King's College Hospital in London is now using an artificial intelligence system during live cardiac procedures. The software, Ultreon 3.0, operates during angioplasty and produces high-definition imaging from inside the coronary arteries. It identifies arterial plaque considered at risk of rupture, then assists cardiologists in positioning stents with a degree of accuracy that manual assessment cannot reliably replicate.

The system does not replace the cardiologist. It processes imaging data in real time, flagging areas of concern and calculating blood flow metrics at a speed that would take a specialist considerably longer to produce unaided. For a procedure where the margin between a successful outcome and a secondary event can be narrow, those seconds carry clinical weight.

Planners at King's are working towards treating around 500 patients per year through the London site before the programme expands nationally. The underlying argument is straightforward: automate the most technically demanding element of visual assessment, reduce the variability that comes with human fatigue or experience gaps, and improve long-term outcomes. Whether that argument holds across a broader rollout will depend on factors that no algorithm can yet account for.

Chief among them is the state of the workforce into which these tools are being introduced. A recent survey of NHS clinical staff found that nearly 90 per cent consider the health service to be in a weak condition. The figure is not incidental. Advanced diagnostic software requires trained operators, institutional familiarity, and the kind of procedural stability that is difficult to maintain when rotas are incomplete and experienced staff are leaving at rates that outpace recruitment. Introducing precision technology into an environment defined by shortage does not automatically resolve the underlying pressure; in some cases it redistributes it.

The fiscal context complicates matters further. The NHS enters 2026 under significant financial strain. Record levels of A&E attendance, persistent inflationary pressures on both procurement and staffing, and an increase in government borrowing are all ongoing issues. The service has met some elective waiting time targets, but the metrics that make headlines are not always the ones that describe operational reality for a district general hospital in the North of England or a trust running at persistent deficit. Rolling out technology that requires licensing agreements, technical maintenance contracts, and staff training across dozens of trusts is an exercise in health economics as much as clinical ambition.

That tension sits at the centre of the government's ten-year health plan, which positions digital integration as a structural priority rather than a supplementary feature. The NHS App is being expanded to allow patients greater control over self-management and appointment data. AI applications are being developed across radiology, pathology, and now cardiac intervention. Officials describe this as a move towards what some in the sector are calling "Physical AI," a model in which computational systems are embedded directly into clinical workflows rather than used as back-office analytics tools.

The cardiac AI programme is, in this framing, a proof of concept. If Ultreon 3.0 delivers measurable improvements in patient outcomes at King's, the data it generates will inform both the clinical case and the commissioning decisions for broader adoption. The NHS has form in piloting technology that stalls at scale. It also has examples of digital tools that moved from pilot to standard care with reasonable speed when the evidence and the institutional will aligned.

What distinguishes the current moment is the scale of expectation. Policymakers, clinicians, and technology companies are all invested in demonstrating that AI can do something meaningful inside the health service rather than adjacent to it. The cardiac catheterisation lab is as concrete a test of that proposition as any. The imaging is either more accurate, the outcomes are either better, or they are not.

For patients awaiting procedures at King's or at the trusts that may follow, the question is less abstract. They are not evaluating a ten-year plan. They are waiting for a diagnosis, and the technology now helping to produce it was, until recently, the subject of conference presentations rather than clinical protocol.

That shift, from aspiration to application, is what makes this particular development worth examining with some care.

‍