1. What does the parameter C control in SVM?
The parameter C in SVM controls the trade-off between the slack variable penalty and the size of the margin. It determines the balance between maximizing the margin and minimizing the classification error. A smaller value of C allows for a larger margin but may result in more misclassifications, while a larger value of C prioritizes minimizing misclassifications but may lead to a smaller margin. The parameter C is crucial in determining the performance and generalization ability of the SVM model.
read more
2. What are the three parts of a neuron?
A neuron consists of three distinct parts: the cell body or soma, dendrites, and the axon. The axon is a long cylinder that originates from the cell body as the axon hillock. In some axons, there is a fatty covering called the myelin sheath, which is formed by flattened glial cells called schwann cells. The myelin sheath is interrupted at intervals by gaps called nodes of Ranvier. Synapses are the junctions between neurons, where action potentials pass from one neuron to another. The synapses consist of presynaptic and postsynaptic regions.
read more
3. What is the synaptic cleft and its role in synaptic transmission?
The synaptic cleft is the gap that separates the pre and postsynaptic membranes in a synapse. It is generally 20-25nm wide and filled with a mucopolysaccharide that 'glues' the pre and postsynaptic membranes together. The synaptic cleft plays a crucial role in synaptic transmission as it allows the exchange of chemical substances between the presynaptic and postsynaptic regions. In the presynaptic region, synaptic vesicles containing transmitter substances are located, while in the postsynaptic region, specific receptors bind these transmitter substances. The synaptic cleft facilitates the diffusion of these substances, enabling the transmission of signals across the synapse. This process is essential for communication between neurons and the overall functioning of the nervous system.
read more
4. What role does the cerebral cortex play in brain functions?
The cerebral cortex, the outer sheet of gray matter covering the brain's hemispheres, plays a crucial role in various brain functions. It is involved in memory, attention, perceptual awareness, thought, language, and consciousness. The cortex is considered the seed of the mind due to its significant contribution to these cognitive processes. The elaboration of the cortex into different areas, known as Brodmann's regions, represents functionally distinct regions of the brain. Each Brodmann number corresponds to a specific area, making them useful for referencing different parts of the cortex. Understanding the relationship between brain functions and Brodmann's regions helps researchers identify the functional cortex divisions and their impact on mental states.
read more
5. Where is the primary visual cortex located?
The primary visual cortex, also known as area 17, is located in the Occipital lobe. It is the first place in the cortex that receives visual signals from the thalamus. This area plays a crucial role in processing visual information and is essential for visual perception. The Occipital lobe is situated at the back of the brain and is responsible for interpreting visual stimuli. Understanding the location and function of the primary visual cortex is vital for researchers studying vision-related disorders and developing treatments for visual impairments.
read more
6. What are the executive functions of the frontal lobe?
The executive functions of the frontal lobe involve recognizing future consequences of current actions, choosing between good and bad actions, overriding and suppressing unacceptable social responses, and determining similarities and differences between things or events. These functions contribute to higher mental processes and help in retaining longer-term memories associated with emotions. The frontal lobe also modifies emotions to fit socially acceptable norms, playing a crucial role in emotional regulation and social behavior.
read more
7. What are the primary and secondary visual cortices in the occipital lobe?
The occipital lobe mainly contains primary and secondary visual cortices, specifically Brodmann areas 17, 18, and 19. These areas are responsible for processing visual information and play a crucial role in creating a coherent view of the environment. The visual information flows from the back of the brain to the front, with the brain expending most of its energy in packaging sensory input from all available modalities. Vision is combined with somatosensory information to give a sense of where one's body is in space. Memory functions in the temporal lobe allow for recognition of the visual perceptions. The processed sensory input finally makes its way to the frontal lobe, where decisions are made regarding the various stimuli.
