FDA clears the first AI that predicts breast cancer recurrence from an MRI: the hard part will be proving its value

By FOX 13 Tampa Bay · October 16, 2026.

The U.S. Food and Drug Administration (FDA) has authorized TumorSight Risk, an artificial intelligence software product from Chicago-based SimBioSys. According to the company, it is the first MRI-based AI tool able to predict breast cancer recurrence.

According to FDA documents cited by the outlet, the software calculates the risk of recurrence at 5 and 10 years. It draws on clinical data (age, race, stage, lymph node involvement and tumor grade) and on MRI scans of patients already diagnosed with invasive breast cancer. It returns a score from 0 to 100 that classifies risk as high, intermediate or low.

SimBioSys says no additional procedures, tissue shipments or specialized hardware are needed because the tool uses data that already exists. Its president and CEO, Stacey Stevens, says that for the first time clinicians can use the biological and spatial information in the MRI, together with pathology and clinical factors, to get a personalized view of risk.

The context shows the scale of the problem. According to the American Cancer Society, about 322,000 new cases of invasive breast cancer will be diagnosed in women in the U.S. in 2026, and some 42,000 will die of the disease.

Our take: the most interesting part is not the AI itself but the fact that it extracts value from data that has already been paid for. An MRI holds far more information than the human eye can sum up in a report, and turning it into a risk estimate without extra tests could make personalized treatment cheaper. That fits the broader thesis: medical AI makes progress when it uses existing data without adding to the patient's burden.

That said, a caveat is in order. FDA authorization of a piece of software is not the same as proof that it improves patient outcomes. The article does not say what population the model was validated on or how accurate it is, and that is the key question: variables such as race are among the inputs, so it matters whether it works equally well across all groups. Until that data is published, performance figures and the adoption timeline remain unknown; the only confirmed facts are the authorization and the company's own claims.

There is also a practical question: what will an oncologist do with an "intermediate" score? A prognostic tool is only useful if it changes decisions, such as intensifying or avoiding treatment, and that will require prospective clinical evidence.

In the long term, tools like this point toward an oncology that treats each patient according to their actual risk rather than averages, a step toward reducing both overtreatment and unexpected recurrences. In the short term, the bar is the same as always in clinical AI: transparent validation across populations before any euphoria.