Systems, Man & Cybernetics - October 2017 - 17

Monitoring the neurophysiological activities involved with motion in a naturalistic environment
using vibration-sensitivity equipment such as functional magnetic resonance imaging [9] or positron emission tomography [10] represents a significant measurement challenge. The electroencephalogram (EEG) is currently the preferred device for noninvasive imaging of human brains in
BCIs as they perform tasks that involve natural movements in a real-world environment [11]. However, in conventional EEG devices, placing the electrodes on the scalp with a conductive gel or
paste is one of the common ways to measure the brain's electrical activity. Nevertheless, such electrodes, termed wet electrodes, require a time-consuming preparation process. Therefore, BCI systems are difficult to apply outside of -laboratory-scale experiments.
Regarding data quality, the measured brain signals are easily contaminated with artifacts originating from noncerebral origins. The amplitude of these artifacts, commonly generated by ocular
and muscle activities, can be quite large and may thus mask the cortical signals of interest, bias the
analysis and interpretation, and affect the performance of the BCI [12], [13]. Several blind source
separation (BSS) techniques have been proposed for signal preprocessing to remove such artifacts.
BSS is called blind because the axes of projection and, therefore, the sources are determined through
the application of an internal measure, which is done without the use of any prior knowledge of the
data structure. For example, independent component analysis [14], which produces the maximally
temporally independent signals available in the EEG recording, is a powerful tool for suppressing
artifacts. However, the iterative process of measuring the independence within multichannel
recordings is computationally intractable. In addition, manually excluding the independent components related to artifacts is still a time-consuming and offline process.
The Use of ERPs in BCI Data Analysis
The use of event-related potentials (ERPs) in BCIs is an effective method of
basic communication [15]. The classification accuracy of the ERP-based
BCIs is reliant on the number of trials used to analyze the data. The influence of infrequent noises can be eliminated from the recorded EEG
data by averaging the ERPs of a large number of repeated trials.
The signal-to-noise ratio (SNR) can be increased by averaging a
progressively larger number of trials. However, the large number of trials reduce the computational speed of the BCIs. An
efficient algorithm to speed up the convergence estimation
of ERPs is highly desirable for BCI applications.
For a BCI system, it is essential to efficiently extract
informative features from multichannel EEG signals.
Many useful approaches for analyzing the rhythmic pattern of EEG signals and extracting quantitative EEG features, such as the amplitude values of EEG signals, band
powers, power spectral density values, and auto regressive parameters, have been introduced to design BCIs. In
the neuroscience field, there has been increasing interest
in studies mapping the human brain connectivity in recent
years. The importance of this research topic was emphasized in the Human Connectome Project [16], which is devoted to investigating the knowledge of the human brain network.
The effective connectivity [17] is one of the most commonly used
measurements to identify the causal and directional relationship
between different brain regions. It is reasonable to assume that the
coupling between spatially separated brain areas can provide complementary information for BCIs.
A considerable amount of multimodality information is often simultaneously
used and recorded in a so-called hybrid BCI system [18]. These distinct information
sources provide various estimations of decision and action from multiaspect data, which
may help improve the system performance. Pelletier et al. [19] illustrate that there is a benefit
to using multimodal information according to the limitations of one modality, which can often be offset by the strengths of another. In addition, optical signals (i.e., near-infrared spectroscopy) are an

O c tob e r 2017

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

Table of Contents for the Digital Edition of Systems, Man & Cybernetics - October 2017