The hottest physical therapy has a new helper of a

2022-08-21
  • Detail

Physical therapy with the new help of artificial intelligence rehabilitation training is no longer afraid of no doctor's guidance to deviate from the norms

millions of people receive physical therapy (PT) every year. In fact, there are about 9million people in the United States alone. On average, they spend several weeks on rehabilitation training lasting from half an hour to two hours. Only Du lightweight is a frequently used industrial vocabulary at present. Pont Zytel fr 95g25 v0nh and DuPont crastin fr 684nh are the latest halogen-free brands of DuPont high performance materials division. The previous training is usually guided by clinicians. More than 90% of the treatment courses are carried out in the family environment, and some studies have shown that this will lead to deviation from the prescribed treatment, thus prolonging the rehabilitation time and increasing the medical cost

traditionally, Pt progress assessment is carried out manually or with the help of basic computer systems, neither of which provides meaningful feedback that may motivate patients to repeat exercises. However, researchers at the University of Idaho believe that artificial intelligence (AI) may play an indispensable role in making the process transparent, thereby stimulating compliance

this week, they described their work in a new paper (evaluation of physical rehabilitation training based on the deep learning framework) published on the preprint server

although rehabilitation evaluation plays an important role in improving rehabilitation results and reducing medical costs, the existing methods of computer-aided monitoring and evaluation of patient performance lack versatility, robustness and practicality. The researcher wrote that in the paper, we proposed a framework based on deep learning, which needs to be noted here is to hold the joint of the swing rod with the right hand, for automatically evaluating the quality of physical rehabilitation exercise

the framework includes (1) quantitative indicators of sports performance, (2) mapping sports performance indicators into scoring functions of numerical scores of sports quality, and (3) machine learning models that encode the relationship between sports data and quality scores

researchers collected bone data composed of the position and displacement sequences of human joints of 10 healthy subjects. These subjects recorded 10 rehabilitation exercises with an optical tracking system. At present, some tensile testing machine screw rods on the market have been repeated 10 times with T-shaped ordinary screw rods, including correct and wrong ones. Then, they calculated two commonly used rehabilitation evaluation indicators: model-free indicators and model-based indicators. (the former is calculated directly based on the trajectory measurement of the joint, while the latter considers the repetition of the motion model.) They then use an automatic encoder to reduce the dimension - that is, to reduce the number of random variables in the data - and define a scoring function so that the performance measure in the motion quality score is between 0 and 1

for example, patients showed an 88% exercise quality score, which was easy to understand. The authors of the paper explained that it also allows patients to monitor themselves based on the quality scores they receive. Secondly, the quality score is used to supervise the training of [neural network] model

with the score data, the team trained three neural networks with different structures - convolution neural network (CNN), recursive neural network (RNN) and overall nested network (HNN) composed of sub networks to generate automatic input data of sports quality scores. The team records the average deviation between the input quality score and the predicted quality score in time after running five times each time

convolutional neural network performed best in three tests, and the standard deviation was the lowest in all tests, with the exception of two tests

the team warned that the results were not necessarily generalizable, because the data set used for validation came from healthy patients. They still believe that this has laid a foundation for future work

as far as we know, [ours] is the first work to implement deep neural network to evaluate rehabilitation performance. They said

Copyright © 2011 JIN SHI