A method and system for analyzing brain-muscle coupling specificity based on conditional covariance

Abstract

The application discloses a brain-muscle coupling specificity analysis method and system based on conditional covariance, which comprises the following steps: firstly, collecting the brain electrical signals and the electromyographic signals of a subject, outputting the brain electrical sequence and the electromyographic sequence of each candidate muscle after pretreatment, forming the restricted joint variable and the complete joint variable, and calculating the restricted joint covariance and the complete joint covariance; secondly, based on the restricted joint covariance and the complete joint covariance, calculating the restricted joint covariance after Riemann aggregation and the complete joint covariance after reconstruction, and obtaining the specificity incremental contribution of the current candidate muscle; finally, obtaining the corresponding specificity incremental contribution of each candidate muscle, and according to the candidate muscle sequence, collecting the specificity incremental contributions, and forming the specificity sequence of the candidate muscle from high to low as the specificity analysis result. The application provides a repeatable quantitative result for comparing the relative incremental information provided by the real prior brain electrical signals on the activities of different candidate muscles.

Description

Technical Field

[0001] This invention belongs to the field of electroencephalogram (EEG) and electromyogram (EMG) signal processing technology, specifically relating to a brain-muscle coupling specificity analysis method and system based on conditional covariance. Background Technology

[0002] Electroencephalography (EEG) reflects the electrical activity of the cerebral cortex, while electromyography (EMG) reflects the electrical activity generated by muscle contraction. When subjects perform limb movements, there is a measurable temporal correlation between motor-related EEG activity and peripheral muscle activity. Simultaneous recording of EEG signals and EMG signals from multiple candidate muscles is used to analyze the functional connection between central nervous system activity and candidate muscle activity. Existing brain-muscle coupling analysis mainly uses methods such as brain-muscle coherence, mutual information, transfer entropy, and directional prediction based on restricted and intact models to characterize the synchronization relationship, statistical dependence, or directional information transmission between EEG and EMG signals.

[0003] However, larger brain-muscle correlation, covariance, or directional prediction results do not equate to prior EEG providing additional predictive information. The current activity of candidate muscles is usually influenced by their past activity, and different muscles have varying degrees of predictability. Without controlling for EEG history vectors, the observed brain-muscle connections will also include the effect of the muscle's own temporal continuity. Therefore, it is necessary to compare the conditional residual variance under the same current EEG values ​​and EEG history vectors, with and without prior EEG history vectors, to determine whether the prior EEG history vector still provides additional predictive information.

[0004] The above comparison also requires that the restricted joint covariance remain consistent before and after adding the prior EEG history vector. Limb movements typically involve multiple stages, and the covariance formed by the joint variables of each stage, after shrinking and positive definite transformation, can be used as a symmetric positive definite matrix for cross-stage Riemann aggregation. If the restricted joint covariance and the complete joint covariance are Riemann aggregated independently, the principal submatrix corresponding to the restricted joint variables in the complete joint covariance after independent Riemann aggregation may no longer be equal to the restricted joint covariance after Riemann aggregation. In this case, the change in conditional residual variance not only reflects the addition of the prior EEG history vector but may also be mixed with the differences caused by the two independent Riemann aggregations. Therefore, it is necessary to maintain the consistency of the restricted joint covariance after cross-stage Riemann aggregation.

[0005] Furthermore, under limited sample conditions, even if there is no correct temporal or individual correspondence between EEG and the current candidate muscle, factors such as common trends and similar spectral structures may still cause a decrease in conditional residual variance. If cyclically shifted EEG, future-shifted EEG, time-reversed EEG, phase-randomized EEG, or cross-subject EEG can still produce a large substitution increment contribution, then this substitution increment contribution cannot be attributed to a correct correspondence between the true prior EEG and the current candidate muscle. Therefore, current technology still lacks a method that can simultaneously utilize EEG history vectors to control the temporal continuity of the candidate muscle itself, maintain the consistency of restricted joint covariance, and complete the calibration of the five types of substitution EEG with the maximum substitution increment contribution. Consequently, it is difficult to accurately determine whether the true prior EEG can provide additional predictive information for candidate muscle activity. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a method and system for brain-muscle coupling specificity analysis based on conditional covariance. The method sequentially performs synchronous data processing, construction of joint variables and joint covariance, conditional covariance decomposition, cross-stage Riemannian aggregation and structure-preserving reconstruction, determination of real-prior EEG increment contributions and five types of alternative EEG calibration, and candidate muscle-specific ranking.

[0007] In one aspect, this invention provides a brain-muscle coupling specificity analysis method based on conditional covariance, the specific steps of which are as follows: Step 1: Synchronous Data Processing. Simultaneously collect EEG signals and EMG signals from at least two candidate muscles during the subject's limb movement. Map the EEG and EMG signals to a common time axis and unify them to a preset analysis sampling rate to obtain time-aligned EEG sequences and EMG sequences for each candidate muscle within the common analysis interval. Based on preset temporal boundaries, exclude positions at both ends of the sequence that cannot form complete temporal samples, determine common effective sampling positions, and define the action stage boundaries according to a unified action analysis time window.

[0008] Step 2: Construction of Joint Variables and Joint Covariance. Using the output of Step 1 as input, construct an electromyography (EMG) history vector and a prior EEG history vector for each candidate muscle. Combine the current EMG value and the EMG history vector to form a restricted joint variable. Add the prior EEG history vector to the restricted joint variable to form a complete joint variable. Subsequently, calculate the complete joint covariance only once using the complete joint variable sample at each action stage, and extract the principal submatrix corresponding to the restricted joint variable from the same complete joint covariance as the restricted joint covariance.

[0009] Step 3: Conditional Covariance Decomposition, Cross-Stage Riemann Convergence, and Structure-Preserving Reconstruction. Using the restricted joint covariance and complete joint covariance of each action stage output from Step 2 as input: perform block decomposition and conditional covariance decomposition to obtain the cross-reconstruction coefficient matrix and the preceding EEG conditional covariance; perform cross-stage Riemann convergence on the restricted joint covariance and the preceding EEG conditional covariance under shared time weights, and linearly weight the cross-reconstruction coefficient matrix; perform structure-preserving reconstruction based on the above conditional covariance decomposition relationship. Step 3 outputs the Riemann-converged restricted joint covariance and the reconstructed complete joint covariance, which are then used as input for Step 4.

[0010] Step 4: True Prior EEG Incremental Contribution and Five Types of Alternative EEG Calibration. Using the restricted joint covariance after Riemann aggregation and the reconstructed complete joint covariance output from Step 3, calculate the restricted conditional residual variance and the complete conditional residual variance to obtain the true prior EEG incremental contribution. Construct five types of alternative EEGs, obtaining five alternative incremental contributions for each using the same process. Subtract the largest alternative incremental contribution from the true prior EEG incremental contribution to obtain the specific incremental contribution of the current candidate muscle.

[0011] Step 5: Candidate Muscle Specificity Ranking. Steps 2 through 4 are performed on each candidate muscle to obtain the corresponding specificity increment contribution. These specificity increment contributions are then summarized according to a pre-defined candidate muscle order, and ranked from highest to lowest to form the candidate muscle specificity ranking, which serves as the specificity analysis result. Step 5 ultimately outputs the candidate muscle name, specificity increment contribution, and candidate muscle specificity ranking; the candidate muscle specificity ranking represents the relative specificity of different candidate muscles to the real-time EEG under the same action background.