read more
8. What are the three functional areas of the cerebral cortex?
The cerebral cortex contains three kinds of functional areas: sensory areas that provide for conscious awareness of sensation, motor areas that control voluntary motor functions, and association areas that integrate diverse information for purposeful action. Each hemisphere is chiefly concerned with sensory and motor functions of the opposite side of the body. Although largely symmetrical in structure, the two hemispheres are not equal in function, with specialization in cortical function in each hemisphere. No functional area of the cortex acts alone, and conscious behavior involves the entire cortex in one way or the other. Different sensory inputs are processed at different parts of the cortex, and particular cortex areas are involved in higher mental functions such as memory, control of attention, complex planning, and reasoning. When sensory inputs are processed or when other mental processing takes place, the corresponding regions of the cortex are particularly active, which may alter EEG signal.
read more
9. What are the characteristics used to classify EEG techniques?
EEG techniques can be classified based on spatial resolution, temporal resolution, intrusiveness, resources required for operation, physiological parameters monitored, and applicability as portable devices. These characteristics help determine the suitability of EEG techniques for specific research or medical purposes. Spatial resolution refers to the ability to distinguish between different brain regions, while temporal resolution measures the ability to capture rapid changes in brain activity. Intrusiveness refers to the level of invasiveness of the technique, with non-invasive methods like EEG being preferred. Resources required for operation consider the equipment and expertise needed to conduct EEG measurements. Physiological parameters monitored include brain activities, and applicability as portable devices refers to the ease of use and mobility of the EEG equipment. Overall, these characteristics help researchers and medical professionals choose the most appropriate EEG technique for their specific needs.
read more
10. What causes EEG recordings?
EEG recordings are the summation of electrical fields produced by millions of interconnected neurons. Dendrites, axons, and cell bodies of neuronal components produce currents due to the following reasons: 1. All cortical neurons have the same orientation, dendrites near the surface and axons projecting inward. 2. Sodium entering dendrites during neuronal firing leaves the outside of the dendrites negatively charged. The architecture of the brain varies with different locations, causing EEG to vary depending on the location of the recording electrodes. The potential differences between two points of the scalp are different due to the small electric potentials induced by single neurons, the need for synchronized neural activity, and the radial arrangement of apical dendrites in the cortex. EEG activity reflects the summation of synchronous activity of thousands or millions of neurons with similar spatial orientation, radial to the scalp. Rhythmic activities in EEG are categorized into five general types, each with distinct frequency ranges and correlations with patterns of behavior.
read more
11. What is the 10-20 system for EEG electrode placements?
The 10-20 system is an internationally accepted standard for electrode placements introduced in 1957 by the International EEG Federation. It ensures that all cortical regions, which might show interesting EEG patterns, are covered. The system requires three anatomical reference points to be determined before electrode positions can be introduced. These reference points include Nasion, which is the intersection of the frontal and two nasal bones of the human skull. The 10-20 system is widely used in research and clinical settings to ensure consistent and accurate EEG recordings.
read more
12. What is the significance of the 10-20 system in electrode placement?
The 10-20 system is an international standard for electrode placement on the scalp during electroencephalography (EEG) procedures. It uses anatomical landmarks on the skull to ensure consistent and accurate electrode placement across different individuals and laboratories. This system avoids placing electrodes near the eyeballs, which can interfere with the EEG signals. By following the 10-20 system, researchers can obtain reliable and comparable EEG data. Additionally, the system allows for the placement of up to 256 additional electrodes for specialized applications, providing more detailed information about brain activity. The 10-20 system is crucial for obtaining high-quality EEG data and ensuring the validity of research findings.
read more
13. What are the two types of EEG electrodes?
The two types of EEG electrodes are polarized (or reversible) electrodes, usually made of precious metals like gold or stainless steel, and non-polarized (or irreversible) electrodes, typically made of silver (Ag) with a thin silver-chloride (AgCl) layer on top (Ag/AgCl electrodes). Polarized electrodes are less commonly used due to their higher resistance, while non-polarized electrodes are state-of-the-art with lower resistive components. Both types are essential for EEG recording and play a critical role in determining the quality of the EEG signal.
