Publication: Visual stimuli-based dynamic commands with single channel electroencephalography for reactive brain-computer interface applications
Loading...
Date
2023-05-01
Authors
Teo Jia Hui
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The brain-computer interface (BCI) technology is widely used to control robotic devices using electroencephalogram (EEG) signals. Normal eye blinks are usually treated as artifacts within the recorded EEG signals but voluntary eye blinks which will cause distinct signal deflections can be introduced into BCI applications as control commands. This study designed a new reactive BCI paradigm using single-channel EEG device which combined visual stimuli and voluntary eye blinks to work on motorized actuators with different speed profiles. This proposal consisted of an EEG decoder that applied on machine learning methods to improve the accuracy of the system. Multilayer perceptron (MLP) neural network and Gaussian Process model (GPM) had been proposed to minimize the mismatches between required and actual transmitted commands. Results from thirty subjects showed that the GPM could achieve the highest 90% accuracy and lowest 1.51cm/s mean absolute error whereas MLP showed slightly better performance (86% accuracy and 1.76cm/s mean absolute error) than other comparing methods. The implementation of Hanning filter showed improvement to minimize unwanted errors in the system. Female-based model could perform slightly better than generic model on female-only test data set whereas male-based model performed similarly as generic model on male-only test data set. While other learning models were susceptible to different types of stimuli, the proposed GPM could perform consistently to demonstrate its high generalization and make it a more suitable option over MLP and other methods on such BCI system.