<p>About 1.7% of the US population live with paralysis or reduced muscle function. Functional electrical stimulation (FES) has been widely used in the biomedical field to increase the functionality of atrophied muscles. The goal of this research was to design, build, and test a neural prosthesis that uses functional electrical stimulation to improve gait in people with muscle weakness. The overall objective of this project was to quantify the gait tracking performance of the third generation prosthesis, and to develop the next generation model by implementing an artificial neural network that automatically controlled the electrical muscle stimulator. The fourth generation prosthesis was programmed to use sensor feedback from three inertial measurement units (IMUs) and four force sensitive resistors (FSRs) to predict the correct stimulation time. The IMUs were used to track leg movement during gait with each IMU containing a triaxial accelerometer and a triaxial gyroscope.&nbsp; Data from the two sensors were combined using a complementary filter that indicated the amount of reliance given to each source of input data.&nbsp; Accelerometers have the advantage of providing a direct measure of joint angle, but they are susceptible to body segment accelerations and noise caused by impact.&nbsp; Gyroscopes are less susceptible to noise but determine the joint angle by integrating angular accelerations over time, so they are subject to sensor and processor delay.&nbsp; FSRs were used to track the forces exerted by the foot at different stages of the gait cycle. These data are quite precise, but the sensor acts more like a switch indicating the start or end of pressure application.&nbsp; Data from these sensors can lead to erroneous results when gait speeds change because only initiation and completion is recorded.&nbsp; Combining data from all types of sensors can overcome the limitations of any individual sensor type.&nbsp; In the neural prosthesis there are many more input parameters than output parameters. To overcome this redundancy a neural network was used to aggregate the input parameters to make a logical decision that incorporated all.&nbsp; The neural network had seven input nodes, one for each of the sensors. There were thirteen hidden layer nodes based on the practice of using one less than two times the number of input nodes.&nbsp; There was one output node to which a threshold was applied to effect binary classification of the input parameters, and activation of the electrical stimulus. Results showed that it was possible to program a highly accurate neural network from the received data of the sensors with a prediction errors of 1.0%, 1.0% and 1.1% for three separate evaluations of the 2767 points classified during the gait cycle.&nbsp; In conclusion, the neural network was found to effectively determine when the stimulator should be activated and when it should be deactivated.&nbsp; Further research is being conducted to improve the performance of the IMUs to reduce the noise and time delay in the input data and to evaluate the performance of the device when used by people with muscle weakness.&nbsp;</p>

Neural Network Controlled Stimulation of a Neural Prosthesis

Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2021/volume/85543

ASME 2021 International Mechanical Engineering Congress and Exposition (IMECE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.