Santé · AI-guided surgeryL’IA a observé en direct des chirurgiens londoniens retirer une tumeur cérébraleRédigé 7 septembre 2026 à 21:22 UTCRegarderTranscription0:00An AI watched a brain operation live and flagged where not to cut.0:04An 11 millimeter tumor was pressing on optic nerves and threatening blindness, and0:08pituitary surgery normally carries a 25 to 50 percent risk of incomplete removal and0:13a 0.5 to 2 percent risk of major vessel injury.0:17Can the guidance that saves sight for one patient be shown to work across hospitals,0:20tumors, and surgeons?0:22The mass was small and benign, but in the most dangerous place, 11 millimeters on0:26the pituitary, compressing the optic apparatus where millimeters decide vision and0:31life.0:32Removing it meant navigating hidden vessels and nerves barely distinguishable under0:36blood and tissue.0:37The operation happened in May 2026 at the National Hospital for Neurology and0:41Neurosurgery in London on a 48-year-old man from Bedfordshire, verified as Rhys0:47Hibbert, noted in the initial source record as Rhys Hibbett.0:50The case was first reported on 26 August at 2307 UTC, described in follow-on0:56coverage on 27 August as the first reported live AI-assisted brain tumor removal.1:01The system was built over almost 10 years by surgeons and engineers at University1:05College London, funded by the National Institute for Health and Care Research, with1:09additional support from Google.1:11Technical work was led by Sophia Banno, surgical work by Professor Hani Marcus and1:15neurosurgeon Daniel Kahn.1:17The change is from memory and microscope alone to machine vision alongside the1:21surgeon.1:22The AI analyzed the live endoscope feed in real time, scanning the field,1:26identifying blood vessels and nerves, and advising on avoidance while the surgical1:30team kept full control.1:31That live moment followed a long lab path.1:34Hundreds of videos trained the model, more than 200 recordings early on, then 6401:39images from 64 videos in a 2024 offline study, building what the team described as1:4410 months of video experience, the equivalent of 10 years for a trainee before1:48simulated studies and safety checks allowed a move from retrospective analysis to1:52real-time advice in theater.1:53The tool has no publicly disclosed product name.1:56Coverage has used the descriptive phrase surgical chat GPT.2:00In theater, it acted as an extra set of eyes for anatomy that is hardest to2:04distinguish when blood and tissue obscure the field.2:07In this single case, the aims aligned, complete excision, vision preserved, energy2:11returned, back to work within weeks, and no reported complications past eight weeks.2:16But a small, benign tumor does not answer how the system handles larger malignant2:20tumors, different locations, bleeding, poor image quality, or different surgeons in2:24hospitals.2:25No specific regulatory approval from the medicines regulator, NICE, or NHS England2:30has been identified.2:31The operation ran within an NIHR-funded clinical trial, and full details of the2:36simulations, surveys, and safety interlocks remain summarized, not released.2:40The wider risk the team is trying to reduce is well quantified.2:44Standard pituitary surgery leaves a one in four to one in two chance of incomplete2:48removal and a 0.5 to 2% chance of major vessel injury, with vision and life directly2:54at stake if optic nerves or nearby arteries are damaged.2:57The most consequential question now is scale.3:00The team is working toward larger trials to test whether live guidance improves3:04safety, completeness of removal, and vision outcomes across many patients, and3:08whether a decade of video training can become routine assistance in brain tumor3:11surgery or remains a one-patient demonstration.Lire l’articleChronologie de l’histoireHeyDexLe format court ↗
