Page 30 - BH-2-2

Brain & Heart Predictive modeling using electroencephalogram

module for analysis. It gathers information and saves it in 3.4. User application
a file. The user application is in charge of regulating the external
3.2.1. Feature extraction processes that are affected by the signals sent from the
signal processing module. Keystroke filter and connector
Several crucial aspects of the EEG signals have been studied filter stand out as particularly useful built-in application
to analyze the motor imagery signals, with each capability modules because they allow the control signals to be
proving effective. Band powers, in particular, play a routed to a different program using automated keystrokes
significant role in EEG processing. By selectively utilizing and user datagram protocol (UDP) sockets, respectively.
spectral features from the alpha and beta-frequency bands, Games can be controlled by accessing the UDP port and
motor imagery has a profound effect on these regions of using the control signals sent to it.
the brain during EEG processing. The steps involved in
analyzing motor imagery signals using band powers include: 3.5. Operator
(i) Removing any artifacts from the raw EEG data. The graphical interface is designed to provide users with
(ii) Selecting the relevant inputs, such as F3 and F4 convenient access to the various modules previously
electrodes.
(iii) Applying 8–12 Hz and 18–35 Hz band-pass filters to discussed. Users have the ability to start, pause, and restart
the process, as well as adjust its settings, all while viewing
each path. 24
(iv) If N represents a time series, then it can be calculated the system’s parameters and real-time analysis results.
using Equation I, where α is the number of trails and γ 4. Experimental results
is the number of channels.
4.1. Data set description
N = α × γ × 2 (I) The BCI IV Competition-I dataset, accessible through
http://www.bbci.de/competition/iv/#dataset1, comprises
(v) To determine the average band power, an FFT must be publicly available EEG signals obtained from a task in which
used. the participants were instructed to focus on the muscles
The average method involves taking the mean and in their left and right hands and feet. Spectral features,
standard deviation of the time series data from a single including band powers, average power, and RMS power,
experiment, making it the most basic feature extraction play a crucial role in deriving valuable information from
method. Due to the oscillating nature of EEG signals and these EEG signals. These features were probably chosen for
the possibility that averaged values across classes may be the BCI IV Competition-I dataset because they are useful in
similar, this method is not particularly helpful for BCI differentiating between motor imagery tasks such as left- and
applications, particularly for motor imaging. On the other right-handed movements. These features are extracted by
hand, the RMS method calculates the average and standard computing them from each trial’s or time segment’s EEG
deviation of a signal by squaring it, finding its average, recordings, which are then fed into ML models that are
and then evaluating its square root using Equation II to responsible for either classifying or decoding the intended
get root-mean-square, where x denotes the mean of the motor imagery tasks. The dataset comprises data from 15
2
values x . individuals engaged in cued motor imagery tasks across four
i
classes: left hand, right hand, foot, and tongue. According to
RMS = x 2 (II) a 10-to-20 montage, electrodes were attached to the linked
mastoids T7 and T8 through the femoral arteries (F3, F4,
FC5, FC6, F7, and F8). The impedance of each electrode
3.3. Signal processing was maintained at 10 kΩ or less. The data were digitized
at 300 Hz using an A/D converter connected to a personal
In two distinct steps – feature extraction and translation computer. The dataset includes EEG recordings collected
– the signal processing module transforms the raw EEG from 15 individuals aged 18 to 36, while they performed
data into categorized results usable by output devices. cognitive tasks. Each class has 60 trials, and there are a total
Each phase is implemented separately from the other
and is subdivided into filters. Two filters, one spatial and of 60 channels in this data set. This EEG data feature was
one temporal, make up the feature extraction phase. In sampled at 250 Hz and filtered between 1 and 50 Hz with
the translation stage, the first filter employed is the linear a Notch filter enabled. Figure 3 depicts the positions of the
classifier, followed by a normalizer that adjusts the outputs electrodes that were used to record the experiment’s results.
to maintain a zero mean and a predetermined range of In this experiment, participants were seated in
values. overstuffed armchairs and instructed to visualize

Volume 2 Issue 2 (2024) 6 doi: 10.36922/bh.2819