read more
14. What are the basic requirements for a bio potential amplifier?
A bio potential amplifier must meet several basic requirements to ensure accurate and safe EEG signal amplification. Firstly, the measured signal should not be distorted, and the physiological process being monitored should not be influenced by the amplifier. The amplifier should provide the best possible separation of signal and interferences. Additionally, the amplifier must offer protection against damages from high input voltages, such as those encountered during defibrillator or electrosurgical instrumentation application. It should also protect the subject from any hazard of electric shock. These requirements ensure the integrity and safety of the EEG signal amplification process.
read more
15. What are the common types of EEG artifacts?
EEG artifacts are interfering waveforms added to the EEG signal during recording sessions. They are generally termed as Artifact or Noise and are not due to brain activity. Artifacts can be caused by subject movements, heart, muscle, and eye electrical activity. There are two types of artifacts: technical and biological. Technical artifacts can be avoided with correct equipment operation and subject preparation, while biological artifacts are unavoidable. Figure 3.9 shows waveforms of common EEG artifacts, including eye blinks, muscle artifacts, and cardiac artifacts. Understanding and identifying these artifacts is crucial for accurate EEG analysis and interpretation.
read more
16. How to remove technical artifacts in EEG data?
Technical artifacts in EEG data can be removed by decreasing electrode impedance, using shorter electrode wires, and applying built-in notch filters in the EEG amplifier. Proper grounding of the subject can also help avoid common technical artifacts. Fluorescent lights emitting radiation noise can be replaced with special incandescent or white LED bulbs. Any remaining artifacts that cannot be removed at the hardware level should be eliminated using software preprocessing techniques. Additionally, placing additional electrodes for monitoring eye blinks, eye moments, and electrical contamination due to heart beats and muscle activities can improve artifact discrimination. Two main techniques are used in BCI research to handle artifacts: discarding affected segments of EEG data and using separate channels for automatic eye activity detection and rejection. Blind source separation methods, such as Principal Component Analysis (PCA), Signal fractional analysis (SFA), and Independent Component Analysis (ICA), are popular for removing artifacts. However, regression methods may perform better for a small number of EEG channels. Recent work suggests that regression methods are appropriate for EOG reduction as well.
read more
17. What are the typical settings for high-pass and low-pass filters in EEG signal processing?
The typical settings for high-pass filter in EEG signal processing are 0.5-1 Hz, and for low-pass filter, they are 35-70 Hz. These settings help filter out slow artifacts, such as electrogalvanic signals and movement artifacts, and high-frequency artifacts, such as electromyography signals, respectively. Additionally, frequencies between 48-62Hz should be removed to avoid technical equipment-induced waves. A notch filter can be used to remove specific frequencies, such as those caused by electrical power lines (50/60 Hz).
read more
18. What is the purpose of feature vector construction in BCI?
Feature vector construction in BCI aims to reduce dimensionality while preserving important signal features. It involves extracting features and constructing feature vectors from digitized signals, which are often of high dimensionality. Various digital signal processing methods are used to extract features and construct feature vectors. Dynamic classification is also employed, where features are extracted from several time segments to build a temporal sequence of feature vectors. This sequence can be classified using a dynamic classifier. In this investigation, PCA, Bandpowers, and Downsampling with and without scaling were implemented for feature vector construction. The goal is to select the most appropriate classifier for a given BCI system.
read more
19. What is the main advantage of PCA?
The main advantage of PCA is its ability to reduce dimensionality and compress data while retaining essential characteristics. By identifying patterns in high-dimensional data, PCA simplifies complex data sets, making it easier to analyze and interpret. This reduction in dimensions helps filter out noise and reveal hidden structures within the data. PCA is a powerful tool used in various fields, such as neuroscience and computer graphics, due to its simplicity and non-parametric nature. It provides guidance on how to transform a noisy data set into a lower-dimensional representation, preserving the most meaningful information. Principal component analysis finds orthogonal axes that best represent the data's trends, allowing for a more focused analysis of the data's underlying structure.