[0012] On the other hand, the present invention also provides a brain-muscle coupling specificity analysis system based on conditional covariance, comprising the following modules: The output acquisition module is used to simultaneously acquire the subject's electroencephalogram (EEG) and electromyogram (EMG) signals, and outputs time-aligned EEG sequences and EMG sequences of each candidate muscle after preprocessing.

[0013] The joint covariance module forms restricted joint variables and complete joint variables based on the time-aligned EEG sequences and the electromyography sequences of each candidate muscle, and calculates the restricted joint covariance and complete joint covariance accordingly.

[0014] The joint covariance reconstruction module calculates and outputs the restricted joint covariance and the reconstructed full joint covariance after Riemann aggregation, based on the restricted joint covariance and the full joint covariance.

[0015] The specific incremental contribution module uses the restricted joint covariance after Riemann aggregation and the reconstructed complete joint covariance to calculate the conditional residual variance and obtain the specific incremental contribution of the current candidate muscle.

[0016] The specificity analysis module is used to obtain the specific incremental contribution corresponding to each candidate muscle, summarize the specific incremental contributions according to the order of candidate muscles, and form a specificity ranking of candidate muscles from high to low, which is used as the specificity analysis result.

[0017] The present invention has the following beneficial effects: 1. This invention constructs a restricted joint variable using electromyography (EMG) history vectors, and adds an earlier electroencephalogram (EEG) history vector to construct a complete joint variable based on the same EMG history vector. This allows the actual contribution of the earlier EEG increment to represent the predictive information further provided by the earlier EEG history vector after the predictive effect of the EMG history vector is controlled.

[0018] 2. This invention calculates the complete joint covariance only once based on the complete joint variables, and directly extracts the restricted principal matrix as the restricted joint covariance, so that the restricted joint covariance of each action stage is always consistent with the corresponding restricted principal matrix in the complete joint covariance, avoiding the deviation of the principal matrix relationship caused by separate covariance estimation and linear shrinkage.

[0019] 3. This invention addresses the problem that the original principal matrix relationship may be destroyed after nonlinear Riemann aggregation of restricted joint covariance and complete joint covariance. It obtains the cross-reconstruction coefficient matrix and the prior EEG conditional covariance through conditional covariance decomposition, performs cross-stage Riemann aggregation and linear weighting under shared time weights, and then performs structure-preserving reconstruction, so that the reconstructed complete joint covariance still strictly retains the restricted joint covariance after Riemann aggregation as its restricted principal matrix.

[0020] 4. This invention utilizes cyclic shift EEG, future shift EEG, time-reversal EEG, phase randomization EEG, and cross-subject EEG to obtain five types of substitution increment contributions, and takes the maximum value among them as the maximum substitution increment contribution; the specific increment contribution is obtained by subtracting the maximum substitution increment contribution from the real advance EEG increment contribution, so that a positive value clearly indicates that the real advance EEG increment contribution is higher than all five types of substitution increment contributions.

[0021] 5. This invention calculates the specific incremental contribution of multiple candidate muscles during the same action process, and under unified EEG processing conditions, unified action stage and unified calculation process, the candidate muscles are ranked from high to low according to the specific incremental contribution, which serves as the specific analysis result. This provides a repeatable quantitative result for comparing the relative incremental information provided by real preceding EEG to the activity of different candidate muscles. Attached Figure Description

[0022] Figure 1 The overall flowchart of the brain-muscle coupling specific analysis method; Figure 2 A schematic diagram is constructed for the joint variables and joint covariance; Figure 3 A schematic diagram of conditional covariance decomposition, cross-stage Riemannian aggregation, and structure-preserving reconstruction. Figure 4 A schematic diagram of five types of alternative EEG structures; Figure 5 A schematic diagram of the contribution of real-prior EEG increments, five types of alternative EEG calibrations, and candidate muscle-specific ranking. Detailed Implementation

[0023] In one aspect, this invention provides a brain-muscle coupling specificity analysis method based on conditional covariance, such as... Figures 1 to 5 As shown, the specific steps are as follows: Step 1: Simultaneously collect EEG signals and EMG signals from at least two candidate muscles during the subject's limb movement. Map the EEG and EMG signals to a common time axis and unify them to a preset analysis sampling rate to obtain a time-aligned EEG sequence and EMG sequences for each candidate muscle within the common analysis interval. Eliminate positions at both ends of the sequence that cannot form complete time-series samples based on preset temporal boundaries, determine the common effective sampling positions, and determine the action stage boundaries according to a unified action analysis time window. Step 1 outputs the time-aligned EEG sequence, EMG sequences for each candidate muscle, the common effective sampling positions, and the action stage boundaries, which are then used as input for Step 2.

[0024] Step 2: Using the output of Step 1 as input, construct an electromyography (EMG) history vector and a prior electroencephalogram (EEG) history vector for each candidate muscle. Combine the current EMG value with the EMG history vector to form a restricted joint variable. Add the prior EEG history vector to the restricted joint variable to form a complete joint variable. Then, at each action stage, calculate the complete joint covariance only once using samples of the complete joint variable, and extract the principal submatrix corresponding to the restricted joint variable from the same complete joint covariance as the restricted joint covariance. Step 2 outputs the restricted joint covariance and complete joint covariance for each candidate muscle at each action stage, and uses this as input for Step 3.

[0025] Step 2.1: Using the candidate muscle electromyography (EMG) sequences output in Step 1, select consecutive EMG sampling values ​​before each common valid sampling location to construct an EMG history vector: In formula (1), Indicates the first The candidate muscles in the first The electromyographic history vector corresponding to each current sampling position; Indicates the first The candidate muscle before the current sampling position. Electromyography (EMG) sampling values ​​at each sampling location, of which Take 1, 2, ... in sequence ; This indicates the electromyography (EMG) sampling value. Indicates the candidate muscle number. Indicates the current sampling position number. Indicates the electromyographic history order. (The text within square brackets is missing.) The electromyographic scalars are arranged horizontally in chronological order from most recent to oldest, with the top right corner showing the scalars. Indicates transpose, therefore Vie It is a column vector. Formula (1) yields... It can be used in both formula (3) and formula (4).

[0026] Step 2.2: At the same common valid sampling location as in Step 2.1, using the time-aligned EEG sequence output in Step 1, sample from multiple preset advance delay positions preceding the current sampling location, and construct an advance EEG history vector according to the advance delay order: In formula (2), Indicates the first The preceding EEG history vector corresponding to the current sampling location; This indicates that the fixed motor-related EEG channel set is in the first... A multi-channel EEG column vector is formed by sampling locations, with each EEG channel arranged in a fixed order; Indicates the first The number of sampling points with a preset lead time delay. Take 1, 2, ... in sequence , This represents the total number of preceding delays. The multi-channel EEG column vectors corresponding to each preceding delay position are arranged sequentially according to the preceding delay order, forming the preceding EEG history column vector. Formula (2) yields... Only the complete joint variables are entered.

