AI watches live brain surgery in world first at London hospital

Neurosurgeons in London have removed a brain tumour while an artificial intelligence system watched the operation as it happened and flagged what not to touch, a world first for live AI-assisted surgery.

The patient, 48-year-old Rhys Hibbert from Bedfordshire, was slowly going blind from an 11mm growth on his pituitary gland. He came out of theatre with his sight intact and, within a week, was walking without his sticks. The operation was carried out in May at the National Hospital for Neurology and Neurosurgery, part of University College London Hospitals, and made public only on 27 August, once Hibbert had recovered. Prof Hani Marcus performed it alongside surgical resident Danyal Khan, with the AI running on a second screen throughout. The clinical trial is funded by the National Institute for Health and Care Research and Google.

The pituitary tumour was found by accident after a seizure on a lunchtime walk.

Until December 2024, Hibbert was a fit customer services manager and an ambulance volunteer who walked two miles every lunchtime. On one of those walks he lost consciousness and started fitting in the road. Scans found a non-cancerous tumour on his pituitary gland, an organ roughly the size of a marble. It was left alone at first, then began pressing on his optic nerves. He lost his lower field of vision and started using walking sticks after tripping once too often.

The AI read the live surgical video feed instead of pre-operation brain scans.

Surgeons reach the pituitary through the nose using an endoscope, a camera on a thin stick. The gland sits millimetres from the carotid arteries and the optic nerves, and anatomy differs from person to person. The average UK surgeon does 10 to 20 of these operations a year, working from scans studied beforehand. Even then, there is a 25 to 50 per cent chance of leaving tumour behind and a 0.5 to 2 per cent risk of injuring a major blood vessel. This system watched the feed as it streamed, colour-coding the vessels and nerves it recognised and marking where it was safest to cut. Think face unlock on a phone rather than a robot surgeon. It was built at the UCL Hawkes Institute, runs on Nvidia's Clara IGX platform, and was trained on hundreds of annotated pituitary operations. "There are no road signs inside our head," Hibbert said.

UCL now wants a bigger AI surgery trial and a tool surgeons can argue with.

Marcus has been clear that the surgeons kept full control. The longer-term ambition is closer to a ChatGPT for the operating theatre, a second opinion a surgeon can consult or overrule, and a larger trial is being planned. Eight weeks on, Hibbert's sight is still sharpening. "It feels like I've got a 360-degree panoramic view of everything around me," he said.