read more
20. What is the role of eigen vectors in PCA?
In PCA, eigen vectors play a crucial role in determining the principal components of a dataset. The transformation matrix P, which is used to express the columns of X in terms of new basis vectors, is formed by the eigen vectors of the covariance matrix CX. These eigen vectors represent the principal components of X, which are the directions in which the data varies the most. By finding the eigen vectors of the covariance matrix, we can identify the maximum variances in the dataset and transform the data accordingly. The rows of the transformation matrix P are the eigen vectors, and they serve as the principal components that capture the essential information in the dataset while minimizing the covariance between principal components.
read more
21. How are band powers calculated in BCI?
Band powers are calculated by first transforming the EEG signal into the frequency domain using Fourier transformation. The power spectrum is then calculated, and the frequency range is divided into bands with equal band width. The band width can be optimized for maximum classification accuracy. The total power for each band is determined, and a feature vector is constructed using these band power values. In detail, if X is a matrix with column vectors X1, X2, ..., Xm representing data from m EEG channels, the Fourier transform of each column vector Xi is denoted as XiY. This process allows for the construction of feature vectors using band powers, which is popular in the BCI community.
read more
22. How is the power spectrum matrix P calculated?
The power spectrum matrix P is calculated by multiplying the element-wise product of matrices i Y and i Y, and then squaring the result. The formula is P = (1/2) * (i i Y * i Y)^2. The matrix P represents the power spectrum of a signal X, where i is the frequency index and Y is the conjugate of the signal. The matrix P has dimensions m x m, where m is the number of frequencies. The power spectrum matrix P is then divided into bands, with each band containing b frequencies. The feature vector F has dimension m and is formed by the kth band's power spectrum matrix Pk, which is given by the equation Pk = (1/b) * sum(j=1 to bk) Pj, where Pj represents the power spectrum matrix of the jth frequency in the kth band. This process allows for the analysis of the frequency components of a signal and their corresponding power levels.
read more
23. What is down sampling in feature vector construction?
Down sampling is a method to reduce feature vector dimensions by collecting data points at regular intervals. It involves constructing a vector by gathering data points from a column vector Xi, which contains n EEG data points for a specific channel. The down sampling process is achieved by sampling the data at a regular rate, resulting in a down sampled feature vector Fi for each channel. The overall feature vector is represented as mFmFmFmFmFm. This technique simplifies the data and reduces its dimensionality, making it easier to process and analyze.
read more
24. What classification techniques are used in brain computer interface systems?
Several classification techniques are employed in brain computer interface systems. The most widely used methods include Discriminant Analysis (both linear and nonlinear), support vector machines (linear and nonlinear), nearest neighbor classifiers, Neural networks (NN), Multilayer perceptron, Nonlinear Bayesian classifiers (such as Bayes quadratic), and Hidden Markov Models (HMM). In this study, the researchers employed Discriminant Analysis (both linear and nonlinear), support vector machines (both linear and nonlinear), and k-th nearest neighbor classifiers. These techniques are crucial for classifying cognitive tasks in brain computer interface systems, and their effectiveness is discussed in detail in the following sections.
read more
25. What is the purpose of Linear Discriminant Analysis (LDA)?
Linear Discriminant Analysis (LDA) aims to find the optimal linear combination of features that maximizes the separation between two or more classes of objects or events. It achieves this by maximizing the ratio of between-class variance to within-class variance in a given dataset. This ensures maximal separability and helps in distinguishing different classes effectively. LDA is closely related to Fisher's linear discriminant and is widely used in statistics and machine learning for classification tasks.
read more
26. What is the scatter matrix in feature space x?
The scatter matrix in feature space x, denoted as S, represents the covariance matrix of the data points in the feature space. It is a square matrix that captures the variance and covariance between different features in the dataset. By solving the general eigenvalue problem, we can obtain the eigenvectors and eigenvalues of the scatter matrix. The eigenvectors, denoted as w, represent the directions in the feature space that capture the most variance in the data. The eigenvalues, denoted as T, represent the amount of variance captured by each eigenvector. The scatter matrix plays a crucial role in dimensionality reduction techniques such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA).
