Your skull contains no road signs. That is precisely the gap a team of London neurosurgeons asked an AI tool to help close back in May — and it’s the phrase the patient reached for once it was over.
The procedure took place at the National Hospital for Neurology and Neurosurgery, and the details have only now been made public. Rhys Hibbert, 48, arrived with a large tumor sitting on his pituitary gland and left with his vision unharmed.
“When I came round… I could see everything in the room clearly,” Hibbert said. Walking unaided had been impossible for him before the operation without glasses or walking sticks.
What the software was actually doing
As the surgeons worked the growth loose, the tool read live footage of the procedure and marked out critical brain structures — nerves, blood vessels. Put bluntly, its job was to stop them from straying somewhere they shouldn’t.
Surgeons fed a camera in through the nose to a position at the base of the skull. Working from that feed, the system followed the instruments in real time, worked out the likeliest locations of concealed vessels and nerves, and lit up the safe corridors for pulling the tumor out. Facial recognition, essentially, aimed at anatomy no one can lay eyes on.
Control of the procedure never left human hands. Still, the tool was critical in steering that work — a distinction every account of this case leans on heavily.
Why the pituitary is the hard one
Its neighbors are the arteries supplying blood to the brain and the optic nerves that govern sight. “Going a millimeter wrong can make a critical difference,” health officials said. Blindness, stroke and death are the ways it goes wrong.
The tumor Hibbert had was benign, which alters the stakes without altering the geometry. It lay at the base of the brain, out of sight, and reaching it meant carving an exact route that harmed nothing en route.
The training data is the whole story
Hundreds of surgical videos of pituitary tumor removals went into training the system. For each one, researchers traced around the vessels and nerves by hand so the model could form an accurate picture of their positions — painstaking, human, thoroughly unglamorous labor.
It’s that annotation work that gives the claim its weight. The AI system “has been exposed to a breadth of surgical examples that would take a surgeon many years to encounter,” said Sophia Bano, an associate professor in robotics and AI at University College London and the technical lead for the tool.
The part nobody has answered yet
And when it gets something wrong? No published answer to that appears here, which ought to nag at you slightly.
Medicine’s relationship with AI is already fraught, with more doctors turning to these tools for clinical notes and symptom lookups. The operating theater is a far steeper drop. Ahead of a pituitary tumor removal, the norm is for a surgeon to pore over brain scans until each patient’s particular anatomy is second nature.
The worry among experts is over-reliance on the software, particularly with a new cohort of medical students growing up alongside it. Skip years of pattern recognition with a tool, and you skip the years that build it.
Hibbert isn’t conflicted
“From a patient safety perspective, I can see absolutely the benefit. There are no road signs inside our head,” he said.
His verdict on the operation itself: “It’s given me my life back.” Sticks and glasses got him through the door. That’s the metric that matters, sample size of one or not.