AI helps people with paralysis to move again
Procedure analyzes residual nerve activity and is now available for researchers to use and develop further free of charge
A new procedure should help people with nerve damage or amputations to regain at least some of their motor abilities. An AI algorithm assesses and interprets the residual nerve activity in the affected part of the body. Often, all it needs is a few minutes of training before a patient is able to move the fingers of a virtual hand or a prosthesis on command. The method developed at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) is now available for researchers to use and develop further free of charge. An article recently published in the journal Science Advances* indicates the potential of this open source solution.
Prosthetic hands or legs controlled by nerve impulses have been available for several years now. However, it often takes a while until people feel confident in using them. The learning process is considerably shorter in the method developed by FAU. It is based on the premise that before their accident or illness, the patients were generally able to move normally. They have internalized the necessary motor commands over many years. The approach called “MyoGestic” aims to make use of this advantage.
Virtual hand demonstrates the movement
Open source software MyoGestic
The core of the invention is a cuff containing 32 electrodes. In the case of a hand amputation, it can be pulled over the stump and record nerve activity. This activity pattern is interpreted by software and converted into movement. “During the training phase we work with two virtual hands displayed on a computer screen,” explains Raul Sîmpetru. The early career researcher is working on his doctoral degree at the Professorship of Neuromuscular Physiology and Neural Interfacing directed by Prof. Dr. Alessandro Del Vecchio. Together with his PhD colleague Dominik Braun, he is the main author of the current study. “One of the hands shows a movement that the patient should copy. The other hand then shows the result based on the interpretation by the AI algorithm.”
Ideally, both hands should move in an identical manner. 32 electrodes are not sufficient to completely record the complex activity pattern of motor nerves. However, if the test person imitates the displayed movement several times after another, the artificial intelligence learns to interpret this incomplete data correctly. “For example, we had a patient with a hand amputation who was able to bend and stretch each individual finger however she wanted using the artificial hand after just five minutes,” says Sîmpetru.
Even people with paraplegia can benefit
The method even works for people with paraplegia. Often, not all nerve fibers are severed. This means that the brain can still send electric signals to the muscles in spite of the injured spine. However, the signals are too weak to trigger the required movement. Now, AI can learn to interpret these weak electrical impulses correctly. “It is more difficult if the injury happened so long ago that patients can no longer remember which commands they have to send from their brain to the relevant part of the body,” explains the FAU researcher. “Or if it is simply not possible to reproduce certain signals correctly.”
In these cases it is possible to perform what is known as “remapping”. For example, the patient may no longer be able to replicate the command “bend the index finger”, but is able to replicate the (less frequently used) command “bend the little finger”. This can then be used to control the index finger. However, the patient must learn that the relevant command now no longer moves the little finger.
Low costs
In the study, the researchers were able to demonstrate that their procedure is highly effective, and it is also extremely low cost. All that is needed to train the AI is a standard cuff fitted with electrodes that does not have to be tailored specifically to the patient. “Of course, our method does not only work for controlling virtual hands, it can also be used to move a computer mouse or a prosthesis,” underlines Prof. Dr. Del Vecchio. Unlike most standard prosthetics, finely coordinated movements are also possible, for example to grasp a tomato without squashing it.
The method is open source, that means it can be used and developed further without having to pay any license fees. The working group has placed their test protocols and the AI algorithm they used online for anyone to access. The aim is that this will allow the procedure to be optimized further and made ready for practical application, allowing paraplegics and people who have suffered an amputation or a stroke to benefit.
*DOI: 10.1126/sciadv.ads9150
Further information:
Prof. Dr. Alessandro Del Vecchio
Professorship for Neuromuscular Physiology and Neural Interfacing
Phone: + 49 9131 85 70940
alessandro.del.vecchio@fau.de