The FDA cleared the first MRI-derived breast recurrence risk software
The Food and Drug Administration granted 510(k) clearance to TumorSight Risk, AI-enabled software
that estimates breast cancer recurrence risk from breast MRI, pathology data and clinical factors.
SimBioSys, the developer, announced the clearance on October 15, 2026. It is the first prognostic
recurrence tool built on MRI imaging to receive that kind of clearance in the United States.
The indication is specific: adult women with hormone receptor positive (HR+), HER2-negative
breast cancer, stages I through IIIA. The software returns a 0 to 100 score that places the patient
in a low, intermediate or high risk category for recurrence at five and ten years.
What spatial modelling adds to pathology
The technical premise of TumorSight Risk is to extract information already present in the exam
that a conventional report does not quantify. According to the company, its spatial AI modelling
combines tumour characteristics with the context of surrounding tissue and clinical factors to
derive the score.
That matters because risk stratification in HR+/HER2- breast cancer is currently dominated by
tissue-based genomic signatures — assays that require a sample, shipping to a central laboratory,
significant cost and turnaround time. An alternative derived from an exam that has already been
performed changes the logistics: no additional procedure, no tissue shipment, no specialised
hardware. Its weight therefore comes less from standalone accuracy and more from where it sits in
the workflow.
The evidence submitted
Clearance was supported by a blinded, retrospective, multi-institutional validation study of
2,129 patients, with participating sites including Mayo Clinic, Emory and Stanford. In findings
presented at the American Society of Clinical Oncology (ASCO) conference earlier in 2026,
researchers reported that each 10-point increase in the score was associated with a 22% increase in
recurrence risk.
The submission also included a precision study in which 119 of 120 images processed successfully,
with 99.4% agreement across replicate runs, and a human factors study in which 30 physicians and
staff completed all critical tasks. The product builds on TumorSight Viz, the company’s previously
cleared surgical visualisation platform.
What 510(k) clearance means — and what it does not
The regulatory mechanism deserves explaining, because it is routinely misread. The 510(k) pathway
demonstrates substantial equivalence to a device already on the market; it is not an approval based
on a prospective clinical trial with a survival endpoint. What was assessed is analytical performance
and equivalence, not proven clinical benefit in real-world use.
In practice that defines the tool’s place: it is a prognostic aid to be weighed alongside other
clinicopathologic factors, not a replacement for the treatment decision. Anyone running breast
protocols already knows this distinction from other contexts — it is the same logic we discussed
when we showed that
detection and risk models are not interchangeable.
What it means for imaging department routine
For the radiologist, the most immediate practical effect is not on the report but on exam
quality. Once MRI feeds a prognostic model, acquisition standardisation stops being a preference and
becomes a requirement: consistent dynamic contrast-enhanced protocol, stable temporal resolution,
reproducible positioning and intact DICOM metadata. Variation between scanners and between protocols
is the main enemy of any imaging model, and departments already managing
breast MRI screening protocols for dense breasts
know how much work that takes.
There is also a workflow consequence: this class of tool pulls radiology closer to the oncology
decision, which requires a new conversation with breast surgery and medical oncology about who
orders it, who interprets it and where the score is recorded. That is not a software problem.
Cautions before generalising
Three limitations stand out. The first is the population: HR+/HER2- stage I to IIIA is a large but
specific slice, and nothing in the submission licenses extrapolation to triple-negative or
HER2-positive disease. The second is the retrospective nature of the validation — the sample is
robust, but it does not answer whether the score changes management or improves outcomes. The third
is the first-of-its-kind claim: “first MRI-derived prognostic software” is a company statement
repeated by trade press, not an independently verified fact.
It is also worth remembering how crowded breast risk prediction has become. Models that read risk
from mammography have already shown the ability to anticipate diagnoses over long horizons, as we
covered in
AI flags breast cancer risk a decade early,
and new breast imaging AI clearances have been frequent — the most recent being the
clearance of AI for breast ultrasound.
What distinguishes TumorSight Risk is not predicting risk: it is predicting recurrence after
diagnosis, a different clinical problem with a different audience and a different consequence.
What to watch over the coming months
The real test will be head-to-head comparison against established genomic signatures in
independent cohorts, not merely the association between score and recurrence. The question an
oncologist will ask is concrete: in how many cases does the imaging score agree with the genomic
assay and, when they disagree, which one is right? Until that answer exists, the tool is likely to
occupy a complementary role — useful for triage and for cases where tissue assays are unavailable,
but not yet a substitute.
