Development and Performance Evaluation of a Neural Signal-based Assistive Computer Interface

This paper presents the development and performance evaluation of a human-computer interface that enables a limb-disabled person to access a computer via neural signals. For this purpose, electromyogram (EMG) signals were extracted from four muscles on the lower arm, and signal statistics (namely the mean and variance) were used for a filtering process. Six patterns were then classified through the application of a supervised multilayer neural network trained by a backpropagation algorithm. To extract the user's intentions, such as cursor movements and a clicking, the authors applied the neural network in the classification of the six patterns. In addition, an on-screen keyboard was developed so that letters of the Roman and Korean alphabets could be keyed into the computer. Finally, to confirm the appropriation of the developed computer interface, the authors applied Fitts' law in an experimental study to evaluate the performance of the computer interface. The experimental results show that the computer interface had an index of performance, or bandwidth, of 1.299. However, although the developed EMG-based human-computer interface had a lower index of performance than a mouse, it provides an alternative means of computer access for those with a disabled upper limb.