[0027] The current electromyographic value and the electromyographic history vector obtained by formula (1) constitute a restricted joint variable. Then, the prior electroencephalographic history vector obtained by formula (2) is added to this restricted joint variable to form a complete joint variable: In formula (3), Indicates the first The candidate muscles in the first Constrained joint variables at each sampling location, superscript Indicates restriction; This is the current electromyography (EMG) value. This is the electromyography history vector obtained from formula (1). Formula (3) sets the current electromyography value... Placed in electromyographic history vector The components together form a column vector. . In formula (4), Represents a complete joint variable for the same candidate muscle and the same sampling location, with superscript indicating the variable. Indicates completeness; For prior EEG history vectors. Formula (4) restricts joint variables. The components and prior EEG history vector The components are arranged in the order shown in the formula, together forming a column vector. Therefore, with compared to, Only the prior EEG history vector was added. The components included. The complete joint variables obtained by formula (4) are used in step 2.3 to calculate the complete joint covariance.

[0028] Step 2.3, as follows Figure 2 As shown, within the common effective sampling location and action phase boundary output in step 1, complete joint variable samples of formula (4) are collected according to the action phase. The sample covariance is calculated and linearly shrunk to improve the numerical stability of covariance estimation under finite sample conditions and to facilitate obtaining the positive definite covariance matrix required for subsequent Riemann aggregation, thus obtaining the complete joint covariance: In formula (5), Indicates the first The candidate muscles in the first The complete joint covariance of each action phase after linear contraction; The sample covariance is calculated from the complete joint variable sample of the same candidate muscle and the same action phase using formula (4). This represents the linear contraction coefficient, with a value ranging from 0 to 1; and They represent the sample covariance respectively. The weights relative to the target of the identity matrix contraction; Represents the matrix trace operation; Represents a complete joint variable dimensionality; express An identity matrix of order 1; Indicates the candidate muscle number. Indicates the sequence number of the action phase, superscript This indicates completeness. Formula (5) yields... As the complete joint covariance of this action phase, it is used in formula (6) to extract the restricted joint covariance.

[0029] In formula (6), Indicates the first The first candidate muscle, the first Constrained joint covariance of each action phase, superscript Indicates restriction; The complete joint covariance obtained from formula (5); This represents the set of coordinate positions of the restricted joint variable within the complete joint variable; This indicates that the variances are selected simultaneously from the complete joint covariance. The principal submatrix formed by the corresponding rows and columns. Formula (6) is directly derived from the same complete joint covariance of Formula (5). No second covariance estimation is performed. Step 2 final output and And use it as input for step 3.

[0030] Step 3: Using the restricted joint covariance and complete joint covariance of each action stage output in Step 2 as input: Step 3.1 Perform block decomposition and conditional covariance decomposition to obtain the cross-reconstruction coefficient matrix and the prior EEG conditional covariance; Step 3.2 Perform cross-stage Riemann aggregation on the restricted joint covariance and the prior EEG conditional covariance under shared time weights, and perform linear weighting on the cross-reconstruction coefficient matrix; Step 3.3 Perform structure-preserving reconstruction based on the above conditional covariance decomposition relationship, such as... Figure 3 As shown. Step 3 outputs the restricted joint covariance after Riemann aggregation and the complete joint covariance after reconstruction, and uses them as input for step 4.

[0031] Step 3.1: Perform block decomposition and conditional covariance decomposition on the complete joint covariance for each action phase. Following the variable order of "restricted joint variables—prior EEG history vector," divide the complete joint covariance output from Step 2 into blocks: In formula (7), The output of step 2 The first candidate muscle, the first The complete joint covariance of each action phase; For constrained joint covariance; This is the cross-covariance matrix between the restricted joint variables and the prior EEG history vectors, with rows corresponding to the restricted joint variable dimension and columns corresponding to the prior EEG history vector dimension. for transpose; For the preceding EEG covariance block, superscript This represents the EEG portion. Formula (7) divides the same complete joint covariance into restricted joint covariance. Cross covariance matrix and preceding EEG covariance blocks The three are used for the conditional covariance decomposition of formula (8): In formula (8), Indicates the first The first candidate muscle, the first The cross-reconstruction coefficient matrix of each action phase; by It can be seen that this matrix, together with the restricted joint covariance, determines the cross covariance matrix. In numerical implementation, this result is obtained by solving linear equations, without explicitly forming the cross covariance matrix. . This represents the prior EEG conditional covariance, derived from the prior EEG covariance block. Subtract the portion that can be explained by the restricted joint variables Obtain, and with They have the same EEG dimension. Through the above decomposition, the restricted joint covariance and the prior EEG conditional covariance can be used as positive definite covariance matrices for cross-stage Riemann aggregation, and the cross-reconstruction coefficient matrix can be linearly weighted using the same shared time weight, thereby providing input for subsequent structure-preserving reconstruction. Equation (8) output and and the same action phase Both serve as inputs for step 3.2.

[0032] Step 3.2: Set shared time weights and perform cross-stage Riemann aggregation and linear weighting. Set a set of shared time weights for each stage of the same action process: In formula (9), Indicates the first Shared time weights for each action phase Indicates the sequence number of the action phase. Indicates the total number of action stages; This indicates that the weights of each shared time are non-negative. This indicates that the sum of all shared time weights is 1. Formula (9) defines the same group. Used for formula (10) processing , and : In formula (10), Indicates the first Candidate muscle span Constrained joint covariance after Riemann convergence of each action phase; This represents the prior EEG conditional covariance after Riemannian convergence; This represents the cross-reconstruction coefficient matrix after linear weighting; For the first Shared time weights for each action phase This represents the total number of action phases. From formula (6), and From formula (8); The matrix logarithm of a symmetric positive definite matrix. Represents the symmetric matrix exponent. Formula (10) outputs the above three cross-stage quantities and serves as the input for step 3.3.

[0033] Step 3.3: Perform structure-preserving reconstruction using the three cross-stage quantities obtained in Step 3.2. First, reconstruct the cross-covariance matrix and the preceding EEG covariance block, then construct the reconstructed complete joint covariance: In formula (11), Indicates the first The cross-covariance matrix of the reconstructed candidate muscles is given by formula (10). and Multiplying them together yields the result. This represents the reconstructed anterior EEG covariance block, which is the Riemann-aggregated anterior EEG conditional covariance. and Together they form a whole. The rows correspond to the restricted joint variable dimension, and the columns correspond to the preceding EEG history vector dimension. Complete the opposite row and column correspondence. Formula (11) yields and Enter formula (12) and... Together they constitute the complete joint covariance after reconstruction. : The following structure preservation relationship is obtained from the block reconstruction form of formula (12): In formula (13), The first one obtained from formula (12) The complete joint covariance of each candidate muscle after reconstruction; This represents the set of positions of the restricted joint variable within the complete joint variable; This indicates selecting from the reconstructed complete joint covariance. The corresponding row and column bounded principal submatrices are obtained; The constrained joint covariance after Riemann aggregation is obtained by formula (10). Formula (13) shows that the reconstructed full joint covariance is directly preserved. As its restricted principal submatrix, this relation does not introduce new computational costs. Because The covariance of the preceding EEG conditions after Riemann convergence remains positive definite, and the covariance of the reconstructed complete joint EEG remains positive definite. Step 3 Output and This is used as input for calculating the conditional residual variance in step 4.

[0034] Step 4.1 Calculate the restricted conditional residual variance and the complete conditional residual variance using the restricted joint covariance after Riemann aggregation and the reconstructed complete joint covariance output from Step 3, and obtain the true prior EEG increment contribution. Step 4.2 Construct five types of surrogate EEGs, obtain five surrogate increment contributions for each using the same process, and subtract the largest surrogate increment contribution from the true prior EEG increment contribution to obtain the specific increment contribution of the current candidate muscle. Step 4 outputs the specific increment contribution and uses it as input for Step 5.

