KR102463667B1 - Mental state classification method based on weighted cross-frequency coupling information of EEG data - Google Patents

Abstract

The present invention relates to a brain state classification method based on weighted cross-frequency coupling information of EEG data. The present invention comprises the steps of: receiving the EEG data from the measurement electrode by the receiving unit; performing segmentation and preprocessing on the EEG data measured by the preprocessing unit; and decomposing the preprocessed EEG data by the decomposing unit into low-frequency and high-frequency envelope signals and generating, by a matrix generator, a covariance matrix of the decomposed signal.

Description

[0001] Mental state classification method based on weighted cross-frequency coupling information of EEG data

The present invention relates to a brain state classification method based on weighted cross-frequency coupling information of EEG data, and in particular, based on Weighted Cross-Frequency Coupling (WCFC), which combines the power of EEG data and cross-frequency coupling information. It relates to a method of classifying a user's brain state with only a small number of electrodes.

The brain-computer interface (BCI) system based on brain waves is a system that operates devices such as computers by classifying brain states using only brain wave activity. A characteristic signal generated when performing a specific task is extracted and used as an input signal. A typical classification problem using the BCI system is motor imagery classification. Event-related desynchronization in which the rhythm of Mu (8-14 Hz) and beta (15-30 Hz) decreases when imagining hand movements is found in EEG measured near the scalp near the motor cortex . Based on these EEG characteristics, the imaginary direction of hand movement is classified. However, it is not easy to predict the direction of movement when only the power (or magnitude) information of these brain action potentials is used. In fact, the brain state is projected on various information in addition to the power information of the EEG. For example, cross-frequency coupling, which exchanges information through relationships with different frequencies, is related to various functions of the brain. Conventional methods for classifying brain states have a disadvantage in that a large number of measuring electrodes are required.

As a related prior art, there is Republic of Korea Patent No. 10-1293446 "EEG classification method and device for imagining movement" It is only disclosing a classification method.

The problem to be solved by the present invention is to provide a method for classifying and predicting brain states by maximizing information even in a small number of electrode environments based on weight-based cross-frequency coupling information that exchanges information through relationships with different frequencies. there is in

The brain state classification method based on the cross frequency coupling information of the EEG data of the present invention comprises the steps of: a receiving unit receiving EEG data from a plurality of measurement electrodes; and a feature extracting unit weighted cross frequency information (WCFC) from EEG data extracting a feature through frequency coupling), combining a plurality of features extracted by the feature combining unit, and classifying the brain state in frequency bands from the features combined by the feature combining unit by the classifying unit. .

The brain state classification system based on cross-frequency coupling information of electroencephalogram data of the present invention includes a receiver for receiving electroencephalogram data from a plurality of measurement electrodes, and weighted cross-frequency coupling (WCFC) in the electroencephalogram data. It includes a feature extracting unit for extracting features through , a feature combining unit for combining a plurality of features extracted from the feature extracting unit, and a classification unit for classifying brain states in frequency bands from the features combined in the feature combining unit.

According to the present invention, there is an effect of improving the accuracy of brain state classification and prediction using cross-frequency coupling information of EEG data measured with only a small number of electrodes.

Beyond the limit of brain state classification using only the power information of EEG, it is easy to predict the direction of imagined motion by using not only power but also cross-frequency combined information.

1 is a flowchart illustrating a brain state classification method using weight-based cross-frequency information in EEG data according to an embodiment of the present invention.
2 is a flowchart illustrating a method of extracting a feature through weight-based cross frequency information of a feature combining unit according to an embodiment of the present invention.
3 is a block diagram of a brain state classification system using weight-based cross-frequency information in EEG data according to an embodiment of the present invention.
4 to 7 are exemplary diagrams illustrating a brain state classification method using weight-based cross-frequency information in EEG data according to an embodiment of the present invention.

The specific structural or functional description of the embodiments according to the concept of the present invention disclosed in this specification is only illustrated for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention are It may be implemented in various forms and is not limited to the embodiments described herein.

Since the embodiments according to the concept of the present invention may have various changes and may have various forms, the embodiments will be illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosed forms, and includes all modifications, equivalents, or substitutes included in the spirit and scope of the present invention.

The terms used herein are used only to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as “comprise” or “have” are intended to designate that a feature, number, step, operation, component, part, or combination thereof described herein is present, but one or more other features It is to be understood that it does not preclude the possibility of the presence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a brain state classification method using weight-based cross-frequency information in EEG data according to an embodiment of the present invention.

Referring to Figure 1, the receiving unit of the brain state classification device receives the EEG data from a plurality of measurement electrodes (S101). The feature extraction unit extracts features from weighted cross-frequency coupling information (WCFC) from the received EEG data (S103). In this case, the weighted cross-frequency combination information means information that combines the power of EEG data and cross-frequency coupling information, and the cross-frequency that maximizes the characteristics of each group in order to classify signals of two conditions. Coupling and power characteristics can be extracted. The weighted cross-frequency combining information (WCFC) is a method of generating a feature, and the problem can be classified by using the generated feature.

