Refined Myoelectric Control in Below-Elbow Amputees Using Artificial Neural Networks and a Data Glove

Abstract

To develop a system for refined motor control of artificial hands based on multiple electromyographic (EMG) recordings, allowing multiple patterns of hand movements. Five subjects with traumatic below-elbow amputations and 1 subject with a congenital below- elbow failure of formation performed 10 imaginary movements with their phantom hand while surface electrodes recorded the EMG data. In a training phase a data glove with 18 degrees of freedom was used for positional recording of movements in the contralateral healthy hand. These movements were performed at the same time as the imaginary movements in the phantom hand. An artificial neural network (ANN) then could be trained to associate the specific EMG patterns recorded from the amputation stump with the analogous specific hand movements synchronously performed in the healthy hand. The ability of the ANN to predict the 10 imaginary movements offline, when they were reflected in a virtual computer hand, was assessed and calculated. After the ANN was trained the subjects were able to perform and control 10 hand movements in the virtual computer hand. The subjects showed a median performance of 5 types of movement with a high correlation with the movement pattern of the data glove. The subjects seemed to relearn to execute motor commands rapidly that had been learned before the accident, independent of how old the injury was. The subject with congenital below-elbow failure of formation was able to perform and control several hand movements in the computer hand that cannot be performed in a myoelectric prosthesis (eg, opposition of the thumb). With the combined use of an ANN and a data glove, acting in concert in a training phase, amputees rapidly can learn to execute several imaginary movements in a virtual computerized hand, this opens promising possibilities for motor control of future hand prostheses.