NEW YORK, July 12 — Under the hood of a new artificial intelligence system developed at New York University sits a tool that may soon give clinicians a faster, more accessible read on breast cancer recurrence risk. The technology, which combines digital images of tumour tissue with basic clinical information, can match or even outperform widely used genomic tests that currently guide treatment decisions. The detail that matters: those genomic tests analyse gene activity in tumour tissue to estimate the likelihood of the cancer returning.
The NYU team’s AI alternative offers a potentially faster path to the same answer. Breast cancer treatment has advanced over recent decades, and many patients now survive and go on to live full lives.
Recurrence remains a serious concern. Doctors frequently face difficult choices about how aggressively to treat the disease. Some patients receive chemotherapy that may not be necessary.
Others may need more treatment than they receive. Krzysztof J.
Geras, who led the study, stated: “Breast cancer is not a single disease, and decisions about how aggressively to treat it are often difficult.” The technical reality is that the AI system was trained on digital images of tumour tissue combined with basic clinical information. No exotic inputs, no proprietary genomic assays. Just images and standard information, processed through machine learning models that learned to spot patterns associated with recurrence.
What this means in practice: a pathologist could upload a digital slide and get a risk estimate back faster than genomic tests sometimes require. For patients sitting in the anxious gap between diagnosis and treatment plan, speed is not a luxury. The NYU team’s findings show the tool matching or exceeding the accuracy of those established genomic tests.
That is a significant claim, because the genomic tests are deeply embedded in clinical guidelines. They determine whether a woman with early-stage breast cancer receives chemotherapy or hormonal therapy alone.
An AI that can replicate or improve on that stratification without the cost and turnaround time of genetic sequencing would change the economics and logistics of breast cancer care. Geras and his colleagues are careful not to overstate. The study is published, the data is public, and the next step is prospective validation in clinical settings.
But the direction of travel is clear: artificial intelligence, trained on the kind of tissue slides that hospitals already produce as a matter of routine, may soon give oncologists a second opinion that arrives before the patient leaves the clinic. The system does not replace the doctor.
It gives the doctor better information, faster. That, as ever, is the point. Consult your doctor for medical advice.