read more
27. What is the role of Support Vectors in Support Vector Machines (SVM)?
Support Vectors are the sample points closest to the separating hyperplane in Support Vector Machines (SVM). They play a crucial role in determining the orientation of the hyperplane. The aim of SVM is to orient the hyperplane in such a way that it is as far as possible from the closest members of both classes. By considering only the Support Vectors, we can define two planes, H1 and H2, that these points lie on. The hyperplane is equidistant from H1 and H2, and the distance between the hyperplane and these planes is known as the SVM's margin. Maximizing this margin is essential for achieving optimal classification. The optimization process involves finding the values of w and b that minimize the margin while satisfying the constraints, using Lagrange multipliers and Quadratic Programming (QP) optimization techniques.
read more
28. What is the significance of the dot product in the Dual form of L D?
The dot product in the Dual form of L D is crucial as it simplifies the calculation process. Instead of computing the entire L P formulation, the Dual form only requires the dot product of each input vector x i. This simplification is essential for implementing the Kernel Trick, which allows for non-linear separability in the data. By utilizing the dot product, the Dual form enables efficient optimization and calculation of the Support Vector Machine (SVM) parameters, leading to improved classification performance.
read more
29. How to handle non-linearly separable data in SVM?
To handle non-linearly separable data in SVM, we relax the constraints slightly by introducing a positive slack variable. This allows for misclassified points and combines equations into a form that accommodates non-linear separability. The slack variable, denoted as 'i', ranges from 1 to N, where N represents the total number of data points. The equations are combined into 0 1 . i i i b w x y, ensuring that i b w x y i i i 0 1 is satisfied. This approach enables the SVM methodology to extend its capabilities and effectively handle data that is not fully linearly separable.
read more
30. How does kNN classification work in BCI systems?
kNN classification in BCI systems assigns an unseen point the dominant class among its k nearest neighbors within the training set. It uses a metric distance, such as Euclidean distance, to determine the nearest neighbors. With a high value of k and sufficient training samples, kNN can approximate any function, enabling it to produce nonlinear decision boundaries. However, kNN algorithms are sensitive to the curse of dimensionality, which can lead to failure in BCI experiments. Despite this, kNN may prove efficient in BCI systems with low-dimensional feature vectors, as demonstrated in some studies [27].
read more
31. What are the three terms that describe classification error?
The three terms that describe classification error are noise, bias, and variance. Noise represents the noise within the system and cannot be reduced. Bias represents the divergence between the estimated mapping and the best mapping. Variance reflects the sensitivity to the training set used. To achieve the lowest classification error, both bias and variance should be low. However, there is a natural bias-variance tradeoff, where stable classifiers tend to have high bias and low variance, while unstable classifiers have low bias and high variance. This tradeoff can explain why simple classifiers sometimes outperform more complex ones. Techniques like stabilization techniques, combination of classifiers, and regularization can be used to reduce variance and improve classification accuracy.
read more
32. What are the commonly used mental tasks in BCI systems?
The commonly used mental tasks in BCI systems include Baseline M1, Mental Arithmetic M2, Mental Letter Composing M3, Visual Counting M4, and Geometric Tasks with Figure Rotation. Baseline M1 involves the subject not performing a specific mental task while relaxing and thinking of nothing in particular. Mental Arithmetic M2 requires subjects to solve non-trivial multiplications without vocalizing or moving. Mental Letter Composing M3 involves mentally composing a letter to a friend or relative without moving or vocalizing. Visual Counting M4 requires subjects to imagine a blackboard and visualize numbers being sequentially written on it. Geometric Tasks with Figure Rotation involve subjects imagining the generation of words beginning with the same random letter and rotating figures in their mind.
