Thakor Lab YouTube page
 
 
 
Brain-Computer Interfaces

& Neuroprostheses
EMG Pattern Recognition
Pattern recognition is an advanced control scheme for myoelectric prosthetic devices. Standard myoelectric control schemes utilize surface EMG electrodes located over specific muscle sites to control individual degrees of freedom. Typical myoelectric setups utilize dual site control with signals collected from two EMG electrodes, one placed over of the flexor muscles of the forearm and the other over the extensors. This simple binary control input scheme works well for a small number of activities, such as opening and closing a prosthetic hand, but quickly becomes unintuitive and unwieldy when trying to control multiple prosthetic components or access multiple grips with myoelectric triggers. To control multiple degrees of freedom with a dual site setup the user must use complicated muscle activation patterns, such as a short activation of the flex and extend muscle simultaneously (called a co-contraction) or double pulses, which are often difficult to master.
 
EMG pattern recognition
 
Pattern recognition utilizes an array of electrodes (typically 8) placed around the residual limb in non- specific locations to collect a larger general sample of EMG signals to establish characteristic patterns specific to a movement or grip. When pattern recognition is used for prosthesis control the amputee imagines moving their phantom limb to achieve a particular hand and/or arm position, which then correspond to muscle contractions detected by the EMG electrode array. As the surface EMG is sampled from all channels, certain time-based features are extracted from each signal to establish patterns that are used to calibrate the pattern recognition algorithm. After the calibration period, a linear discriminant analysis (LDA) is performed to establish classifiers that tie EMG patterns provided by the user to the specific movements and positions used during calibration. During use the prosthesis is sent control commands that correspond to output from the LDA. Meaning, when the amputee moves their phantom limb into a calibrated position, command signals are sent to the prosthesis to mimic that movement or position. In this way, pattern recognition restores intuitive control by moving the arm in the same way that the amputee imagines.
 
Go to Top
 
 
   

Copyright © 2008 JHU Neuroengineering & Biomedical Instrumentation Lab
All rights reserved.