AI boost for low cost medical robot magnetic location system

May 18, 2020 //By Nick Flaherty
Researchers at UCSD have used machine learning to boost the accuracy of a $100 magnetic positioning system for a medical robot
Researchers at UCSD have used machine learning to boost the accuracy of a $100 magnetic positioning system for a medical robot

Researchers at the University of California San Diego in the US have developed a simple, affordable system to track the location of a flexible medical robot inside the human body and used machine learning to boost the accuracy. There is a video below.

The researchers embedded a magnet in the tip of a flexible medical robot that can be used in delicate places inside the body, such as arterial passages in the brain. "We worked with a growing robot, which is a robot made of a very thin nylon that we invert, almost like a sock, and pressurize with a fluid which causes the robot to grow," said researcher Connor Watson. Because the medical robot is soft and moves by growing, it has very little impact on its surroundings, making it ideal for use in medical settings.

The researchers then used existing magnet localization methods to develop a computer model that predicts the robot's location. The researchers know how strong the magnetic field should be around the magnet embedded in the robot. This relies on four sensors that are carefully spaced around the area where the robot operates to measure the magnetic field strength, and the field strength determines the position of the tip of the robot.

The whole system, including the robot, magnets and magnet localization setup, costs around $100.

"Continuum medical robots work really well in highly constrained environments inside the body," said Prof Tania Morimoto. "They're inherently safer and more compliant than rigid tools. But it becomes a lot harder to track their location and their shape inside the body. And so if we are able track them more easily that would be a great benefit both to patients and surgeons."

However Morimoto and Watson went a step further, training a neural network to learn the difference between what the sensors were reading and what the model said the sensors should be reading. As a result, they improved localization accuracy to


Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.