Qureight raised $20M (led by Molten Ventures) to build a general-purpose 3D foundation model of the chest, then adapt it to asthma, lung cancer and more in 1-2 months instead of a year. It works inside drug trials — cutting imaging decisions from 2 weeks to 48 hours — and offers AI ‘synthetic control arms’, the promising, regulator-dependent claim to watch. Clients include AstraZeneca and Bristol Myers Squibb.
Most medical-AI imaging startups pick a disease, build a model, sell it. Cambridge’s Qureight raised $20 million to do something structurally different: build one general-purpose 3D ‘foundation model’ of the human chest, then spin disease-specific models out of it in one to two months instead of a year. It’s the foundation-model idea, applied to a lung.
Most medical-AI imaging startups do the same thing: pick a disease, build a model to detect it, sell it. A Cambridge company called Qureight has raised $20 million to do something structurally different — build one general-purpose 3D “foundation model” of the human chest, and then spin disease-specific models out of it in one to two months instead of the year it takes from scratch. It’s the foundation-model idea, borrowed from language AI, applied to a lung.
What they built
Qureight’s core is a 3D model of the chest that creates detailed maps of airways and blood vessels and tracks how disease progresses over time. From that base, the company says it can build models for asthma, lung cancer, pulmonary hypertension and bronchiectasis quickly, building on its existing lung-fibrosis work. The $20M Series B was led by Molten Ventures, taking total funding past €27M. The founders are physicians — CEO Muhunthan Thillai is a chest physician — and the technical bench includes a former HP platform head and a machine-learning professor from Imperial College London.
Why the foundation-model approach matters here
The insight is the same one that transformed language AI. Instead of training a narrow model per task, you train one large general model of the domain, then adapt it cheaply to specific tasks. Qureight is betting that a chest is a chest — the anatomy, the airways, the vasculature are shared across diseases — so a single strong 3D model of it can be specialised to many conditions far faster than building each from zero. If that holds, the economics of medical imaging AI change: the expensive part (the foundational model) is built once, and each new disease is a cheap adaptation rather than a fresh year-long project.
The other clever choice is where Qureight sits. Rather than selling diagnostic software to hospitals — the crowded lane where rivals like Oxford’s Brainomix compete — it operates inside clinical trials themselves. That’s a different and arguably better business: pharma companies running drug trials need to measure whether a treatment is working, and imaging is how you see it. Qureight already counts AstraZeneca and Bristol Myers Squibb as clients.
The genuinely interesting bit: synthetic control arms
The most striking claim is speed and a technique called “synthetic control arms.” On speed: the company says it cuts imaging-based decisions in a trial from around two weeks to 48 hours. On synthetic control arms: it offers AI-generated substitutes for the real control group in a trial — the patients who get a placebo so you have something to compare the treated group against.
That is a big idea with a big asterisk. Control groups are expensive, slow to recruit, and ethically fraught (someone has to get the placebo). If an AI model of disease progression could reliably stand in for some of that comparison, trials would get faster and cheaper, and fewer patients would need to be placed on placebo. But “reliably” is carrying enormous weight. A synthetic control arm is only as trustworthy as the model’s ability to predict how untreated patients would have progressed — and if it’s wrong, you get a false read on whether a drug works, which is the single most consequential error in medicine. Regulators will scrutinise this heavily, and rightly.
The honest boundaries
Two cautions. First, “build a disease model in one to two months” is a company claim about its own pipeline, not an independently verified benchmark — the foundation-model approach is sound in principle, but the quality of each spun-out model still has to be proven per disease. Second, synthetic control arms are promising and unsettled; they will live or die on regulatory acceptance and on evidence that the model’s predicted progression matches reality closely enough to base drug approvals on. This is frontier methodology, not settled practice.
The read
Qureight is a smart application of the foundation-model pattern to medical imaging, and its position inside clinical trials rather than hospital software is a shrewder business than the crowded diagnostics market. The speed gains are credible and valuable. The synthetic-control-arm idea is the one to watch — potentially transformative for the cost and ethics of drug trials, and potentially dangerous if the models aren’t as good as they need to be. That’s the right tension for a company at this frontier: a real efficiency breakthrough wrapped around a claim that regulators, not press releases, will have to validate.
Reporting on a venture funding announcement as covered on 30 July 2026, based on reporting by The Next Web. Capabilities and speed figures are company-stated and not independently verified; synthetic control arms are an emerging, regulator-dependent methodology. Not medical or investment advice.
