EEG-based BCI Systems in Neuropsychiatric Diseases

181

The most common reference types are average reference, linked mastoids, ref-

erence electrode standardization technique (REST), or a specific EEG channel

such as Cz, or FCz [35].

Downsampling and segmentation are the other two steps in pre-processing.

Downsampling is a technique used in digital signal processing to reduce the

sampling rate by an integer factor which causes a decrease in the resolution

and file size and speeds up the further analysis. On the other hand, it can

lead to the loss of information as well. Segmentation, also called epoching, is

the division of the EEG data into blocks, often aligned with specific events or

tasks.

7.3.4

Feature Extraction Methods for EEG-based BCI

EEG reflects distinct characteristics of brain signals and feature extraction

is a signal processing stage in BCI, which captures the relevant information

to explain mental conditions. Various feature extraction approaches could be

utilized to develop EEG-based BCI systems.

If we endeavor to gain a more comprehensive understanding, it would

be better to start with the properties of EEG. An EEG signal is formed

by five crucial different natural frequencies (oscillatory activity), which are

traditionally divided into delta (1–3 Hz), theta (4–7 Hz), alpha (8–13 Hz), beta

(15–30 Hz), and gamma (above 30 Hz) frequency bands. These fundamental

EEG bands are believed to indicate distinct functional processes within the

brain. Brain functions are not only associated with oscillatory activities but

also with functional connections within the brain [36, 37]. Various parameters

can be used to characterize any oscillation and include important information.

These are the oscillation’s (i) frequency, (ii) amplitude, and (iii) phase [38].

All these parameters of EEG are represented in different domains that

help in analyzing and understanding the characteristics of brain activity [37].

One of them is time domain representation which examines the EEG signal

over time for understanding the temporal aspects of brain activity. The sec-

ond representation is the frequency domain that decomposes the signal into

various frequency bands, including delta, theta, alpha, beta, and gamma, and

provides insights into the dominant frequencies associated with different cog-

nitive states. Furthermore, the other informative representation is called the

time-frequency domain which combines both time and frequency information,

providing a more comprehensive view of EEG data [32].

In each domain, there exist diverse types of analyses [29]. In the EEG liter-

ature, the most popular time domain analysis is the Event-Related Potential

(ERP) which is a time-locked average of EEG activity in response to specific

stimuli or events. Various sensory, cognitive, or motor stimuli have the po-

tential to elicit ERPs. The second most common analysis of the time domain

is the digital filtering of EEG to determine the oscillatory activity in differ-

ent frequency bands. To gather statistical features, amplitude (magnitude of

EEG voltage fluctuations over time) and latency (the time delay between the