Can AI Find Breast Cancer Years Earlier Than Radiologists?

The most provocative breast imaging story this week is not about convenience, but lead time. If AI can identify breast cancer on DBT substantially earlier than human readers, it could alter the entire logic of screening by shifting the goal from detection at first visibility to detection at earliest measurable signal.

That kind of claim deserves scrutiny, because earlier detection is only meaningful if it translates into better outcomes without overwhelming the system with false alarms. In imaging, an algorithm that finds more subtle abnormalities can also create more work, more biopsies, and more anxiety unless it is paired with strong risk calibration and clinical triage rules.

But the interest is real because DBT has already become a rich data source for machine learning. Tomosynthesis offers more volumetric information than standard mammography, which makes it attractive for models that can learn subtle patterns across slices. If those patterns reliably precede conventional radiologic findings, AI could become less of a reviewer and more of a predictive biomarker.

The practical test will be whether these algorithms can show durable benefit in prospective studies and across diverse screening populations. The field has learned that high retrospective performance is not enough; the next benchmark is whether earlier flagging changes treatment timelines, stage at diagnosis, and ultimately survival in a way that justifies broad adoption.