[0035] Step 4.1: Calculate the conditional residual variance based on the two types of covariance output in Step 3. Both formulas (3) and (4) place the current EMG value as the first component of the joint variable; therefore, both types of covariance can be divided according to the order of "current EMG value - predictor variable". The predictor variable under the constrained condition is the EMG history vector, while the predictor variable under the complete condition is the EMG history vector and the prior EEG history vector. The conditional residual variance is calculated using the following formula: In formula (14), Represents a column vector of given predictor variables Current electromyography value Conditional residual variance; For prediction Historical information exists under both conditions. Under restricted joint variables Only including electromyographic history vectors, under complete joint variables This includes both electromyographic history vectors and prior electroencephalographic history vectors. During calculation, they are respectively... and corresponding The joint covariance after Riemannian aggregation is divided into two parts. ;in, The variance of the corresponding row and column is , The row and The covariance corresponding to the column is , The row and The covariance corresponding to the column is , The covariance matrix corresponding to the row and column is , This represents its inverse; in numerical implementation, the corresponding product is achieved by solving a linear equation. Substituting the above block divisions into formula (14), we get ;Will Substituting the corresponding blocks into formula (14), we get ; Calculate the contribution of real preceding EEG increments: In formula (15), Indicates the first The actual leading EEG increment contribution of each candidate muscle; Indicates the candidate muscle number; Represents the natural logarithm; and These are the variances of the constrained residuals and the variances of the intact residuals obtained from formula (14), respectively; subscript Indicates restriction, subscript Indicates completeness; This represents a preset lower limit for positive values, used to prevent abnormal values ​​caused by conditional residual variance being zero or too small. When Less than hour, Increase. The actual preceding EEG increment contribution obtained by formula (15) is used as the input for step 4.2.

[0036] Step 4.2: Construct five types of alternative EEGs and calculate the contribution of the substitution increment, such as... Figure 4 The five alternative EEG sequences are all constructed based on the time-aligned EEG sequence output in step 1, with only the EEG source or temporal structure of formula (2) changed; the EEG sequences of each candidate muscle, the common effective sampling location, and the boundary of the action phase remain unchanged. The specific construction is as follows: (1) Cyclic shift EEG. Suppose that a certain EEG channel sequentially contains the 1st to the 2nd EEG within the common analysis interval. 1 sample value, of which This represents the total number of EEG sampling points in that interval; a non-zero integer cyclic shift amount is predetermined. and satisfy The original EEG sequence was changed from the first... Starting from the first sample value, they are arranged sequentially up to the second. Each sample value is then used to take the first to last sample values ​​from the original EEG sequence. Each sample value is appended to the end in its original order to obtain a cyclically shifted EEG with the same length as the original sequence. All EEG channels use the same... This alters the original time correspondence between EEG and current EMG.

[0037] (2) Future Displacement EEG. Let the future displacement amount be... There are 10 analysis sampling points, among which It is a positive integer and satisfies For the first common valid sampling location The time position, the time-aligned EEG sequence, the first The multichannel EEG column vectors at each time location are used as future misshifted EEG data at the [number]th time location. The values ​​at each time position; then, according to the advance delay formula (2) Constructing the prior EEG history vector, the actual EEG locations used are: .because If the delay is greater than the maximum precedence time, the location is still at the current electromyographic location. Afterwards; the common valid sampling positions output in step 1 have been reserved to meet the requirements. The end boundary.

[0038] (3) Time-reversed EEG. Each EEG channel within the common analysis interval is processed separately; if the original EEG sequence is in the order of 1st, 2nd, ..., 1st... The time-reversed EEG takes the nth sample value in sequence. , No. ... the first sampled value, where This indicates the total number of sampling points in the common analysis interval. The time-reversed EEG continues to use the common valid sampling locations and action phase boundaries output from step 1.

[0039] (4) Phase randomized EEG. Each EEG channel in the time-aligned EEG sequence is independently subjected to Fourier transform, keeping the amplitude, DC component and Nyquist frequency component of each frequency component unchanged; the remaining positive frequency components are assigned random phases generated by preset random seeds, and the corresponding negative frequency components are set according to the conjugate symmetry relationship; the phase randomized EEG with the same length as the original sequence is obtained by inverse Fourier transform, and the same standardization rules as the time-aligned EEG sequence are adopted.

[0040] (5) Cross-subject EEG. Keeping the candidate muscle electromyography sequences of the current subject unchanged, according to the pre-fixed subject sorting rules, select EEG segments from other subjects that have a fixed set of motor-related EEG channels, consistent channel order and analysis sampling rate, and meet the recording length requirements, as the cross-subject EEG of the current subject. The EEG segments of each other subject use the common effective sampling location and movement phase boundary of the current subject.

[0041] The above five types of surrogate EEGs only replace the EEG source in formula (2); the candidate muscle electromyography sequences, common effective sampling locations, and action phase boundaries output in step 1, as well as the remaining parameters and processing flow in steps 2 to 4.1, remain unchanged. For cyclic shift EEG, future shift EEG, time-reversal EEG, and phase randomization EEG, the corresponding surrogate EEGs are used to replace the real EEGs, and the complete conditional residual variance is calculated according to the same processing flow as the real EEGs. Then, the corresponding surrogate incremental contribution is calculated according to formula (15). For cross-subject EEGs, the EEGs of each other subject are used to replace the real EEGs of the current subject, and the surrogate incremental contribution of a cross-subject EEG is calculated according to the same processing flow and formula (15). When multiple other subjects are used, the median of the multiple cross-subject EEG surrogate incremental contributions is taken as the surrogate incremental contribution of this type. Then, the following formula is used to calculate the surrogate incremental contribution of the next subject. Specific incremental contribution of each candidate muscle: In formula (16), Indicates the first The specific incremental contribution of each candidate muscle; Indicates the candidate muscle number; , , , and These represent the surrogate incremental contributions of cyclic shift EEG, future shift EEG, time-reversal EEG, phase randomization EEG, and cross-subject EEG, respectively; superscript , , , , and They are respectively identified as true, cyclic shift, future shift, time reversal, phase randomization, and cross-subject condition; This indicates taking the maximum value among the five categories of incremental substitution contributions. Equals the difference between the actual preceding EEG increment contribution and the maximum surrogate increment contribution: when At that time, the contribution of the actual preceding EEG increment was higher than the contribution of all five types of substitution increments; when At that time, the actual preceding EEG increment contribution did not exceed at least one alternative increment contribution. Equation (16) It is step 4 for the first The final output of each candidate muscle is used as the input for step 5.

[0042] Step 5: Perform steps 2 to 4 for each candidate muscle to obtain the corresponding specific incremental contribution; summarize the specific incremental contributions according to a pre-fixed candidate muscle order, and form a candidate muscle specificity ranking from high to low, as the specificity analysis result. Step 5 finally outputs the candidate muscle name, specific incremental contribution, and candidate muscle specificity ranking; the candidate muscle specificity ranking represents the relative specificity of the real preceding EEG to different candidate muscles under the same action background.

