FDA Clears MRI-Based AI That Scores 10-Year Breast Cancer Return Risk

For many women with early breast cancer, a breast MRI is already part of the work-up before surgery. Software cleared by the US Food and Drug Administration (FDA) can now turn that scan, plus five routine clinical facts, into an estimate of how likely the cancer is to come back within five and ten years.

The product is TumorSight Risk, made by SimBioSys, a Chicago company. The FDA cleared it through the 510(k) route on 25 September 2026, and SimBioSys announced the clearance on 15 October, calling it “the world’s first AI-powered, MRI-derived prognostic tool” for breast cancer recurrence. That “first” is the company’s own claim. The tool drew wider attention on 26 October when NVIDIA featured it on X, describing it as “Powered by NVIDIA technology” without saying more about its role.

This explainer sets out what the software does, who it is meant for, and how far the published evidence goes.

What the software actually reads

The FDA’s clearance letter and 510(k) summary describe a two-step process. First, the patient’s dynamic contrast-enhanced (DCE) MRI, in which images are taken as an injected contrast dye moves through the breast, goes into TumorSight Viz, an earlier SimBioSys product that maps the tumour and surrounding tissue. TumorSight Risk then takes measurements from that map and feeds them, with the patient’s age, race, clinical tumour stage, clinical nodal stage and tumour grade, into what the FDA summary calls “a regression-based machine learning model”.

The output is a score from 0 to 100. A score above 35 counts as high risk, above 20 to 35 as intermediate, and 20 or below as low. The report also shows the recurrence rates seen at five and ten years among patients with similar scores in the validation study. Hormone receptor and HER2 status are checked only to confirm a patient qualifies; they are not used to compute the score. The FDA summary says the tool works with GE and Siemens scanners at 1.5 and 3 tesla.

SimBioSys says the cloud-based software needs “no additional procedure, tissue shipment, or specialized hardware”. That is the practical pitch. The tissue-based genomic tests used today can, the company says, add “cost, logistical requirements and turnaround time”.

Who it is for, and how it was cleared

The FDA’s indications limit TumorSight Risk to adult women with hormone receptor-positive, HER2-negative breast cancer, either node-negative (N0) or with limited spread to nearby nodes (N1), at Stage I to IIIA. SimBioSys says this subtype makes up about 75% of breast cancers in the US. The tool is for prescription use only, and its stated purpose is “to inform a physician in prognostic risk-based decisions in conjunction with other relevant clinicopathological factors”. The 510(k) summary says plainly that it is not intended to establish a diagnosis.

A 510(k) clearance means the FDA judged the device “substantially equivalent” to one already on the market. Here the comparison device, or predicate, is ArteraAI Breast from Artera, which reads stained tissue slides from a surgical specimen and estimates the risk of distant spread. TumorSight Risk does a similar job from imaging taken before treatment, which is why SimBioSys stresses it can be available before surgery. It joins other FDA-cleared AI aids pitched as support for clinicians rather than stand-alone answers, such as the Neurora autism assessment aid TSN covered earlier this month.

Recurrence risk matters because it drives some of the hardest choices in this disease, including whether to add chemotherapy and whether to continue hormone therapy beyond five years. Cutting the chance of a cancer coming back is the target of new treatments too, as with the personalised mRNA melanoma therapy that cut recurrence or death risk by 42% in a phase 3 trial.

What the 2,129-patient study shows

The clearance rests on one validation study of 2,129 women treated at Mayo Clinic, Emory University, Stanford Medicine and Mercy Medical Cedar Rapids, all diagnosed between 2010 and 2024, and all with an MRI before treatment. The FDA summary says the model was built on a separate set of 1,133 patients from six other US sites and locked before validation. Recurrence data were sealed until the device had produced its scores.

