Women with abnormal mammograms often face weeks of uncertainty before receiving additional diagnostic tests or a definitive diagnosis. Researchers at the University of California, San Francisco (UCSF) and the University of California, Berkeley have now demonstrated how artificial intelligence can significantly shorten that waiting period by identifying patients at the highest risk of breast cancer and prioritising them for accelerated follow-up care.
The study, published in npj Digital Medicine, evaluated an AI-supported workflow that enables high-risk patients to move from screening mammography to diagnostic imaging and, when necessary, biopsy within a much shorter timeframe than is typically possible.
“This is really an exciting time,” said first author Maggie Chung, MD. “This moves us closer to personalized care, where we can tailor a plan so that each patient gets the right intervention at the right time.”
Identifying high-risk patients
The researchers used Mirai, an open-source AI model developed by study co-author Adam Yala, PhD, a data scientist at UC Berkeley. The model was trained on hundreds of thousands of mammograms linked to cancer outcomes and is designed to detect subtle imaging patterns associated with future breast cancer risk.
The team applied the model to more than 4,100 screening mammograms performed at Zuckerberg San Francisco General Hospital and Trauma Center. Mirai identified 525 women—approximately 12.7 percent of those screened—as being at elevated risk.
These patients were offered immediate interpretation of their mammograms and, when indicated, same-day diagnostic imaging. Some women requiring biopsies were also able to undergo the procedure on the same day.
Substantial impact
According to the researchers, the impact on waiting times was substantial. Diagnostic evaluation times were reduced from several weeks to approximately one hour. For women ultimately diagnosed with breast cancer, the average wait for a biopsy fell from more than two months to fewer than ten days.
Importantly, Mirai does not function as an autonomous diagnostic system. Instead, it serves as a triage tool that helps radiologists and clinicians determine which patients are most likely to benefit from expedited assessment.
“This is a powerful example of how AI can be a collaborative partner for physicians,” said Yala. “It shows how we can improve care when we bring clinicians and data scientists together to design these systems.”
Personalised screening pathways
Before introducing the workflow, researchers analysed more than 114,000 archival mammograms to ensure that the system could reliably identify high-risk patients without overwhelming available diagnostic capacity.
The findings support a broader shift toward risk-based screening and diagnostic pathways. Rather than following a largely uniform screening schedule, future breast cancer programmes may increasingly tailor follow-up strategies to individual patient risk profiles.
“Right now, many women follow the same screening schedule but their individual risk can be very different,” Chung noted. “AI risk assessment gives us the chance to identify the women most likely to benefit from expedited care and get them what they need.”
AI’s growing role in breast cancer care
The UCSF study is the latest example of how AI is being integrated into breast cancer screening and diagnostics. Earlier this year, ICT&health reported on a large UK study involving more than 175,000 women that found Google's AI system could detect breast cancer on screening mammograms with accuracy comparable to, and in some cases exceeding, that of human radiologists.
Conducted across multiple NHS screening programmes, the study demonstrated that AI could increase cancer detection rates, reduce false-positive findings and lower radiologists’ workload by more than 30 percent. Researchers suggested that such systems could help address growing workforce shortages while improving screening efficiency and enabling radiologists to focus on more complex clinical tasks.
In 2025, ICT&health also covered research from UCLA Health showing that AI may help identify interval breast cancers—tumours that emerge or become apparent between routine screening examinations. In that retrospective study of nearly 185,000 mammograms, AI flagged more than three-quarters of cases that were later associated with interval breast cancer.
Researchers estimated that AI-assisted screening could potentially reduce the number of interval breast cancers by approximately 30 percent. However, they also emphasised that further validation is required, particularly regarding the accuracy of AI-generated image markers and the interpretation of suspicious findings.
Supporting, not replacing clinicians
Taken together, the studies illustrate how AI is increasingly being explored across the entire breast cancer pathway—from risk prediction and screening support to earlier detection and faster diagnosis.
Researchers consistently emphasise that these technologies are intended to augment, rather than replace, clinical expertise. Instead of making autonomous decisions, AI systems are increasingly being positioned as decision-support tools that help clinicians prioritise patients, identify subtle abnormalities and allocate scarce healthcare resources more effectively.
As healthcare systems worldwide continue to face growing demand for breast cancer screening and diagnostic services, AI-supported workflows may become an important component of efforts to improve both efficiency and patient experience.
