At Zuckerberg San Francisco General Hospital and Trauma Center, women with abnormal mammograms no longer face weeks of uncertainty waiting for a diagnosis—thanks to an AI triage tool called Mirai that researchers at UC San Francisco and UC Berkeley have now put into real-world practice.
The waiting game for women with suspicious mammograms is brutal. An abnormal scan means scheduling additional imaging, then waiting for more appointments, and potentially enduring two months or more before getting answers about a biopsy. That stretch of anxiety compounds the stress of an already frightening medical situation. But Mirai, an open-source AI model developed by UC Berkeley data scientist Adam Yala, Ph.D., is changing that timeline by identifying which patients need immediate attention—and getting them answers the same day.
Trained on hundreds of thousands of mammograms linked to actual patient cancer outcomes, Mirai learns to spot subtle patterns that signal higher risk. When researchers deployed the model to analyze more than 4,100 screening mammograms at the San Francisco hospital, it identified 525 women—about 12.7% of those screened—as high-risk. These patients received their mammogram interpretation immediately after imaging and had access to additional diagnostic work and sometimes even biopsies on the same day they were scanned. For those ultimately diagnosed with breast cancer, the average wait time for a biopsy plummeted from more than two months to fewer than 10 days.
Maggie Chung, MD, the study's first author, emphasizes that this acceleration matters profoundly. "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," she said in describing the work published in npj Digital Medicine. The model reduced diagnostic wait times from several weeks to roughly an hour—a transformation that turns months of worry into a single afternoon of answers.
Importantly, Mirai doesn't replace radiologists or make diagnoses independently. Instead, it works as a collaborative partner, helping physicians identify patients who benefit most from expedited evaluation. Yala stresses this partnership: "It shows how we can improve care when we bring clinicians and data scientists together to design these systems." Before launching the program, researchers analyzed more than 114,000 archival mammograms to ensure the AI would catch enough high-risk patients without overwhelming the clinic with unnecessary expedited referrals.
The real innovation here is moving beyond one-size-fits-all screening schedules. Right now, most women follow identical screening intervals regardless of their individual risk—a blunt instrument when risk varies dramatically from person to person. AI risk assessment opens a path toward truly personalized medicine, where a woman's screening frequency and intervention strategy match her actual likelihood of developing cancer. For some, that might mean routine screening; for others flagged as high-risk by Mirai, it means the kind of same-day evaluation that turned this San Francisco hospital into a model for faster, less anxious diagnosis.
As this work spreads beyond one institution, it points toward a future where artificial intelligence doesn't replace human expertise but amplifies it, ensuring that every patient gets not just care, but the right care at the right moment.