TL;DR

Queen Victoria Hospital in East Grinstead has been running Harrison.ai’s chest X-ray tool in its radiology department since July, using it to flag films carrying potentially significant findings and push them up the queue. It recognises as many as 124 findings and runs quietly inside the existing workflow. A qualified clinician still writes every report.

Prioritisation, not autonomy

The distinction the trust draws is worth taking seriously, because it defines the risk profile. The software does not issue a diagnosis; it surfaces what it has spotted to the radiologist or reporting clinician, who decides what follows. Nothing bypasses a human. What changes is the order films are looked at — which, when the constraint is reporting capacity rather than imaging capacity, is where the practical gain sits.

Dr Mohamed El-Belihy, the trust’s clinical lead for imaging and a consultant radiologist, said the tool helps the department prioritise films for review, with an experienced clinician still assessing each one before a report goes out. He thanked the imaging network covering Surrey, Sussex and Frimley for supporting the rollout.

Part of a national programme

This is a small trust adopting a tool already running in other NHS organisations and abroad, so its significance is as evidence of pace rather than novelty. Ministers committed £20m in June to putting AI chest X-ray analysis into every English trust by 2029, covering upwards of four million films a year.

Harrison.ai’s UK footprint is widening in parallel. Royal Surrey announced work in May on analysing images at the moment of capture, flagging suspected cancers for immediate attention. More consequentially, Oxford University Hospitals is running a study into whether AI can report chest X-rays on its own, testing model output against a year of imaging from Oxford and Manchester.

Looking forward

That Oxford study is the one to watch, because it tests the step this deployment deliberately avoids. Radiology reporting backlogs remain among the strongest business cases for clinical AI in the NHS, and triage tools like this one capture part of the benefit without moving the liability. Whether autonomous reporting can follow depends on evidence that does not yet exist.