives and unnecessary biopsies. If AI tools can reliably catch signs of cancer that human radiologists miss on routine reads, it raises the possibility that earlier detection is achievable without changing screening frequency at all, simply by adding an AI second read to scans already being taken.
Consistent With Broader Real-World Evidence
This retrospective finding builds on a growing body of prospective evidence. In real-world implementation studies of AI-supported mammography screening at scale, radiologists working with AI assistance have achieved breast cancer detection rates of 6.7 cancers per 1,000 women screened, about 17.6% higher than the 5.7 per 1,000 detection rate achieved without AI support, according to population-level screening data. Separately, comparative studies have found AI-assisted double reading can match or exceed the cancer detection performance of two independent radiologists reading the same scans, the long-standing gold standard in several European screening programs, while potentially reducing radiologist workload.
The Case for Caution
Radiologists and screening researchers have urged caution against over-interpreting retrospective “AI would have caught it” studies, since flagging a suspicious area on a scan after the fact, once the outcome is already known, is a different and easier task than making a blind, prospective call on a live patient during actual clinical practice. Critics also note that increasing sensitivity to catch more true cancers earlier risks increasing false positives, which can lead to unnecessary biopsies, patient anxiety, and additional imaging costs, and that AI-flagged findings still require radiologist confirmation before triggering any clinical action.
A Billing and Access Problem Slowing Adoption
Even as evidence mounts for AI’s clinical value, practical barriers remain. Because AI-assisted mammography readings are still relatively new, there is currently no dedicated insurance billing code for the AI portion of the reading, according to reporting on the technology’s rollout, and some clinics have begun charging patients an additional $40 to $100 out-of-pocket fee for an AI-assisted read, a cost that falls unevenly on patients depending on where they get screened and whether they can afford the added expense.
What It Means and What’s Next
The Seattle findings add to mounting pressure on radiology practices, insurers, and regulators to formalize how AI-assisted mammography fits into standard screening protocols and how it should be reimbursed. If professional societies and insurers move to cover AI reads as a standard part of screening rather than an optional add-on, the technology could shift from a paid upgrade available mostly to patients who can afford it into a universal safeguard layered onto routine mammograms. Researchers say the next step is larger prospective trials that track whether earlier AI-flagged detection actually translates into earlier treatment and better survival outcomes, rather than simply moving the diagnosis date earlier without changing the course of the disease.
