Background
After the human sees the response of the device and makes a change to the device using the brain signal (Joseph, 2010). Then, the depth of the electrodes in the semi-invasive BCI system is implanted between invasive and non-invasive BCI.
Problem Statements
Aims and Objectives
The main parts of the human brain
The Lobes of the human brain
Brain Signals
It is strong when the person is in inner focus, prayer, meditation, daydreaming and spiritual awareness. It can measure when the person is simultaneously processing information from different parts of the brain.
EEG Headset
So, before using the EEG headphones, the researcher had to measure the skull size of the subjects. The nasion is the bridge of the nose from the lowest point between nose and forehead.
Types of Artifacts
Its amplitude will correlate with the EEG signal with the strength of the muscle contraction (Bit Brain, 2020). Body movements such as head movements, arm movements or leg movements will affect the contact quality of the electrode with the scalp and destroy the EEG signal (Bit Brain, 2020).
Blind Source Separation (BSS)
Independence Components Analysis (ICA)
Thus, full rank means that the number of linear independence columns or rows is the largest possible for a matrix of the same dimension (Stat Trek, n.d.). The number of detected signals must be greater than or equal to the number of independent signals. The first ambiguity is that ICA cannot determine the variance or the energies of the independence components.
In the ICA model, A and S are unknown, so it will take the unit variance of the source signal component as 1: 𝐸{𝑠𝑖2} = 1. Next, the second ambiguity is that ICA cannot determine the order of independence of the component. Thus, the model can arbitrarily change the order of expressions and call any independent component first.
AP-1 is the new mixture matrix and PS are the original independence signals with different orders (Hyvarinen & Oja, 2000).
Regression Method
The adaptive filter is a filter that can remove the reference signal from the observed signal. There are two components that group the adaptive filter, namely the digital filter and the adaptive algorithm, as shown in Figure 2-15. It has to estimate the desired signal 𝑠̂ and produce 𝑘 the optimal noise 𝑛̂ by reference noise 𝑛𝑘 𝑘 to cancel the artifact to get the pure signal 𝑠𝑘 from the polluted signal 𝑦𝑘 as shown in Equation 2.6 and Equation 2.7.
Since the noise 𝑛𝑘 and the pure signal 𝑠𝑘 are uncorrelated, the last term can be omitted and form the equation as shown in Equation 2.10. 𝐸(𝑠̂) represents the estimated signal power, 𝑘2 𝐸(𝑠𝑘2) represents the signal power, and 𝐸((𝑛𝑘− 𝑛̂)𝑘 2) represents the residual noise power. If we want to achieve the maximum signal-to-noise ratio, the total output of the canceller must be minimized (Adaptive Filters, 2015).
BCI important elements for neurofeedback application
There is a study that mentions that the effect of training in a virtual environment is more effective than in a real environment. Neurofeedback training consists of two groups, namely training in a virtual environment and training in a real environment. As the results in Figure 2.22 and Table 2.4 show, the virtual environment is more improved than the real environment.
All results showed more improvement in the virtual environment, except for the forward digit span test. The reason the virtual environment has more improvements is because the real environment has hardware limitations, it cannot provide competition like the AI players. So the virtual environment is one of the important elements that should be included in the neurofeedback system (Dong-Kyun, Min-Ho, John & Seong-Whan, 2019).
In the result shown in Figure 2.25, group 2 used 3D play stimulus content because the neurofeedback has more prefrontal alpha asymmetry than the 2D play stimulus content in group 1.
Summary
Other reasons for these results are that the 3D stimulus game has a high degree of difficulty and complexity than the 2D stimulus game and the participants need to receive more information to make decisions in the 3D game. So the brain will have better exercise when playing the 3D game (Yasir, Syed, Syed, Muhammad, Syed, 2019). The difference between the ICA and regression method is the ICA can be used in the unknown mixing process, but cannot cancel the noise that is not linear.
There are three important elements to improve the performance of the neurofeedback BCI system, which are interactive, virtual stimulus and 3D stimulus. In the virtual stimulus, the person can compete with the AI and the prize in real time. As a result, the actual effectiveness of the stimulus will be weaker than the virtual one.
The brain will gain more complex information from the 3D stimulus, so participants can have better training in the 3D stimulus.
Overview
This project has two main parts, pre-processing in which the EEG signals were recorded and artifacts were removed from the data in real-time, and NFT in which the subject received real-time feedback from his brain. In the neurofeedback training, the frequency of the alpha band was used, which is from 8 to 13 Hz, and the difference was observed before and after the neurofeedback training.