[0043] On the other hand, the present invention also provides a brain-muscle coupling specificity analysis system based on conditional covariance, for implementing the above-mentioned brain-muscle coupling specificity analysis method, including an output acquisition module, a joint covariance module, a joint covariance reconstruction module, a specific increment contribution module, and a specificity analysis module. The output acquisition module is used to simultaneously acquire the subject's EEG and EMG signals, and outputs time-aligned EEG sequences and EMG sequences of each candidate muscle after preprocessing. The joint covariance module forms restricted joint variables and complete joint variables based on the time-aligned EEG sequences and EMG sequences of each candidate muscle, and calculates the restricted joint covariance and complete joint covariance accordingly. The joint covariance reconstruction module calculates and outputs the restricted joint covariance after Riemann aggregation and the reconstructed complete joint covariance based on the restricted joint covariance and the complete joint covariance. The specific incremental contribution module calculates the conditional residual variance using the restricted joint covariance after Riemann aggregation and the reconstructed complete joint covariance to obtain the specific incremental contribution of the current candidate muscle. The specificity analysis module is used to obtain the specific incremental contribution corresponding to each candidate muscle, summarize each specific incremental contribution according to the candidate muscle order, and form a specificity ranking of candidate muscles from high to low as the specificity analysis result.

[0044] Example: This embodiment is executed sequentially according to steps 1 to 5 in the invention content. The following specific parameters are used to illustrate the implementation of formulas (1) to (16). The same technical relationships as in the invention content directly refer to the original formulas, and no new formula numbers are set again.

[0045] Step 1: Synchronize data processing.

[0046] In this embodiment, the raw sampling rate of the EEG signal was 1000 Hz, and the raw sampling rate of the EMG signal was 2148.148193359375 Hz. The frontal midline electrode site from the international 10-20 system was selected. Central midline electrode site Left central electrode site Right central electrode site Left central apex electrode site and the right central apex electrode site Six motor-related EEG channels; among them , , , , and All electrode sites were identified using the international 10-20 system standard. Candidate muscles included the quadriceps femoris, tibialis anterior, biceps brachii, and anterior deltoid. A common time axis was established based on the original sampling times of the two types of signals, and linear interpolation was used to unify the EEG and EMG signals to a 100 Hz analysis sampling rate.

[0047] Standardized scores were applied to the six resampled EEG channels. For the electromyographic channels of each candidate muscle, standardized scores were applied first, then the absolute values ​​were taken and smoothed using a moving average window with a length of 5 analysis sampling points. Standardized scores were then applied again to obtain the time-aligned EEG sequences and the electromyographic sequences of each candidate muscle.

[0048] Each valid session retains a maximum of 600 seconds of shared analysis interval. To ensure that all shared valid sampling locations have complete temporal boundaries, 10 analysis sampling points are reserved at the beginning of the sequence and 20 analysis sampling points are reserved at the end of the sequence; only locations that simultaneously meet the above boundary conditions and have valid EEG and EMG are included in the shared valid sampling locations. Subsequently, the shared valid sampling locations are divided into 5 consecutive and non-overlapping action phases according to time sequence. Step 1 outputs the time-aligned EEG sequence, the EMG sequences of each candidate muscle, the shared valid sampling locations, and the action phase boundaries.

[0049] Step 2: Construction of Joint Variables and Joint Covariance. Using the time-aligned EEG sequences, EMG sequences of each candidate muscle, common effective sampling locations, and action phase boundaries output from Step 1, restrictive joint variables, complete joint variables, complete joint covariance, and restricted joint covariance are constructed.

[0050] For the th candidate muscle, two consecutive EMG samples are taken before the th common effective sampling position, so the EMG history order in formula (1) is 2; at a 100 Hz analysis sampling rate, the two samples correspond to 10 ms and 20 ms before the sampling position of the current EMG value, respectively. The EMG history vector obtained by formula (1) is simultaneously included in both restricted joint variables and complete joint variables.

[0051] For a fixed set of motion-related EEG channels, the total number of precedence delays in formula (2) is 3, and the number of sampling points for the three precedence delays is 2, 5, and 10 respectively, corresponding to 20 milliseconds, 50 milliseconds, and 100 milliseconds before the current EEG position. Multi-channel EEG column vectors for the three precedence delay positions are obtained according to the fixed channel order, and precedence EEG history vectors are constructed according to the precedence delay order. The precedence EEG history vector is only included in the complete joint variables and not in the restricted joint variables.

[0052] According to formula (3), the current EMG value and the EMG history vector constitute a restricted joint variable. According to formula (4), the prior EEG history vector is added to the restricted joint variable to form a complete joint variable. Therefore, the two joint variables have the same current EMG value, EMG history vector, and common effective sampling location. The only difference is that the complete joint variable adds the prior EEG history vector.

[0053] For each candidate muscle and each movement phase, use all of the phase's data. First, calculate the sample covariance. Then, linear shrinkage is performed according to formula (5). The shrinkage coefficient in this embodiment... The weight is fixed at 0.08, meaning the sample covariance weight is 0.92 and the target weight for identity matrix shrinkage is 0.08; the shrinkage covariance is then positively definitely corrected using a preset lower limit of positive eigenvalues. Subsequently, the sample covariance is directly processed from the same... Extracting restricted principal submatrices Step 2 ultimately yields the restricted joint covariance for each candidate muscle and each movement phase. and complete joint covariance The two types of covariance mentioned above are then directly assigned to step 3.

[0054] Step 3: Conditional covariance decomposition, cross-stage Riemann aggregation, and structure-preserving reconstruction. Using the constrained joint covariance and full joint covariance of each action stage output in Step 2, conditional covariance decomposition, cross-stage Riemann aggregation under shared time weights, and structure-preserving reconstruction are performed sequentially.

[0055] For each obtained in step 2 First, according to formula (7), it is divided into restricted joint covariance. Cross covariance matrix and preceding EEG covariance blocks Then calculate the cross-reconstruction coefficient matrix according to formula (8). Covariance of prior EEG conditions In numerical implementation, the result is obtained by solving linear equations. The equivalent result is that the inverse matrix is ​​not explicitly formed, and for Perform positive fixation repair.

[0056] Total number of action stages in this embodiment The weight of the shared time in formula (9) is set uniformly, i.e., from stage one to stage five. All are 0.2. Formula (10) uses the logarithmic Euclidean weighted mean to perform cross-stage Riemann aggregation on the restricted joint covariance and the covariance of the prior EEG conditions, and uses the same weight of 0.2 to perform linear weighting on the cross-reconstruction coefficient matrix.

[0057] Obtained using formula (10) , and First, obtain the reconstructed cross covariance matrix according to formula (11). and the reconstructed preceding EEG covariance block Then, the reconstructed complete joint covariance is obtained according to formula (12). Formula (13) shows The principal submatrix corresponding to the restricted joint variables is equal to Step 3 Output and And use it as input for step 4.

[0058] Step 4: True Prior EEG Incremental Contribution and Five-Type Alternative EEG Calibration. The true prior EEG increment contribution is calculated using the restricted joint covariance after Riemann aggregation and the reconstructed complete joint covariance output from Step 3, and then calibrated using five-type alternative EEG.