The feature combining unit combines a plurality of extracted features (S105). The model forming unit preforms a machine learning model through the EEG data for learning, and the classification unit classifies the brain state in frequency bands from the features combined in the feature combining unit through the machine learning model (S107). The machine learning model can determine which group the data belongs to according to the measured values of the independent variables through a linear discriminant analysis (LDA), but is not limited thereto.

2 is a flowchart illustrating a method of extracting a feature through weight-based cross frequency information of a feature combining unit according to an embodiment of the present invention.

Referring to FIG. 2 , in the brain state classification method based on the cross-frequency coupling information of the EEG data, the receiver first receives the EEG data from the measuring electrode (S201).

The pre-processing module of the feature extraction unit performs segmentation and pre-processing on the measured EEG data (S203). The pre-processing module performs data segmentation on the measured EEG data, and pre-processes it by passing through a common average reference and a band filter of 1 Hz to 200 Hz. The electroencephalogram data may be divided into 0.5 seconds to 3.5 seconds after stimulation with a window size of 2 seconds, but the present invention is not limited thereto and may vary depending on the data. For each trial and each channel, the frequency power value can be estimated at 1Hz resolution using MATLAB's pwelch() function, and can be used as a feature vector for classification analysis. In particular, classification analysis can be performed by composing feature vectors for each frequency band in order to compare classification accuracy for each frequency band most used in EEG analysis.

The low frequency used in the feature vector construction uses a frequency of 4 Hz from 4 Hz to 40 Hz and a movement size of 2 Hz, and a high frequency can be compared and analyzed using a 20 Hz interval from 40 Hz to 140 Hz and a movement size of 10 Hz, but the present invention is not limited thereto. .

The decomposition module of the feature extraction unit decomposes the pre-processed EEG data into low-frequency and high-frequency envelope signals (S205). The decomposition module can decompose the signal into frequency bands by the following equation.

The covariance matrix generation module of the feature extraction unit generates a covariance matrix of the decomposed signal (S207). The covariance matrix generation module generates a covariance matrix by the following equation.

의

주파수 사이의 상관관계를 나타낸다.The covariance matrix is the a-th column of C, the b-th row C _a,b is a multi-frequency matrix

of

It shows the correlation between frequencies.

를 통해 얻은 고유벡터를 활용하여 가중치벡터를 만든다. 가중치투영행렬모듈(137)은 아래의 수식에 의해 가중치벡터를 생성한다.The weight projection matrix generation module of the feature extraction unit generates a weight projection matrix based on the generated covariance matrix (S209). The weight projection matrix generation module is an objective function

A weight vector is created using the eigenvector obtained through The weight projection matrix module 137 generates a weight vector by the following equation.

The feature extraction module of the feature extraction unit projects the decomposed signal into the generated weight projection matrix to extract features (S211). The feature extraction module extracts features by the following equation.

3 is a block diagram of a brain state classification system using weight-based cross-frequency information in EEG data according to an embodiment of the present invention. The brain state classification system 10 is composed of a brain state classification device 100 and a measuring electrode 200 .

The brain state classification apparatus 100 includes a receiving unit 110 , a control unit 120 , a feature extraction unit 130 , a feature combining unit 140 , a model forming unit 150 , and a classification unit.

The receiver 110 receives the electroencephalogram data from the measurement electrode 200 .

The control unit 120 controls the processing of the process related to the operation of the brain state classification device, and controls the operation of each component.

The feature extracting unit 130 extracts features by receiving the EEG data from the plurality of electrodes by the feature extracting unit. That is, the feature extraction unit 130 may extract weight-based cross-frequency combination information from the EEG data. The feature extraction unit 130 includes a preprocessing module 131 , a decomposition module 133 , a covariance matrix generation module 135 , a weight projection matrix generation module 137 , and a feature extraction module 139 .

The pre-processing module 131 may receive the EEG data measured from the measuring electrode 200 to perform pre-processing.

The decomposition module 133 decomposes the pre-processed EEG data into low-frequency and high-frequency envelope signals. The decomposition module 133 decomposes the EEG data into various frequency bands such as low frequency and high frequency. The decomposition module 133 may decompose a signal into a frequency band by the following equation.

The covariance matrix generation module 135 generates a covariance matrix of the decomposed signal. The covariance matrix generation module 135 performs normalization to adjust the scale of each decomposed signal, and generates a covariance matrix. The covariance matrix generation module generates a covariance matrix by the following equation.

의

주파수 사이의 상관관계를 나타낸다.The covariance matrix is the a-th column of C, the b-th row C _a,b is a multi-frequency matrix

of

It shows the correlation between frequencies.

