There is a quiet room in Chestnut Hill, Massachusetts where a woman lies still for a mammogram she has had a dozen times before. Nothing about the machine looks different. What is different is what happens after. The image now passes through software that does not look for a tumor. It looks for the future.
That software is Clairity Breast, and the company behind it is a 16-person outfit in Boston that has spent the last several years arguing with one of medicine's oldest habits: waiting. For half a century, the mammogram has been a rear-view mirror - a way to catch cancer that already exists. Clairity built a windshield. Same picture, opposite direction.
It is a deceptively small idea wrapped around a hard one. The small idea: you already took the photo, so ask it a better question. The hard one: teaching a neural network to read texture and pattern in breast tissue that no human eye reliably catches, and convincing the U.S. Food and Drug Administration that the answer is trustworthy enough to put in front of a patient. The first part is software. The second part is why most companies like this never ship.
Same image. New question. That is the whole company in four words.
- the Clairity premise, paraphrasedA diagnosis that arrives too late, to the wrong people
Breast cancer screening has a blind spot, and it is not the technology. It is the math we wrap around it. Most risk calculators in clinical use - the Tyrer-Cuzick model, the Gail model - lean heavily on family history, age, and reproductive factors. They were largely developed and validated on white, European women. For everyone else, they are a rough guess wearing a lab coat.
Dr. Elizabeth Mittendorf, chief of breast surgery at Beth Israel Deaconess, put the problem plainly: the older calculators "were developed and validated in white, European women, and so they don't really apply more broadly." Meanwhile, the disease has been showing up more often in younger women and in women with no family history at all - exactly the people the standard tools are worst at seeing.
So you have a screening system that is excellent at finding cancer that is already there, attached to a risk system that is mediocre at predicting who will get it and unfair about who it predicts well. The gap between those two things is measured in late-stage diagnoses. That gap is the entire reason Clairity exists.
Some of the other risk calculators we currently use were developed and validated in white, European women, and so they don't really apply more broadly.
- Dr. Elizabeth Mittendorf, Beth Israel Deaconess Medical CenterA radiologist who got tired of being surprised
Dr. Connie Lehman spent more than two decades as a breast imaging radiologist at Massachusetts General Hospital. She holds an MD and a PhD from Yale, an honorary MD from Harvard, and somewhere north of 300 peer-reviewed papers with her name on them. She is, by most measures, exactly the person you would want reading your mammogram.
And she kept getting surprised. Patient after patient, younger than the textbook said, with none of the risk factors the calculators asked about, sitting across from her with a new diagnosis. Lehman's bet was that the information to predict those cancers was already in the images she had been reading her whole career - she just could not see it, and neither could anyone else, because human eyes were never built for it. A deep-learning model might be.
It is a tidy irony that the breakthrough required a radiologist to admit the limits of radiologists. Lehman built the model with collaborators tied to MIT and Mass General Brigham, trained it on hundreds of thousands of real mammograms with known outcomes, and then did the unglamorous thing: she founded a company to get it through the FDA and into clinics, rather than leaving it as a celebrated paper nobody could use.
With Clairity Breast, we can now use the same images to predict who may be at risk.
- Dr. Connie Lehman, Founder, ClairityOne score, from a picture you already have
Clairity Breast takes a standard 2D screening mammogram and returns a single number: an estimate of the woman's risk of developing breast cancer over the next five years. No extra scan. No blood draw. No genetic test. No additional radiation. The input is an image already sitting in the medical record.
Underneath, it is a deep convolutional neural network trained on more than 400,000 mammograms, each paired with five years of follow-up so the outcomes were actually known. The model was then validated against roughly 77,000 additional mammograms across a deliberately diverse population - the part the old calculators skipped - and outperformed conventional risk assessment. The point of pixel-level analysis is that it reads the tissue itself, not a checklist of demographics.
How a paper became a product
Clairity milestone reel · 2020 → 2026
There is a roadmap behind the single product, too. Clairity Breast 3D extends the model to tomosynthesis exams. Clairity Heart points the same deep-learning framework at cardiovascular risk - because the mammogram, it turns out, may quietly hold clues about more than one disease. The thesis scales: routine images are an underused dataset, and most of them are already taken.
The ProofThe numbers, and the people willing to stake a clinic on them
Validation studies are easy to wave at and hard to read. Here is the version that matters: a model trained and tested on hundreds of thousands of images, with known five-year outcomes, that beats the calculators currently in use - and was specifically checked to work across populations the old tools failed.
The evidence base, by the numbers
Why a hospital was willing to deploy it
Bars scaled to the training set, which dwarfs everything else - which is exactly the point. Source: company and press reporting, 2025-2026.
Then there are the people who put their names on it. The Breast Cancer Research Foundation is not just an investor; it is a research partner whose whole creed - "prevention is the ultimate cure" - the product is built to serve. ACE Global Equity and Santé Ventures led the $43 million round in November 2025, with Masimo founder Joe Kiani joining the board. And in February 2026, Beth Israel Deaconess became the first health system to actually hand patients a Clairity score, with its chief of breast surgery running the rollout. Investors can be wrong. A breast surgery chief deploying it on her own patients is a harder thing to fake.
Predictive imaging is the next frontier in oncology.
- ACE Global Equity, lead Series B investorMove the moment of truth earlier
Clairity states its mission without much hedging: expand access to early detection and prevention by using AI to predict breast cancer risk from images women already have. Read between the lines and it is an argument about timing and fairness. Earlier, because a five-year warning is room to act. Fairer, because a model that reads tissue instead of a demographic checklist does not quietly fail the women who were never in the original studies.
It helps that the founder's incentives are unusually aligned with the patient's. Lehman did not arrive at prevention through a market analysis. She arrived at it through twenty years of telling people, too late, that they had cancer. The company is the attempt to have a different conversation - one that starts five years sooner.
Field notes / things that amuse and inform
- It uses the mammogram you already took - no extra scan, no needle, no spit-in-a-tube.
- The founder admitted the limits of expert radiologists to build a tool for radiologists.
- Sixteen people produced a first-in-category FDA authorization. Most "AI health" press releases have larger comms teams.
- The same model is being pointed at the heart next - the mammogram may be a more talkative image than anyone assumed.
- Connie Lehman: MD and PhD from Yale, honorary MD from Harvard, 300+ papers, and now a TIME100 Health name.
If a picture can predict, what else is in the file?
The interesting thing about Clairity is not the breast cancer model. It is the precedent. If a routine mammogram contains a reliable five-year cancer signal, the obvious next question is what other forecasts are hiding in images hospitals already take by the millions. Clairity Heart is the company betting that the answer is "a lot." Every CT, every X-ray, every scan becomes a candidate dataset for prediction rather than just detection.
That is a bigger and more uncomfortable idea than a single product. It reframes the medical image from a record of what is wrong now into a forecast of what could go wrong later. Clairity is not the only company chasing radiology AI - iCAD, Lunit, Volpara and others crowd the field - but it got to the risk-prediction frontier with an FDA stamp first, which in regulated medicine is most of the moat.
Back in that quiet room in Chestnut Hill, the machine still looks the same. The woman gets up, gets dressed, goes home. But now there is a number waiting for her - and her doctor - that did not exist a year ago. It does not promise her anything. It just refuses to let the future stay a surprise. For a disease that has specialized in arriving unannounced, that is the part that changes everything.