[0059] In this embodiment, Represents a column vector of given predictor variables Current electromyography value Conditional residual variance; Indicates the use of prediction Historical information exists under both conditions. Under limited conditions Only includes electromyographic history vectors, under complete conditions This includes both electromyographic history vectors and prior electroencephalographic history vectors. During calculation, they are respectively... and corresponding Divided into two parts, the restricted joint covariance after Riemannian aggregation. and the reconstructed complete joint covariance Divide into blocks; among which, The variance of the corresponding row and column is , The row and The covariance corresponding to the column is , The row and The covariance corresponding to the column is , The covariance matrix corresponding to the row and column is , This represents its inverse; in numerical implementation, the corresponding product is achieved by solving a linear equation. Substituting the above block divisions into formula (14), we get ;Will Substituting the corresponding blocks into formula (14), we get Then, calculate the actual preceding EEG increment contribution according to formula (15). In this embodiment, the same preset positive lower limit is used for the variance of the two conditional residuals.

[0060] Subsequently, five types of alternative EEG sequences are constructed using the time-aligned EEG sequences output from step 1. First, cyclically shifted EEG: assuming the common analysis interval contains... Each analysis sampling point, cyclic shift amount Fixed value 20 and The larger value in; the record length in this embodiment satisfies Each brainwave channel is configured according to the same Perform a cyclical translation with the beginning and end connected. Second, future misdirection EEG: future misdirection amount. With a fixed 20 analysis sampling points and a maximum lead time of 10 analysis sampling points, the following conditions are met. At common effective sampling locations Location of EEG sequence after time alignment The multi-channel EEG column vector, the 20 analysis sampling points reserved in step 1 satisfy the boundary of this operation. Third, time-reversed EEG: for each EEG channel, according to the first... Point, number The order of points up to point 1 is rearranged. Fourth, phase randomization EEG: the amplitude spectrum is maintained for each channel and the phase of the remaining frequency components is randomly changed, and phase randomization EEG is obtained by inverse Fourier transform. Fifth, cross-subject EEG: five other subjects with fixed motor-related EEG channel sets, channel order, analysis sampling rate, and recording length are selected according to a pre-fixed subject number order, and equal-length EEG segments are obtained as cross-subject EEG. The EEG segments of each other subject use the common effective sampling location and action phase boundary of the current subject.

[0061] All five types of surrogate EEGs only replace the EEG source in formula (2), while the other inputs, parameters, and processing procedures remain unchanged. For cyclic shift EEG, future shift EEG, time-reversal EEG, and phase randomization EEG, the corresponding surrogate EEG is used to replace the real EEG, and the complete conditional residual variance is calculated according to the same processing procedure as the real EEG. Then, the corresponding surrogate incremental contribution is calculated according to formula (15). For cross-subject EEG, the EEG of each other subject is used to replace the real EEG of the current subject, and the surrogate incremental contribution of a cross-subject EEG is calculated according to the same processing procedure and formula (15). In this embodiment, when 5 other subjects are used, the median of the 5 cross-subject EEG surrogate incremental contributions is taken as the surrogate incremental contribution of this type.

[0062] Finally, calculate according to formula (16). First, the maximum value among the five types of substitution increment contributions is taken as the maximum substitution increment contribution. Then, the actual preceding EEG increment contribution is used. Subtract the contribution of the largest substitution increment. This indicates that the contribution of the actual preceding EEG increment is higher than the contribution of all five types of substitution increments; This indicates that the actual preceding EEG increment contribution does not exceed at least one surrogate increment contribution. Step 4 is for the first... Individual candidate muscle output-specific increment contribution And use it as input for step 5.

[0063] Step 5: Candidate muscle-specific sorting.

[0064] In this embodiment, steps 2 to 4 are performed on the quadriceps femoris, tibialis anterior, biceps brachii, and anterior deltoid muscles respectively to obtain the corresponding specific incremental contributions. Step 5 summarizes the specific incremental contributions of each muscle according to a pre-fixed candidate muscle order, and forms a specificity ranking of the candidate muscles from high to low specific incremental contributions. At the same time, the candidate muscle names and corresponding specific incremental contributions are output as the specificity analysis results.

[0065] In another embodiment, the cross-stage Riemann aggregation operator of formula (10) can be replaced by the log-Euclidean weighted mean with the weighted Fraser mean under the affine invariant Riemann metric; this replacement only changes the cross-stage Riemann aggregation implementation of the restricted joint covariance and the prior EEG conditional covariance, without changing the input-output relationship of steps 1 to 5, the structure-preserving reconstruction method, and the five alternative EEG calibration procedures.

[0066] To verify the feasibility of the method of this invention and whether each key processing step can achieve the above-mentioned technical effects, synchronous EEG and EMG data from 23 subjects were used in the experiment. After unified data processing, 49 valid sessions were obtained. For each session, the common valid sampling location and 5 movement stages were determined according to step 1. Then, steps 2 to 5 were performed on the quadriceps femoris, tibialis anterior, biceps brachii, and anterior deltoid muscles, respectively. All comparison methods used the same subjects, the same valid sessions, the same candidate muscles, the same common valid sampling location, and the same movement stages, with each subject as an independent statistical unit. The experiment sequentially verified the control effect of EMG history vectors, the consistency of restricted joint covariance, and the three key aspects of five types of alternative EEG calibration.

[0067] First, the controlling effect of the electromyographic history vector was verified. After removing the electromyographic history vector, the subject-level mean difference of the results compared to the full method was 0.00695, with a 95% confidence interval of [0.00416, 0.00995]. Further scaling of the overall electromyographic sequences of each candidate muscle from 0.25 to 4 times, the lowest median Spearman rank correlation coefficient of the candidate muscle-specific ranking obtained by the full method remained at 1.0, and the first candidate muscle remained unchanged. These results indicate that the electromyographic history vector can control the influence of the muscle's own temporal continuity and overall amplitude changes on the contribution of the actual preceding EEG increment.

[0068] Secondly, the consistency of the restricted joint covariance after cross-stage Riemann aggregation was verified. Using the conditional covariance decomposition, cross-stage Riemann aggregation with shared time weights, and structure-preserving reconstruction of this invention, the principal submatrices corresponding to the restricted joint variables in the reconstructed complete joint covariance are completely consistent with the restricted joint covariance after Riemann aggregation, with a consistency error of 0. After canceling the structure-preserving reconstruction, the 99th percentile of the consistency error reached 6.9208. Further implementations were performed using logarithmic Euclidean weighted mean and weighted Fréchet mean under affine invariant Riemannian metric, respectively. In both implementations, this invention outperformed the control scheme that directly performed Riemann aggregation on the complete joint covariance. These results demonstrate that this invention can maintain the consistency of the restricted joint covariance after cross-stage Riemann aggregation.

[0069] Finally, it was verified whether the specific incremental contribution of the true preceding EEG was still retained after calibration using the five types of alternative EEGs. After calibration using cyclic shift EEG, future shift EEG, time-reversal EEG, phase randomization EEG, and cross-subject EEG, 19 out of 23 subjects still obtained positive specific incremental contributions, with a subject-level mean of 0.00442 and a 95% confidence interval of [0.00225, 0.00690]. According to the definition of formula (16), a positive value indicates that the incremental contribution of the true preceding EEG is higher than that of all five types of alternative incremental contributions. The average specific incremental contribution of the standard Granger causality analysis was 0.00060, which is about 7.4 times that of the method of this invention; the average specific incremental contribution of the marginal constraint Wasserstein centroid direct migration method was 0.00034, which is about 13.0 times that of the method of this invention, and the 95% confidence interval of the subject-level paired difference between the two groups was above 0. For 22 subjects with Fugl-Meyer motor function assessment scores, the Spearman rank correlation coefficient between the specific incremental contribution and the Fugl-Meyer score was 0.5584, and the 95% confidence interval of the subject clustering bootstrap method was [0.1575, 0.8363]. These results indicate that, even after controlling for the predictive role of the electromyographic history vector, maintaining consistency of the restricted joint covariance after cross-stage Riemannian aggregation, and calibrating the five types of surrogate incremental contributions, this invention can still extract the specific incremental contribution of the true preceding EEG.