EEG Headset and computer requirement
Sensors Materials Hydrophilic semi-dry polymer Connection Wireless: Bluetooth Low Energy EEG Sampling Rate 128 samples per second per channel EEG Resolution 14 bits with 1 LSB = 0.51μV.
EMOTIV PRO
OpenViBE
It uses the various function boxes shown in Figure 3.13 to process the EEG data into useful information. In the setting for the channel selector shown in figure 3.16, the channels they select are number 4 to number 8. There are 10 channels of data coming from EMOTIV PRO and channel numbers 1, 2, 9 and 10 are not the data of of the EEG signal, so we simply select the 5 channels that comprise the EEG data.
Behind the channel selection box is a temporal filter because the EEG data needs to be limited to a fixed frequency range and to confirm that the sign of the signal remains the same compared to the raw signal and the clean signal. The field frequency for a MATLAB script follows the bits in MATLAB, so MATLAB uses 64-bit and its field frequency will be 64. The MATLAB working directory is the location of the MATLAB files that the program needs to process.
The name of the signal stream in the setup names the LSL import software and the researcher will easily identify the different streaming data.
MATLAB for OpenViBE scripting
Unity3D
In the downloaded files, they provide two sample scenes, namely the LSL Data Receiver and the LSL Data Transmitter. The LSL data receiver will be used in the project, and the prototype interface is shown in Figure 3.23. It can display the stream name, device ID, number of channels, header names and data value.
Next, we make some changes from the prototype version and add a green bar that will slide following the value as shown in Figure 3.24 (Bernard Polidario, 2021). After all the design and C# scripting of the scene was done, the application was built using the build setting shown in Figure 3.25.
Neurofeedback training
MATLAB for data analysis
Budget
Artifact removal by using ICA-REG
The working of ICA-REG
The whole process of the ICA-REG can be divided into three phases: independent component analysis, EOG blinking artifact detection, and the regression method. The first stage of the ICA-REG is to perform independent component analysis in the raw signals to separate them into the independent components and the unmixing matrix W. Kurtosis is a statistical term describing how much the tails of a distribution differ from the tails of a normal distribution (CFI, n.d.).
For finding the artifact EOG IC, these ICs distributions were standardized to zero mean and unit standard deviation. If the value of the Z-score from ICs is more than 1.64, these ICs will detect as artifactual independent components (Kazi A. and Gleb V. T., 2015). It uses the detected artifactual IC to remove the EOG clipping noise from the raw signals.
Finally, the adaptive spatial filter is multiplied with the raw signals to obtain clean signals.
Advantages of using ICA-REG
The difference between the normal ICA and the fast ICA is that the calculation of the fast ICA is lower than the normal ICA, which requires less memory and more efficiency (Stack Exchange, 2017). The classic regression method, it does not use the spatial adaptive filter to cancel the noise, which makes the calculation very expensive and not available for the BCI application. The ICA-REG method uses the spatial adaptive filter to perform artifact removal, so it reduces the load of the calculation and is available to the BCI system.
Result of the ICA-REG
Data Analysis from the NFT training
Correlation
The correlation is the weight of the relationship between the two variables or the signals which means the similarity of these signals. In signal processing, the value of the positive correlation and negative correlation shows how similar the two variables are. In our data analysis, we compare the correlation between the raw signal and the clean signal in each session.
The correlation in the pre-training and post-training session has a high correlation that is at least 0.8895. The high correlation in the pre- and post-training session is because we record the raw and clean signal at the same time. Subsequently, the correlation in the NFT training sessions has a very small value that is between -0.01 and 0.01.
In addition, the high correlation in the pre- and post-training session means that the artifact removal performance is good.
Root Mean Square Error
Alpha band power
Future work and Recommendations
If the interface of the neurofeedback is more interesting, the result of the NFT training can also be increased.
Conclusion
Dostopno na: < https://www.neuroelectrics.com/blog hans-berger-lights-and- shadows-of-the-inventor-of-electroencephalography/>. Dostopno na:
Available at: < https://www.livingwelldallas.com/living_well_dallas_news/how-a-qeeg- session-can-change-your-life/>. Available at:
Available at: < https://www.investopedia.com/ask/answers/032515/what-does-it-mean- if-correlation-coefficient-positive-negative-or-.
