A robot surgeon watched medical videos to learn how to stitch human wounds

Elective surgeries will be cancelled for at least three months (Picture: Getty Images)

A robot "watched" videos of surgery techniques to help it learn how to stitch up human wounds.

In a research project, scientists fed 78 slowed-down films showing medical procedures into an artificial intelligence algorithm, connected to a pair of automated pincers programmed to work with a needle and thread.

The aim of the “imitation learning” study by scientists at the University of California, Berkeley, was to train surgery AI in a similar way to semi-autonomous cars, or the machine learning software assigned to scour complex legal papers.

After the videos had been analysed and the motions had been learned, the suture tasks were broken down for the AI into programmable actions, such as “needle insertion”, “needle extraction” and “needle hand-off”.

This video still shows how the robot's pincers set to work with needle and thread on a fake wound after learning suture (University of California, Berkeley)

Researchers said this enabled their deep learning system, called Motion2Vec, to instruct the robot hands to stitch a fake wound with about 85 per cent accuracy.

It could mean they work under supervision with surgeons in the future to undertake repetitive healthcare tasks, while freeing up human colleagues to concentrate on more complex procedures.

However, the current stitching system is not yet ready to unleash on real patients as it still has a 9.4mm error rate.

Study leader Dr Ajay Tanwani said: “A long-standing goal in artificial intelligence is to learn new skills by observing humans.

"Learning manipulation skills from video demonstrations by imitation can provide a scalable alternative.”