[0070] To avoid miscounting the predictability of the muscles themselves as a contribution of prior EEG, the analysis results were first compared with and without EMG history vectors, as shown in Table 1. The subject-level mean difference between the two settings was 0.00695, with a 95% confidence interval of [0.00416, 0.00995], indicating that EMG history affects contribution estimation and supports controlling for its predictive role in the analysis. Furthermore, after scaling the EMG amplitude to 0.25–4 times its original value, the lowest median Spearman rank correlation coefficient, reflecting ranking consistency, was 1.000, and the number of changes in the first candidate muscle was 0, indicating that the complete method maintains ranking stability under the tested amplitude variations.

[0071] In addition to controlling for EMG history, it is also necessary to ensure that the comparisons before and after EEG use use the same EMG statistical basis. Relevant verification is shown in Table 2. Consistency error measures the degree of deviation between the restricted principal matrix in the complete joint covariance and the aggregated restricted joint covariance. When using structure-preserving reconstruction, this error is 0; when reconstruction is canceled, the 99th percentile of the error reaches 6.9208, indicating that structure-preserving reconstruction maintains consistency in the comparison basis for both conditions. Both log-Euclidean and affine-invariant implementations are superior to the corresponding direct Riemannian aggregation controls; therefore, the number of implementations superior to the controls in the table is 2 / 2.

[0072] Further testing was conducted to determine whether the real prior EEG could retain additional predictive information after calibration with five types of alternative EEGs, and the results were compared with two control methods. The results are shown in Table 3. According to formula (16), the specific incremental contribution is the real contribution minus the maximum alternative contribution, and a positive value indicates that the real contribution exceeds all five types of alternative contributions. The subject-level mean of the present invention is 0.00442, and the 95% confidence interval is [0.00225, 0.00690]; the mean values ​​of standard Granger causality analysis and the marginal constraint Wasserstein centroid direct transfer method are 0.00060 and 0.00034, respectively. The present invention is approximately 7.4 times and 13.0 times higher than the two methods, and the 95% confidence intervals of the paired differences between the two subject-level groups are both above 0, indicating that the specific incremental contribution retained by the present invention in this embodiment is higher than that of the two control methods.

[0073] In addition to the average contribution, the distribution of positive results among the subjects and the association between this indicator and motor function were also observed. The results are shown in Table 4. Nineteen out of 23 subjects obtained positive specific incremental contributions, indicating that most subjects retained positive results after five types of alternative EEG calibration. Among the 22 subjects with motor function assessment scores, the Spearman rank correlation coefficient between the specific incremental contribution and the score was 0.5584, and the 95% confidence interval of the subject cluster bootstrap method was [0.1575, 0.8363], indicating that this indicator was positively correlated with the motor function score in this embodiment.

[0074] Table 1. Control effect of electromyographic history vector

[0075] Table 2 Consistency of Constrained Joint Covariance

[0076] Table 3 Comparison of five types of alternative EEG calibration methods

[0077] Table 4. Positive values ​​of subjects after calibration and clinical relevance.

Claims

1. A brain-muscle coupling specificity analysis method based on conditional covariance, characterized in that, Includes the following steps: Step 1: Simultaneously acquire the subject's EEG and EMG signals, and output the time-aligned EEG sequence and the EMG sequences of each candidate muscle after preprocessing; Step 2: Based on the time-aligned EEG sequence and the candidate muscle EMG sequences, restrictive joint variables and complete joint variables are formed, and the restrictive joint covariance and complete joint covariance are calculated accordingly. Step 3: Based on the restricted joint covariance and the complete joint covariance, calculate the restricted joint covariance after Riemann aggregation and the complete joint covariance after reconstruction. Step 4: Calculate the conditional residual variance using the restricted joint covariance after Riemann aggregation and the complete joint covariance after reconstruction, and obtain the specific incremental contribution of the current candidate muscle. Step 5: Obtain the corresponding specific incremental contribution for each candidate muscle; The specific incremental contributions of each candidate muscle are summarized according to the order of the candidate muscles, and the candidate muscles are ranked from high to low to form a specificity ranking, which is used as the result of the specificity analysis.

2. The brain-muscle coupling specificity analysis method based on conditional covariance according to claim 1, characterized in that, Step 1 is specifically implemented as follows: synchronously collecting the electroencephalogram (EEG) signals and electromyographic (EMG) signals of several candidate muscles during the process of the subject completing limb movements, mapping the EEG signals and EMG signals to a common time axis and unifying them to a preset analysis sampling rate, and obtaining the time-aligned EEG sequence and the EMG sequence of each candidate muscle within the common analysis interval; The positions at both ends of the sequence that cannot form complete temporal samples are excluded according to the preset temporal boundary. The common effective sampling position is determined, and the action stage boundary is determined according to the unified action analysis time window. The time-aligned EEG sequence, the electromyography sequences of each candidate muscle, the common effective sampling position and the action stage boundary are obtained.

3. The brain-muscle coupling specificity analysis method based on conditional covariance according to claim 2, characterized in that, Step 2 is implemented as follows: Step 2.1: Using the electromyography (EMG) sequences of each candidate muscle, select consecutive EMG sampling values ​​before each common effective sampling location to construct an EMG history vector: in, Indicates the first The candidate muscles in the first The electromyographic history vector corresponding to each current sampling position; Indicates the first The candidate muscle before the current sampling position. Electromyography (EMG) sampling values ​​at each sampling location; Indicates the current sampling position number. Indicates the order of electromyographic history. Indicates transpose; Step 2.2: At the same common valid sampling location as in Step 2.1, using the time-aligned EEG sequence, sample from multiple preset advance time delay locations preceding the current sampling location and splice them together to construct an advance EEG history vector. , indicating the first The preceding EEG history vector corresponding to the current sampling location; Obtaining restricted joint variables Complete joint variables ,in, This is the current electromyography (EMG) value; Step 2.3: Within the common effective sampling location and action phase boundary, calculate the sample covariance according to the complete joint variables formed by the action phase and perform linear shrinkage to obtain the complete joint covariance: in, Indicates the first The candidate muscles in the first The complete joint covariance of each action phase after linear contraction; This represents the sample covariance calculated from complete joint variables within the same candidate muscle and the same movement phase. Indicates the linear shrinkage coefficient; Represents the matrix trace operation; Represents a complete joint variable dimensionality; express An identity matrix of order 1; Indicates the sequence number of the action phase, obtained As the complete joint covariance of this action phase; in, Indicates the first The first candidate muscle, the first Constrained joint covariance of each action phase; This represents the set of coordinate positions of the restricted joint variable within the complete joint variable; This indicates that the variances are selected simultaneously from the complete joint covariance. The principal submatrix formed by corresponding rows and columns.