read more
33. What is visual stimulus driven letter imagination?
Visual stimulus driven letter imagination involves subjects imagining letters based on visual stimuli. In this task, participants are shown a letter and then asked to mentally visualize and imagine the letter without physically writing it. This process engages the visual cortex and related brain regions, allowing researchers to study the neural mechanisms underlying visual imagery. By analyzing the brain's electrical activity during this task, researchers can gain insights into how the brain processes and represents visual information. This technique has applications in various fields, including cognitive neuroscience, psychology, and brain-computer interface (BCI) research. Understanding the neural basis of visual imagery can contribute to the development of more effective BCIs and assistive technologies for individuals with communication impairments.
read more
34. What is the purpose of using NPS calibrator 00230?
The NPS calibrator 00230 is used to calibrate the EEG recording system. It injects a precise 16 Hz, 50 mV signal into all 24 amplifiers in the system. The recording software then reads the outputs of the amplifiers and calculates a correction factor to bring the amplifiers to the norm. This calibration process is essential to account for component tolerance and ageing issues that can cause the amplifiers to vary by about 10% from the norm. By calibrating the system, accurate EEG recordings can be obtained, ensuring reliable data for research purposes.
read more
35. What software programs are used for EEG data recording?
Three software programs, LDC, DCon, and IMTE, are used for recording and manipulating EEG data. LDC has calibration features, DCon converts binary data to required formats, and IMTE performs preprocessing, feature vector construction, and classification. These programs are connected via SCSI and have GUIs for parameter tuning and optimization. They save data in separate files and allow for multiple calculations with different parameters. The output files contain all parameters used in each calculation.
read more
36. What steps are involved in preparing subjects for EEG recording?
The steps involved in preparing subjects for EEG recording are as follows: 1. Put the harness on the subject. 2. Choose a proper size electro-cap for the subject. 3. Attach the ear electrodes by applying conducting gel and abrading the skin. 4. Slip the cap onto the subject's head and attach it to the body harness. 5. Ensure impedance between reference electrodes and individual electrodes is below 2.5 Kohm. 6. Move the subject to a chair in front of a blank wall. 7. Connect the wrist band to the subject and ground. 8. Perform test recordings to check for noise in the EEG signal. 9. Use an alarm program to inform the subject about mental tasks and recording periods. 10. Allow subjects to rest and relax as needed. These steps ensure proper setup and minimize noise in the EEG signal.
read more
37. What recording parameters were used in both HS and MI tasks?
In both HS and MI tasks, the same recording parameters and settings were used. These parameters and settings are detailed in Table 6.2. Before starting the experiment, the amplifier system was calibrated as described in the previous chapter. The EEG laboratory at the Institute of Fundamental Studies has a separate EEG recording room divided into two sections to ensure the subject and recordist are well separated. Tube lights in the recording room were turned off during recording sessions to eliminate electrical noise, although the room was dimly lit due to light from the adjacent laboratory area. Additionally, the EEG amplifier and computer used for recording were kept as far apart as possible to reduce noise contamination.
read more
38. What is the optimal performance for mental tasks in BCI?
The optimal performance for mental tasks in BCI varies with the subject and the parameters used. In the study, the best performance for each subject was less than 65% for RH and LH, and less than 70% for DH and UH. The parameters used in preprocessing, feature vector construction, and classification need to be changed to achieve optimal performance for each mental task. In a real BCI system, the channels, classification methods, and optimal parameters can be determined automatically for a given subject. The best performance of each subject for fixed channels, methods, and parameters used in classifying mental tasks is given in tables (6.6) below. For MI, the performance results are presented in tables (6.7, 6.8, 6.9, 6.10) to compare with HS.
read more
39. How did HS perform compared to MI in the study?
In the study, it was evident that HS outperformed MI. Subject 2 achieved a perfect score of 100% for HS, while MI scored 92%. Subject 1 performed well with HS, scoring between 88% and 98%, but poorly with MI, scoring between 75% and 80%. Subject 3 had good performance with HS, scoring between 80% and 90%, while MI scored between 82% and 98%. Overall, the mental tasks introduced in HS performed better than the widely used MI.
read more