A P300-based brain–computer interface aimed at operating electronic devices at home for severely disabled people

Abstract

The present study aims at developing and assessing an assistive tool for operating electronic devices at home by means of a P300-based brain-computer interface (BCI). Fifteen severely impaired subjects participated in the study. The developed tool allows users to interact with their usual environment fulfilling their main needs. It allows for navigation through ten menus and to manage up to 113 control commands from eight electronic devices. Ten out of the fifteen subjects were able to operate the proposed tool with accuracy above 77 %. Eight out of them reached accuracies higher than 95 %. Moreover, bitrates up to 20.1 bit/min were achieved. The novelty of this study lies in the use of an environment control application in a real scenario: real devices managed by potential BCI end-users. Although impaired users might not be able to set up this system without aid of others, this study takes a significant step to evaluate the degree to which such populations could eventually operate a stand-alone system. Our results suggest that neither the type nor the degree of disability is a relevant issue to suitably operate a P300-based BCI. Hence, it could be useful to assist disabled people at home improving their personal autonomy.

An Asynchronous P300-Based Brain-Computer Interface Web Browser for Severely Disabled People

“…Those signals were recorded using 8 active electrodes placed on Fz, Cz, Pz, P3, P4, PO7, PO8 and Oz, according to the International 10-20 System [16] distribution, using a FPz electrode as a ground and referencing the system to the earlobe. This distribution is commonly used to record P300 potentials, mainly generated over the parietal cortex [17]- [19]. Electrodes were connected to a g.USBamp amplifier (g.Tec, Guger Technologies, Austria) with a sampling frequency of 256 Hz.…”

“…When the target's row or column are illuminated, a P300 potential is generated, used for figuring out what the desired command is [3], [11]- [13], [15]. More specifically, each random stimulus lasts for 62.5 ms and then the screen remains unvarying for 125-250 ms [19].…”

“…Because of the high sampling rate of the recordings relative to the low frequency of the P300 response, a dimensionality reduction for removing redundant features is beneficial for the real-time classification [23]. In this case, a subsampling of 20 Hz over an 800 ms window from the stimulus onset is applied [13], [19]. Therefore, each stimulus is considered a vector f of 128 features: 16 samples (20 Hz · 0.8 s)  8 channels.…”

“…Therefore, significant differences (p-value < 0.05) between models with and without the current evaluated feature are assessed to determine whether it provides discriminative information to the model or not. In this case, the discriminant function was restricted to contain a maximum of 60 features [13], [17]- [19], [23], and the selection/elimination criteria was set up as…”

“…, commonly applied in P300-based BCI studies [17], [19], [23], [25], [26]. Once the optimum weight vector w is computed, under the assumption that noise is normally distributed with equal covariances for both classes, the output of SWLDA is a log-likelihood ratio to belong to the positive class (i.e., presence of P300) [27].…”