The company’s headline figures are that women classed as high risk had a 3.3-fold greater risk of recurrence than those classed as low risk, and that each 10-point rise in the score was linked to a 22% rise in relative risk. Across five score bands, the estimated five-year recurrence risk ran from 2.2% for scores of 0 to 10 up to 13.6% for scores of 58 to 100; at ten years the range was 6.2% to 34.2%. The score stayed linked to recurrence after allowing for stage, grade, age, race and treatment, the company says.

The FDA summary adds figures for the three cleared risk groups. Five-year recurrence was 3.0% in the low group, 5.7% in the intermediate group and 9.8% in the high group, rising to 8.6%, 13.4% and 26.4% at ten years.

The results were presented at the 2026 American Society of Clinical Oncology (ASCO) Annual Meeting. The published abstract, led by Mayo Clinic’s Matthew Goetz, used a single high-versus-low cut-off of 47 and reported a hazard ratio of 2.71, with a median follow-up of 6.24 years. It confirms the 22% per 10 points figure. The 3.3-fold figure in the company’s release does not appear in the abstract, and the release does not explain the difference. The abstract names SimBioSys as its funder, and the release notes that “Mayo Clinic has a financial interest with SimBioSys”.

Sorting risk is not the same as changing care

The evidence so far supports a narrow claim: in old records from four US centres, the score sorted women into groups whose cancers came back at clearly different rates, and it added information beyond the usual clinical factors. That is a genuine step for a tool that reuses a scan many patients already have. It does not show that oncologists using the score make better choices, or that any woman lives longer or avoids chemotherapy she did not need. A 510(k) clearance tests equivalence to an existing device, not benefit to patients.

The spread of the numbers also deserves a careful reading. Even in the high-risk group, roughly nine in ten women had no recurrence within five years, according to the FDA figures. The score shifts the odds; it does not tell a woman what will happen to her.

The study population also leaves gaps. SimBioSys says node-positive women have historically had fewer validated options, yet only 15% of the validation group, 318 women, were N1, against 25% in the US figures the FDA table cites. Grade 3 tumours were underrepresented too. And because race is a direct input, with every group other than White and Black coded as “Other”, performance for Asian and other patients will need watching.

None of that undermines the tool’s purpose, which mirrors a wider push to get more from data hospitals already hold, from ESMO’s study of routine blood tests and cancer risk to Sophont’s attempt to read gene activity from standard slides. It simply puts the burden of proof on what comes next.

US-only, with the real test still ahead

The clearance applies to the US market only, and none of the sources mention plans for the UK or Europe, where an FDA clearance does not apply. The release gives no pricing and no list of hospitals using the tool. The next markers to watch are a full peer-reviewed paper on the validation cohort, any head-to-head comparison with existing genomic tests, and, above all, prospective data showing whether a pre-surgery MRI score changes treatment decisions for the better.

Sources

1. SimBioSys press release: “SimBioSys Secures FDA Clearance for TumorSight Risk, Introducing the World’s First AI-Powered, MRI-Derived Breast Cancer Recurrence Risk Assessment” (15 October 2026)
2. FDA 510(k) Premarket Notification database: K260023, TumorSight Risk (decision date 25 September 2026)
3. FDA clearance letter, Indications for Use and 510(k) Summary, K260023 (PDF)
4. Goetz et al., “Prognostic value of an MRI-derived recurrence assay in women with hormone receptor–positive, HER2-negative breast cancer: A retrospective, multi-center analysis”, ASCO 2026 abstract 560, Journal of Clinical Oncology 44 (16_suppl)
5. NVIDIA on X, 26 October 2026

Related stories

• ESMO Study: Routine Health Data Flags Higher Risk of 20 Cancers
• Personalised mRNA Therapy Cuts Melanoma Recurrence or Death Risk 42%
• AI in Medicine: Google’s AMIE in The Lancet, Chai and GSK, and an FDA-Cleared Autism Assessment Aid
• Sophont’s Holon-S1 Aims to Predict Tumour Gene Activity From Routine Slides