를 투영(projection)시킬 가중치 벡터(weighted vector)를 만들기 위해 목적함수

를 세운다. The weight projection matrix generation module 137 generates a weight projection matrix based on the generated covariance matrix. Weight projection matrix generation module 137 is a multi-frequency matrix

The objective function to create a weighted vector onto which to project

set up

를 통해 얻은 고유벡터를 활용하여 가중치벡터를 만든다. 가중치투영행렬모듈(137)은 아래의 수식에 의해 가중치벡터를 생성한다.Weight projection matrix generation module 137 is an objective function

A weight vector is created using the eigenvector obtained through The weight projection matrix module 137 generates a weight vector by the following equation.

The weight projection matrix generation module 137 forms a weight projection matrix by using the weight vector. In this case, the eigenvalues and eigenvectors correspond to each class. Therefore, it is possible to create a weight vector by selecting an eigenvector k corresponding to the eigenvalues at both ends.

The feature extraction module 139 extracts features by projecting the decomposed signal into the generated weight projection matrix. The feature extraction module 139 generates a plurality of features by projecting the decomposed signals on a corresponding weight projection matrix.

The feature extraction module 139 extracts features by the following equation.

The feature combining unit 140 combines a plurality of features extracted by the feature extracting unit.

The model forming unit 150 forms a machine learning model in advance through the EEG data for learning.

The classification unit 160 classifies the frequency band from the features combined in the feature combining unit through the formed machine learning model. The classifier can classify the brain state through an already trained machine learning model.

4 to 7 are exemplary diagrams illustrating a brain state classification method using weight-based cross-frequency information in EEG data according to an embodiment of the present invention.

4 is a diagram for explaining a method of extracting weight-based cross-frequency combination information from EEG data. The EEG data of two classes are separated (a), and each covariance matrix is formed for each class (b), based on this, a weight projection matrix is formed (c), and a signal is applied to the weight projection matrix. N features are extracted by projection.

FIG. 5 is a diagram for explaining the principle of weight-based cross-frequency combining information, and it can be seen that when one type of cross-frequency combining PAC is high (alpha-gamma), the correlation is also high.

6 is a view showing the accuracy of each method, and the embodiment according to the present invention showed higher accuracy than the existing methods for classifying brain states using only power information (BrPSD, NrPSD). In addition, it can be seen that CSP using 64 electrodes and WCFC2 (the present invention) using only two electrodes show similar accuracy.

7 is a diagram of a weight projection matrix of the present invention. The WCFC dominant group has higher accuracy than the CSP, and the CSP dominant group has the opposite case. As a result, there was a significant difference between the weighted projection matrices formed at high frequencies. It is analyzed that there was a difference in accuracy between the two groups due to the low frequency, that is, the cross frequency combination information, not the power information. Since the WCFC dominant group is a subject with cross-frequency coupling information, it is analyzed that higher accuracy was obtained by extracting the WCFC-related information along with the power information.

Although the present invention has been described with reference to the embodiment shown in the drawings, which is merely exemplary, those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Accordingly, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

10; brain state classification system 100; brain state classification device
110; receiver 120; control
130; feature extraction unit 140; feature combination
150; model forming unit 160; classification
200; measuring electrode

Claims (5)

In the brain state classification method based on weighted cross-frequency coupling information of EEG data,
(a) receiving the electroencephalogram data from the measuring electrode by the receiving unit;
(b) extracting features from the EEG data by the feature extraction unit through weighted cross-frequency coupling (WCFC);
(c) combining the extracted features by the feature combining unit;
(d) a brain state classification method based on weighted cross-frequency combining information of EEG data, comprising the step of a classification unit classifying brain states in frequency bands from features combined in a feature combining unit. According to claim 1,
Step (b) is,
performing segmentation and pre-processing on the EEG data measured by the pre-processing module of the feature extraction unit;
decomposing the pre-processed EEG data by the decomposition module of the feature extraction unit into low-frequency and high-frequency envelope signals; and
A brain state classification method based on weighted cross-frequency coupling information of EEG data, comprising the step of generating, by the covariance matrix generation module of the feature extraction unit, a covariance matrix of the decomposed signal. 3. The method of claim 2,
The method of classifying a brain state based on weighted cross-frequency coupling information of EEG data further comprising the step of generating, by the weight projection matrix generation module of the feature extraction unit, a weight projection matrix based on the generated covariance matrix. 4. The method of claim 3,
The method of classifying a brain state based on weighted cross-frequency coupling information of EEG data further comprising the step of extracting features by projecting the decomposed signal into the weight projection matrix generated by the feature extraction module of the feature extraction unit. In the brain state classification system based on cross-frequency coupling information of EEG data,
a receiver for receiving electroencephalogram data from the measuring electrode;
a feature extracting unit for extracting features from the electroencephalogram data through weighted cross-frequency coupling (WCFC);
a feature combining unit combining the features extracted by the feature extracting unit; and
A brain state classification system based on weighted cross-frequency combining information of EEG data, including a classification unit that classifies brain states in frequency bands from features combined in the feature combining unit.

Images