4. The brain-muscle coupling specificity analysis method based on conditional covariance according to claim 3, characterized in that, Step 3 is implemented as follows: Step 3.1: Divide the complete joint covariance into blocks. ,in, This is the cross-covariance matrix between the restricted joint variables and the prior EEG history vectors, with rows corresponding to the restricted joint variable dimension and columns corresponding to the prior EEG history vector dimension. For the preceding EEG covariance block, superscript This refers to the electroencephalogram (EEG) portion; Perform conditional covariance decomposition as and : Indicates the first The first candidate muscle, the first The cross-reconstruction coefficient matrix of each action phase; by get; This represents the prior EEG conditional covariance, derived from the prior EEG covariance block. Subtract the portion that can be explained by the restricted joint variables get; Step 3.2: Set a set of shared time weights for each stage of the same action process. : in, Indicates the first Candidate muscle span Constrained joint covariance after Riemann convergence of each action phase; This represents the prior EEG conditional covariance after Riemannian convergence; This represents the cross-reconstruction coefficient matrix after linear weighting; This represents the total number of action phases. Step 3.3: First, reconstruct the cross-covariance matrix and the preceding EEG covariance block: Indicates the first The cross-covariance matrix of each candidate muscle after reconstruction is composed of... and Multiplying them together yields the result. This represents the reconstructed anterior EEG covariance block, which is the Riemann-aggregated anterior EEG conditional covariance. and Combining; Constitutes the complete joint covariance after reconstruction The structure preservation relationship is obtained through block reconstruction. , The constrained joint covariance after Riemannian aggregation.

5. The brain-muscle coupling specificity analysis method based on conditional covariance according to claim 4, characterized in that, Step 4 is implemented as follows: Step 4.1: Calculate the variance of the conditional residuals. ,in, Represents a column vector of given predictor variables Current electromyography value Conditional residual variance; Indicates the use of prediction Historical information exists under both conditions. Under restricted joint variables Only including electromyographic history vectors, under complete joint variables It includes both electromyographic history vectors and prior electroencephalographic history vectors; during calculation, they are respectively... and corresponding The joint covariance after Riemannian aggregation is divided into two parts. ,in, The variance of the corresponding row and column is , The row and The covariance corresponding to the column is , The row and The covariance corresponding to the column is , The covariance matrix corresponding to the row and column is , Indicates its reversal; will Substituting the above block into the conditional residual variance calculation formula, we get ;Will Substituting the corresponding blocks into the conditional residual variance calculation formula, we get... ; Calculate the contribution of real preceding EEG increments: in, Indicates the preset lower limit of positive numerical values; Step 4.2: Construct five types of surrogate EEGs: cyclic shift EEG, future shift EEG, time reversal EEG, phase randomization EEG, and cross-subject EEG, and calculate the incremental contribution of surrogate EEGs; All five types of alternative EEG only replace The source of the brainwave was then calculated according to the following formula. Specific incremental contribution of each candidate muscle: in, Indicates the first The specific incremental contribution of each candidate muscle; , , , and These represent the substitution increment contributions of cyclic shift EEG, future shift EEG, time-reversal EEG, phase randomization EEG, and cross-subject EEG, respectively.

6. The brain-muscle coupling specificity analysis method based on conditional covariance according to claim 5, characterized in that, The substitution increment contribution for cyclic shift EEG, future shift EEG, time-reversal EEG, phase randomization EEG, and cross-subject EEG is as follows: For cyclic shift EEG, future shift EEG, time-reversal EEG, and phase randomization EEG, the corresponding substitution EEG is used to replace the real EEG. The complete conditional residual variance is calculated according to the same processing procedure as the real EEG, and then the corresponding substitution increment contribution is calculated. For cross-subject EEG, the EEG of each other subject is used to replace the actual EEG of the current subject, and the substitution increment contribution of cross-subject EEG is calculated according to the same processing procedure. When multiple other subjects are used, the median of the multiple cross-subject EEG substitution increment contributions is taken as the substitution increment contribution.

7. The brain-muscle coupling specificity analysis method based on conditional covariance according to claim 6, characterized in that, The five types of alternative EEG structures—circular shift EEG, future shift EEG, time reversal EEG, phase randomization EEG, and cross-subject EEG—are described in detail below. Cyclic shift EEG: Suppose a certain EEG channel sequentially contains the 1st to the 2nd EEG channels within a common analysis interval. 1 sample value, of which This indicates the total number of EEG sampling points in this interval; Predetermine the non-zero integer cyclic shift amount and satisfy ; The original EEG sequence was changed from the first Starting from the first sample value, they are arranged sequentially up to the second. Each sample value is then used to take the first to last sample values ​​from the original EEG sequence. Each sampled value is appended to the end in its original order to obtain a cyclically shifted EEG with the same length as the original sequence. All EEG channels use the same... ; Future Displacement EEG: Let the future displacement amount be... There are 10 analysis sampling points, among which It is a positive integer and satisfies For the first common valid sampling location The time position, the time-aligned EEG sequence, the first The multichannel EEG column vectors at each time location are used as future misshifted EEG data at the [number]th time location. The values ​​at each time position; then according to the advance delay. Constructing the prior EEG history vector, the actual EEG locations used are: ; Time-reversed EEG: Each EEG channel within the common analysis interval is processed separately; If the original EEG sequence is in the order of 1, 2, ..., 1... The time-reversed EEG takes the nth sample value in sequence. , No. ... The first sample value, the time-reversed EEG continues to use the common valid sampling location and action phase boundary output in step 1; Phase randomized EEG: Fourier transform is performed independently on each EEG channel in the time-aligned EEG sequence, keeping the amplitude, DC component, and Nyquist frequency component of each frequency component unchanged. The remaining positive frequency components are assigned random phases generated by a preset random seed, and the corresponding negative frequency components are set according to the conjugate symmetry relationship; the phase randomized EEG with the same length as the original sequence is obtained by inverse Fourier transform, and the same standardization rules as the time-aligned EEG sequence are adopted. Cross-subject EEG: Keeping the candidate muscle electromyography sequences of the current subject unchanged, according to the pre-fixed subject sorting rules, select fixed motor-related EEG channel sets, channel order and analysis sampling rate from other subjects and the EEG segments with the required recording length as the cross-subject EEG of the current subject; the EEG segments of each other subject use the common effective sampling location and action phase boundary of the current subject.

8. A brain-muscle coupling specificity analysis system based on conditional covariance, used to implement the brain-muscle coupling specificity analysis method according to any one of claims 1 to 7, characterized in that, Includes the following modules: The output acquisition module is used to simultaneously acquire the subject's electroencephalogram (EEG) and electromyogram (EMG) signals, and outputs time-aligned EEG sequences and EMG sequences of each candidate muscle after preprocessing. The joint covariance module forms restricted joint variables and complete joint variables based on the time-aligned EEG sequence and the EMG sequences of each candidate muscle, and calculates the restricted joint covariance and complete joint covariance accordingly. The joint covariance reconstruction module calculates the restricted joint covariance and the reconstructed full joint covariance after Riemann aggregation based on the restricted joint covariance and the full joint covariance. The specific incremental contribution module uses the restricted joint covariance after Riemann aggregation and the complete joint covariance after reconstruction to calculate the conditional residual variance and obtain the specific incremental contribution of the current candidate muscle. The specificity analysis module is used to obtain the specific incremental contribution for each candidate muscle. The specific incremental contributions of each candidate muscle are summarized according to the order of the candidate muscles, and the candidate muscles are ranked from high to low to form a specificity ranking, which is used as the result of the specificity analysis.

Images