Improving the common spatial pattern filtering method to improve the efficiency of computer-brain interface systems

Master's thesis in the field

Medical Engineering

Improving the joint spatial pattern filtering method to improve the efficiency of computer-brain interface systems

       Computer-brain interface systems are systems that can translate brain electrical signals related to movement concepts in the human brain into comprehensible computer commands. Therefore, this capability can help many sensory-motor patients and solve their problems in relation to their surroundings to a very high extent. However, the current systems have not yet been able to enter the market commercially due to the lack of high enough accuracy for all people.

In order to improve the accuracy and performance of these systems, it has been experimentally shown that the use of the "joint spatial pattern" method to extract appropriate separating features from brain signals is an optimal method in synchronous computer-brain interface systems, but it still has problems and challenges.  In this research, two solutions have been proposed to modify the non-linear method of the joint spatial pattern: the first solution, which is the personalization of the linear dust algorithm, includes two new combined methods. The first method, by adding signal frequency-spatial information to the common spatial pattern method, has led to the improvement of the algorithm's performance in order to find better distinctive features, and based on these features, the accuracy of data classification increases. In the second method, which is implemented in two different ways, the co-activity value of all channels is injected two by two into the CSP formulation, and then the values ??obtained by a linear kernel function go to another space and finally improve the separation between movement perceptions.

Human brain and its activities

From the point of view of anatomy and physiology, the nervous system and especially the human brain as the main processor, decision maker and controller. All human behavior is known. It can be said that the brain system is the most complex and unknown organ of the human body because the way it functions in topics such as learning, memory, information processing, creating relationships between different concepts, etc., has not been clearly defined yet, and there are many questions for the scientists of the world. Knowing as much as possible how the different parts of the brain work can be effective in solving many neurological, psychological, sensory and motor deficiencies of the human body, and this fact has caused more and more scientists to pay attention to how the human brain works in recent years.

In order to process brain signals, a way to receive them must first be found. Neurons are the smallest data processing unit in the brain, and electrical communication between them is the basis of brain function [1]. The electrical connection between neurons creates electric and magnetic fields and, as a result, electric currents in the volume of the brain, and measuring these fields and currents is one of the ways to record brain activities. Among the methods used in this field, EEG [1] can be mentioned, in its non-invasive type, a set of electrodes recording the electrical signal is placed on the scalp and the electrical signals of the brain are recorded while performing a specific activity. In another method called magnetoencephalography or MEG [2], the magnetic activities of the brain are recorded by means of superconducting recorders placed on the head. But basically, the MEG recording device is very expensive and has high technology. In addition to the methods of recording electrical activities in the brain, there are other methods that record metabolic activities in the brain. In fact, these methods are different medical imaging methods such as fMRI, PET, SPECT, etc. They use and determine which parts of the brain have metabolic activities at any given moment. The advantages of brain metabolic imaging methods include high spatial resolution and the possibility of three-dimensional localization of brain activity sources. But most of these methods are very expensive and not always available.A) sample of EEG signal, b) sample of EEG electrodes and how to record the signal Brain-computer or [3]BCI have been developed to solve this challenge. In fact, they are systems that have the ability to separate brain images (especially the image of body movements) and translate them into commands that can be understood by the computer. Based on the individual's brain signals, this capability can greatly help patients such as those suffering from cerebral-spinal insufficiency such as ALS[4] or other motor insufficiency[5] [2,3]. Because the BCI system provides a communication channel for this type of patients that does not depend on peripheral nerves and muscles. For example, in ALS disease, the neurons related to voluntary movement degenerate over time and the person gradually loses the ability to move his various organs. In the initial stages of this disease, problems arise in moving the hands and feet, chewing and swallowing food, and in the final stages, the patient completely loses the ability to walk and use his hands and feet, and the act of chewing and swallowing becomes very difficult for him, so that the patient may suffocate. Also, at this stage, even the person's breathing faces problems and the person will not be able to speak [4].

    Although in ALS disease, the person loses the ability to move his limbs, this disease does not affect the sensory capabilities (hearing, smell, sight, taste and touch) as well as the mental activities of the person, and the patient has the ability to imagine performing various movements. Therefore, if with the help of the BCI system it is possible to determine from the brain signals what kind of movement the patient imagined, he can be given great help. In this way, such patients, as well as other reasonable patients (who are intellectually healthy) who have motor impairment, can control prostheses and artificial organs and external devices such as light switches, etc. by motor imagination (MI) [6], which greatly expands the personal ability of a person. Moving body parts in the daily life of a healthy person is quite common. Research has shown that when a member of one procedure [7] (such as the left hand) is moved, the range of Mu Rhythm (9-13Hz) and Beta Rhythm (18-22Hz) of the motor cortex of the other side of the brain [8] is significantly reduced, while the rhythm of the motor cortex of the same side of the brain [9] is significantly increased. Furthermore, when a person imagines performing an action (without actually performing the action), they activate the same motor area of ??the brain (and produce the same EEG pattern) as when they actually perform the action [5]. In fact, the performance of computer-brain interface devices is based on this issue and considers the signals of the imagination of performing a movement.

More precisely, it should be said that the performance of almost all BCI systems is such that after going through a training course and learning how to work with signs and the system, people must imagine doing the activity related to the sign by seeing a series of special signs on the monitor in front of them. For example, seeing the arrow on the left, imagine the movement of the left hand, the arrow on the right hand imagine the movement of the right hand. During these visualizations, signals are recorded. It should be noted that in most cases, the exact time of imagining is not known and signaling is done in a period of several seconds during which movement imaginations are performed. For example, if 4 seconds of signaling is done, it will not be clear at which moment of these 4 seconds, the person imagined performing the limb movement. In BCI systems, EEG signals of the person are recorded because these signals have a better synchronization with brain activities and can obtain fast movements of brain information processing in a non-invasive and real-time manner, and also because they are always available and easily used [6]. Although EEG signals are known for having low spatial resolution and high noise level. It has been shown that spatial filter methods are effective in extracting brain activity patterns.

The purpose of this research is that by having the EEG signals of a person, we can identify which corresponding signal belongs to the person's movement perception. Of course, here there is the assumption that the said signal belongs to one of the few predetermined movements. The reason that here the signals are classified into only a few movements is the low quality of the EEG signal and the little information they provide.