Machine learning technologies speed up the process of patient adaptation to bionic prostheses



    Bionic prostheses allow amputees to restore some of the functions of an absent hand, foot, or toe. Over time, bionics is becoming more perfect, respectively, the range of possibilities of prostheses is expanding. However, patients need a long adaptation period in order to learn how to handle their new limb, it is very difficult.

    Help, as it turned out , can machine learning technology. Namely - AI, where reinforcement training is used. The new method has already been tested in clinical trials. A person without a leg (or rather, parts of it below the knee) became a volunteer, which was replaced by an artificial limb.

    In a normal situation, a person has to train for hours to adapt to the use of a technologically advanced prosthesis. The new model works differently - in this case, it adapts to the owner, adapting to the peculiarities of his movements. This prosthesis helps a specialized algorithm that controls the artificial joint.

    The developers say that it is too early to talk about the introduction of the demonstrated technology into medicine, it was just a “demonstration performance”, which allows to judge about the possibilities of machine learning in bionic prosthetics. Scientists have proven that their technology has a high potential, so work continues on it.

    The results of the work were published in the authoritative edition, section IEEE Transactions on Cybernetics. Most likely, the new technology will be the beginning for the development of a number of machine-based methods for “learning” bionic prostheses. This will reduce the time and cost involved in preparing patients for work with conventional prostheses that are not equipped with an AI assistant. The algorithm allows you to adapt the prosthesis to almost any conditions, changes in its mode of operation is carried out automatically, literally on the go.

    The bionic knee has 12 different operating parameters that need to be customized. The AI ​​does this automatically, so instead of a few days of adaptation, we can talk about a couple of hours. In the process, the algorithm begins to “understand” how a person interacts with the electronic-mechanical system, after which there is a quick adjustment of the latter’s modes of operation.

    According torepresentative of the development team, Jenny Si, the body is quite difficult to adapt to an artificial object, embedded in the place of the missing part of the body. In this case, the reaction of the brain and the nervous system can be unpredictable. Machine learning reduces the number and level of problems. Of course, the ideal prostheses are not, but scientists are gradually working to bring together the possibilities and functions of organic and mechanical limbs.

    Artificial intelligence makes this convergence faster. The developers of AI have already proven that it can play better than a human in many games that were previously considered to be exclusively the prerogative of Homo Sapiens. True, there are difficulties. For example, machine learning technologies should be more effective than in the case of learning to play go, when AI sorted out millions of games, perfecting its art. The patient cannot spend hundreds of hours in the laboratory; machine learning technology should squeeze the maximum out of the few tens of minutes of interaction with the person she has.

    In addition, not all possible tests that may be useful for training were conducted in the laboratory. In a typical situation, a patient with a bionic prosthetic leg, without any support, may fall. The AI ​​in this case will receive valuable information that will allow you to avoid falling in the future. But the investigated did not study the fall due to security concerns.

    Whatever it was, but the first results were promising. The technology was able to identify a number of motion patterns, which were then used to adapt the prosthesis to its carrier, and this work was performed fairly quickly.

    In the near future, the developers plan to train the "bionic AI" to climb and descend the stairs. In addition, scientists are developing a wireless system - now the computer unit of the prosthesis is combined with a computer center with a cable, so such a system cannot be called mobile or autonomous. If the data systems will be transmitted through the air, it will greatly accelerate the process of data exchange. The patient himself will be able to move freely around the locations with different conditions, and not walk exclusively on the stand.

    Also popular now: