A method for transforming EEG signals with different electrode distributions based on zero reference technology

Abstract

This invention discloses a method for converting EEG signals from different electrode distributions based on zero-reference technology, which involves instrument measurement technology, signal processing methods, and neuroscience application fields. The invention solves the problem of inconsistent EEG data electrode distribution in multi-center EEG data analysis, enabling efficient analysis and mining of multi-center EEG data and promoting large-scale EEG application research shared across laboratories. It successfully achieves the conversion of EEG signals from different electrode distributions to a unified electrode distribution (target electrode distribution).

Description

Technical Field

[0001] The present invention relates to instrument measurement technology, signal processing method and neuroscience application field, and is a method for converting brain electrical signals with different electrode distributions based on zero reference technology.

[0002] Background

[0003] EEG is a fundamental technology for noninvasively monitoring brain activity. It can directly record the electrophysiological activity of neuronal populations on a headpiece, providing a variety of important and rich information about brain function. Due to its advantages such as light weight, high compatibility with the environment and systems, low cost, wearability, and wirelessness, EEG has been widely used in various fields such as medicine, psychology, education, computing, and biomedical engineering. EEG electrodes are typically placed on a headpiece according to the standard electrode placement method (i.e., the 10-20 system) specified by the International Electroencephalography Society. The 10-20 system defines electrode placement as follows: the distance from the midpoint of the frontal pole to the root of the nose and the distance from the occipital point to the external occipital protuberance each account for 10% of the total length of these lines, with the remaining points separated by 20% of this total length. Other electrode placement methods also exist, such as the 10-10 electrode distribution system and the HydroCel high-density mesh electrode distribution system. It is worth noting that because different EEG equipment manufacturers have developed different hardware and software systems, in actual use, even if manufacturers claim to use a 10-20 system, their actual electrode distributions may be slightly different. For these reasons, EEG data from different electrode distributions cannot usually be mixed, which also limits EEG research to small-scale, single laboratories.

[0004] In recent years, with the continuous advancement of brain projects around the world, large-scale, open, collaborative, and shared multicenter research has gradually become a new research model for understanding and studying the brain. However, because different research institutions typically use different EEG equipment, the diversity of EEG electrode distributions significantly limits multicenter collaborative research and effective data mining of EEG data, hindering large-scale EEG application research shared across laboratories. Therefore, the present invention provides a method for converting EEG signals from different electrode distributions based on zero-reference technology to address this issue.

[0005] The prior art (Yao DA method to standardize a reference of scalp EEG recordings to a point at infinity [J]. Physiological measurement, 2001, 22 (4): 693) proposed zero reference technology. Zero reference technology is a technology that re-references EEG signals to infinity to obtain ideal EEG signals. However, zero reference technology does not take into account different electrode distributions and can only achieve zero reference conversion on the original electrode distribution. The present invention improves the zero reference technology method theory so that it can achieve conversion of EEG signals to different electrode distributions. Summary of the Invention

[0006] The present invention solves the problem of inconsistent EEG data electrode distribution in multi-center EEG data analysis, enables multi-center EEG data to be effectively analyzed and mined, and promotes large-scale EEG application research shared across laboratories.

[0007] The technical solution adopted by the present invention to achieve the above-mentioned object is: a method for converting EEG signals with different electrode distributions based on zero reference technology, wherein the method comprises:

[0008] Step 1: Use the EEG acquisition system to collect the spontaneous EEG potentials of normal subjects, record the potential and scalp distribution coordinates of each lead, and obtain the distribution coordinates of the original electrode. At the same time, the original EEG signals of each lead are represented by an electrical signal matrix;

[0009] Record the signal waveforms and coordinate distribution of all head surface electrodes, where the electrode signal matrix V Inf , expressed as

[0010]

[0011] Among them G N×K is the lead field matrix of the original electrode coordinates with N electrodes, X K×T It is the real source inside the brain of size K×T; represents the instantaneous potential of lead numbered i at time j, i and j are divided into those that satisfy 1≤i≤N,1≤j≤T, N is the number of electrodes in the electrode coordinate electrode, K is the number of equivalent sources, and T represents the number of points in the time series;

[0012] Step 2: Calculate the average reference potential of the EEG signal under the original electrode distribution to obtain the electrical signal matrix based on the average reference under the original electrode distribution; then subtract the average EEG of all leads at each time point from the original EEG at each time point to obtain the EEG time series matrix V based on the average reference. AVG , the specific formula is as follows:

[0013]

[0014] Among them, w is a column vector of size N×1, each element of which is 1, w T is the transpose of w;

[0015] Step 3: Based on the distribution coordinates of the original electrodes, a head model based on a three-layer sphere and an equivalent distributed source model are established to solve the EEG forward modeling problem and calculate the lead field matrix;

[0016] Step 4: First calculate the MP generalized inverse matrix of the lead field matrix based on the average reference, and then calculate the MP generalized inverse matrix G based on the above steps + AVG And step 2 to obtain the electrical signal matrix V based on the average reference AVG , according to the formula

[0017]

[0018] Calculate the approximate equivalent distributed source of the real source X

[0019] Step 5: According to the target electrode distribution coordinates, establish a head model based on a three-layer sphere and an equivalent distributed source model, solve the EEG forward problem and calculate the lead field matrix

[0020] Step 6: Equivalent distribution source obtained according to step 4 And the lead field matrix of all head surface electrodes in step 5 Calculate the electrical signal matrix based on the infinite reference at the target electrode coordinates

[0021]

[0022] Furthermore, the specific method of step 3 is:

[0023] Step S3-1: Select a three-layer concentric sphere model to model the head;

[0024] The head model is built using a three-layer concentric sphere model. A Cartesian coordinate system is established with the midpoint of the left and right ears as the origin, with the line from the origin to the right earlobe as the positive X-axis, the line from the origin to the root of the nose as the positive Y-axis, and the positive Z-axis perpendicular to the XY plane upward. Based on this coordinate system, the relative radius sizes of the cerebral cortex and the inner and outer sides of the skull, as well as the relative conductivity, are measured. The electrode distribution coordinates after removing bad conductors need to be normalized and distributed on the head model. The three-layer concentric sphere model is a spherically symmetric conductor model, which simplifies the head shape into a spherical conductor composed of three concentric spheres. The three concentric spheres represent the inner radius of the skull, the outer radius of the skull, and the radius of the head from the inside to the outside, respectively.

[0025] Step S3-2: Select a dipole source model as the equivalent distributed source. This model uses 6144 dipoles to approximate the distribution of real sources inside the brain.

[0026] Step S3-3, normalizing the distribution coordinates of the head surface electrodes and distributing them on the head model;

[0027] Step S3-4, calculating the lead field matrix based on the forward theory of spherical harmonic spectrum;

[0028] According to the head model established in step S3-1, the equivalent distributed source model established in step S3-2 and the electrode distribution coordinates normalized in step S3-3, and based on the forward theory of spherical harmonic spectrum, the lead field matrix based on the average reference of the original electrode coordinates is calculated.

[0029] The present invention successfully realizes the transformation of EEG signals of different electrode distributions into a unified electrode distribution (target electrode distribution). BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of the EEG electrode coordinate conversion method of the present invention;

[0031] Figure 2 This is a flow chart of lead field matrix calculation of the present invention;

[0032] Figure 3 The head model based on three layers of concentric spheres of the present invention;

[0033] Figure 4 This is an electrode coordinate conversion effect diagram of the present invention;

[0034] Figure 5 Schematic diagram of the method principle of the present invention. DETAILED DESCRIPTION

[0035] Figure 1 The present invention is a schematic flow chart of the method, wherein the specific method steps are as follows:

[0036] Step S1: Record the signal waveforms and coordinate distribution of all head-surface electrodes.

[0037] Record the signal waveforms and coordinate distribution of all head surface electrodes, where the electrode signal matrix (V Inf ), expressed as

[0038]

[0039] where G is the lead field matrix of the original electrode coordinates with N number of electrodes, and X is the real source inside the brain with size K×T. It represents the instantaneous potential of lead numbered i at time j, i and j are divided into those that satisfy 1≤i≤N,1≤j≤T, N is the number of electrodes in the electrode coordinate electrode, K is the number of equivalent sources, and T represents the number of points in the time series.

[0040] Step S2: Calculate the average reference potential (V AVG ).

[0041] Specifically, first calculate the average value v of all leads at each time point AR , the formula is as follows:

[0042]

[0043] Where w is a unit column vector of size N×1, each element of which is 1, w T is the transpose of w.

[0044] Then, the average EEG value of all leads at each time point is subtracted from the original EEG at each time point to obtain the EEG time series matrix based on the average reference (V AVG ), the specific formula is as follows:

[0045]

[0046] Step S3, calculate the lead field matrix (G AVG );

[0047] The lead field matrix G of the original coordinates in the present invention AVG It is calculated by the three-layer head model, equivalent source model and original electrode distribution. The specific calculation process is as follows Figure 2 As shown, the following steps are included:

[0048] Step S3-1: Select a three-layer concentric sphere model to model the head.

[0049] Specifically, the three-layer concentric sphere model is a spherically symmetric conductor model, which simplifies the head shape into a spherical conductor composed of three concentric spheres. The three concentric spheres represent the inner radius S311 of the skull, the outer radius S312 of the skull, and the head radius S313 from the inside to the outside, respectively. Figure 3 As shown, the corresponding radii, calculated by normalizing the head radius, are 0.87 (S411), 0.92 (S412), and 1.0 (S413). Furthermore, the electrical conductivity is uniformly distributed, with the relative conductivity of the cerebral cortex and scalp both being 1.0, and the relative conductivity of the skull being 0.0125.

[0050] Step S3-2: Select a dipole source model as the equivalent distributed source.

[0051] Specifically, the spatial structure of the dipole source model is a triangular network generated by a standard brain template as an EEG dipole source, which contains a total of 6144 radial dipoles, which is equivalent to a total of 6144 equivalent distributed sources.

[0052] Step S3-3: normalize the distribution coordinates of the head surface electrodes and distribute them on the head model.

[0053] Specifically, the distribution coordinates of the head surface electrodes in step S2 are normalized and distributed on the head model.

[0054] Step S3-4, calculating the lead field matrix based on the forward theory of spherical harmonic spectrum.

[0055] Specifically, according to the head model established in step S3-1, the equivalent distributed source model established in step S3-2, and the electrode distribution coordinates normalized in step S3-3, the lead field matrix (G AVG ). In this embodiment, in order to simplify the calculation process, a custom code is used to implement the calculation of the lead field matrix.

[0056] Step S4, calculate the equivalent distributed source

[0057] In order to obtain an equivalent distributed source It is necessary to first calculate the lead field matrix based on the average reference (G AVG )'s MP generalized inverse matrix (G + AVG ); Then, based on the above steps, the MP generalized inverse matrix (G + AVG ) and step S2 to obtain the electrical signal matrix (V AVG ), according to the formula

[0058]

[0059] Calculate the approximate equivalent distributed source of the real source (X)

[0060] Step S5: Calculate the lead field matrix of the target electrode coordinates

[0061] The lead field matrix of the target electrode coordinates can be calculated by following the process similar to step S3. Simply change the input of step S3 to the distribution coordinates of the target electrode, keep other places unchanged, and then calculate again to get the lead field matrix of the target electrode.

[0062] Step S6, calculating the target electrode electrical signal matrix based on the infinite point reference after the electrode coordinate transformation

[0063] The equivalent distribution source obtained in step S5 is known and the lead field matrix of all head surface electrodes in step S6 Substitution

[0064]

[0065] The calculation can be performed to obtain the electrical signal matrix of the target electrode based on the infinite point reference after the electrode coordinate transformation.

[0066] Reconstruction effect

[0067] During the electrode coordinate conversion process, the EEG signals of electrodes with the same name may change. The reasons are as follows: 1) The electrode coordinates of the same electrode name in the original electrode distribution and the target electrode distribution are different.

[0068] 2) The approximations and limitations of the method itself, such as the use of 6144 dipoles to approximate the equivalent source in the brain and the use of the generalized inverse matrix when calculating the approximate equivalent source in step S4.

[0069] Therefore, it is necessary to examine the changes in EEG and quantitative EEG after electrode coordinate conversion. Figure 4 The conversion effect of electrode distribution is demonstrated from two aspects: EEG and quantitative EEG. Figure 4 (A) is a comparison of the EEG signals of the FP1 lead before and after the conversion of the left electrode of the same subject. The black line and red line in the figure are the EEG signals of the FP1 lead from 0 to 500ms under the original coordinates (64 leads) and the target electrode coordinates (10-20 system 61 leads, 10-20-cap61), respectively. It can be seen that the EEG waveforms before and after the electrode coordinate conversion are almost identical. Figure 4 (B) shows the power spectrum topography of the original EEG electrode distribution and the target electrode distribution for the same subject. The columns represent frequency bands, and the rows represent electrode distribution. It can be seen that the reconstructed EEG power spectrum is essentially consistent with the original EEG power spectrum. This figure demonstrates that this method can significantly restore EEG signals and preserve EEG information while transforming the electrode coordinate distribution.

[0070] Figure 5 This is a schematic diagram of the method principles of the present invention. In step S71, an equivalent distributed source is calculated from actual scalp potential recordings using a certain point on the head or the average potential as a reference electrode. In step S72, the lead potential referenced to a point at infinity is reconstructed using an equivalent distributed source that approximates the true source.

[0071] In the field of neuroscience, process S71 is known as inversion, and process S72 is known as forward modeling. The actual calculation steps of these two processes are related to the selected head model, the equivalent distributed source, and the distribution coordinates of the head surface electrodes. In fact, in the field of neuroscience, there are many ways to model EEG power sources, and different modeling methods correspond to different forward and inversion methods. The present invention focuses on reconstructing the bad conduction potential using the EEG inversion and forward modeling processes, so the precise location of the EEG power source is not of concern. Therefore, the modeling of EEG power sources in the present invention only needs to provide one of the ideas, which will not be elaborated in detail here.

[0072] In this embodiment, after executing step S2 and before executing step S3, in order to eliminate the influence of interference sources, the EEG signal needs to be preprocessed. Specifically, it is first passed through a bandpass filter with a frequency band of 0.01-100 Hz, and then a notch filter with a frequency band of 49-51 Hz is used. Next, independent component analysis (ICA) is performed to remove interference components such as artifacts in the EEG signal to ensure that a clean EEG signal is obtained.

[0073] In fact, the above-mentioned EEG data preprocessing methods and processes are not unique, but no matter which process is used, it is sufficient to ensure that a clean EEG signal waveform can be obtained after the preprocessing is completed.

[0074] The present invention requires that the number of electrodes in the target electrode distribution be less than or equal to the number of electrodes in the original electrode distribution. Because EEG signal reconstruction is based on the original EEG signal, if the number of electrodes in the target electrode distribution is greater than the number of electrodes in the original coordinates, although the number of electrodes in the converted EEG signal increases, the amount of truly useful information does not increase.

[0075] In step S2 of the present invention, a potential matrix based on other single-point references can also be calculated. For example, a potential matrix based on the earlobe reference point can be calculated, and the earlobe potential is used instead of the average reference potential.

[0076] After executing step S2 and step S3, it is recommended to preprocess the EEG signal (e.g., removing bad conductors, bandpass filtering, artifact removal, etc.) to eliminate the influence of interference sources to obtain a clean EEG signal. The above-mentioned EEG data preprocessing method and process are not unique, but no matter which process is used, it is sufficient to ensure that a clean EEG signal waveform can be obtained after the preprocessing is completed.

[0077] In step S31 of the present invention, the three-layer concentric sphere model is selected to model the head, and other models can also be selected to model the head, for example, a real head model can be used to model the head.

[0078] The present invention also provides a method for converting from electrode reference or average reference to infinite reference. Specifically, in addition to converting the electrode coordinate distribution, step S6 actually converts the EEG signal based on the average reference or other reference into a signal based on infinite reference.

[0079] The method for calculating the MP generalized inverse matrix of the lead field matrix based on the average reference in step S4 of the present invention is not unique; specifically, the MP generalized inverse matrix can also be calculated using the maximum rank decomposition method.

Claims

1. A method for transforming EEG signals with different electrode distributions based on zero reference technology, wherein: The method comprises: Step 1: Use the EEG acquisition system to collect the spontaneous EEG potentials of normal subjects, record the potential and scalp distribution coordinates of each lead, and obtain the distribution coordinates of the original electrode. At the same time, the original EEG signals of each lead are represented by an electrical signal matrix; Record the signal waveforms and coordinate distribution of all head surface electrodes, where the electrode signal matrix V Inf , expressed as Among them G N×K is the lead field matrix of the original electrode coordinates with N electrodes, X K×T It is the real source inside the brain of size K×T; represents the instantaneous potential of lead numbered i at time j, i and j are divided into those that satisfy 1≤i≤N,1≤j≤T, N is the number of electrodes in the electrode coordinate electrode, K is the number of equivalent sources, and T represents the number of points in the time series; Step 2: Calculate the average reference potential of the EEG signal under the original electrode distribution to obtain the electrical signal matrix based on the average reference under the original electrode distribution; then subtract the average EEG of all leads at each time point from the original EEG at each time point to obtain the EEG time series matrix V based on the average reference. AVG , the specific formula is as follows: Among them, w is a column vector of size N×1, each element of which is 1, w T is the transpose of w; Step 3: Based on the distribution coordinates of the original electrodes, a head model based on a three-layer sphere and an equivalent distributed source model are established to solve the EEG forward modeling problem and calculate the lead field matrix; Step 4: First calculate the MP generalized inverse matrix of the lead field matrix based on the average reference, and then calculate the MP generalized inverse matrix G based on the above steps + AVG And step 2 to obtain the electrical signal matrix V based on the average reference AVG , according to the formula Calculate the approximate equivalent distributed source of the real source X Step 5: According to the target electrode distribution coordinates, establish a head model based on a three-layer sphere and an equivalent distributed source model, solve the EEG forward problem and calculate the lead field matrix Step 6: Equivalent distribution source obtained according to step 4 And the lead field matrix of all head surface electrodes in step 5 Calculate the electrical signal matrix based on the infinite reference at the target electrode coordinates 2. The method for converting EEG signals with different electrode distributions based on zero reference technology as claimed in claim 1, characterized in that: The specific method of step 3 is: Step S3-1: Select a three-layer concentric sphere model to model the head; The head model is built using a three-layer concentric sphere model. A Cartesian coordinate system is established with the midpoint of the left and right ears as the origin, with the line from the origin to the right earlobe as the positive X-axis, the line from the origin to the root of the nose as the positive Y-axis, and the positive Z-axis perpendicular to the XY plane upward. Based on this coordinate system, the relative radius sizes of the cerebral cortex and the inner and outer sides of the skull, as well as the relative conductivity, are measured. The electrode distribution coordinates after removing bad conductors need to be normalized and distributed on the head model. The three-layer concentric sphere model is a spherically symmetric conductor model, which simplifies the head shape into a spherical conductor composed of three concentric spheres. The three concentric spheres represent the inner radius of the skull, the outer radius of the skull, and the radius of the head from the inside to the outside, respectively. Step S3-2: Select a dipole source model as the equivalent distributed source. This model uses 6144 dipoles to approximate the distribution of real sources inside the brain. Step S3-3, normalizing the distribution coordinates of the head surface electrodes and distributing them on the head model; Step S3-4, calculating the lead field matrix based on the forward theory of spherical harmonic spectrum; According to the head model established in step S3-1, the equivalent distributed source model established in step S3-2 and the electrode distribution coordinates normalized in step S3-3, and based on the forward theory of spherical harmonic spectrum, the lead field matrix based on the average reference of the original electrode coordinates is calculated.

Images