Cortical muscle coupling analysis method based on information flow decomposition

Abstract

The invention discloses a cortex and muscle coupling analysis method based on information flow decomposition. The method comprises the following steps: collecting and pre-processing EEG signals and EMG signals of multiple channels; constructing a cortical muscle function connection network through an ordinal number division conversion method; using a Helmholtz-Hough-Ketaria decomposition method to decompose and remove a circulating flow in the cortex muscle function connection network to obtain a residual gradient flow network; analyzing and comparing topological characteristics of the cortex muscle function connection network and the gradient flow network by using node access intensity; performing causal hierarchy analysis on the gradient flow network; the node input intensity and the causal hierarchy are used as classification features, and a support vector machine is used as a classifier for classification. According to the method, a high-credibility cortical muscle function connection network is constructed through an ordinal number division conversion network, the network is decoupled into a circulation flow and a gradient flow through Helmholtz-Hough-Cotela decomposition, and finally, the information transmission and causal relationship in a cortical muscle coupling system is intuitively reflected by utilizing causal hierarchical division.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of brain-muscle electro-signal coupling analysis, in particular to a coupling analysis framework based on network decomposition and causal hierarchy. BACKGROUND

[0002] Cortico-muscular coupling (CMC) is a phenomenon of information interaction between the brain cortex and muscles through neural pathways, which is specifically manifested as that the brain sends control instructions to drive muscle activity through the motor cortex, and the feedback information of the muscles reversely regulates the neural oscillation of the functional area of the cortex. The coupling process reflects the "command-execution-feedback" closed-loop mechanism in the motor control system, and the strength, directionality and time-frequency characteristics of the coupling process are closely related to the type of motor task, the level of force output and the state of the nervous system. Research on cortico-muscular coupling is of great significance for understanding the human motor nervous system and rehabilitation evaluation of patients with motor dysfunction.

[0003] Cortico-muscular coupling analysis mainly uses electroencephalography (EEG) and electromyography (EMG) to study the relationship between the cortex activity and the muscle group activity. Current CMC analysis mainly focuses on measuring the coupling relationship between specific single-channel EEG signals and task-related single-channel EMG signals, and has certain limitations. In addition, methods such as coherence and mutual information are difficult to identify the direction of information transmission.

[0004] Functional connectivity network is regarded as an effective tool for revealing the relationship between the cortex and the muscles as a whole. By abstracting the EEG and EMG channels as network nodes and connecting the relationship between specific channels in the form of edges, the topological characteristics of the coupling relationship can be further explored by using complex network theory. However, the authenticity of the functional connectivity network depends on the robustness of the causal measurement method, and even a small amount of false connections will damage the authenticity and effectiveness of the functional connectivity network. In addition, the complexity of the nervous system itself also makes it difficult for the functional connectivity network to clearly reflect the specific mode of neural information transmission. Although some network threshold and sparsification methods are used for network analysis, most of the methods do not involve the underlying mechanism of the functional connectivity network. Therefore, a new type of cortico-muscular coupling analysis method is needed. SUMMARY

[0005] The traditional method based on specific channel pair coupling analysis cannot comprehensively reflect the control mode of the cortical muscle, the method based on the functional connection network is difficult to effectively capture the information transmission directionality dynamics in the cortical muscle coupling system due to the network complexity, and the ordinary sparse method cannot eliminate the network connection redundancy caused by the information circulation transmission from the bottom mechanism.

[0006] To achieve the above purpose, the technical scheme adopted is:

[0007] A cortical muscle coupling analysis method based on information flow decomposition, comprising the following steps:

[0008] Step 1, collecting and preprocessing multi-channel EEG signals and EMG signals;

[0009] Step 2, constructing a cortical muscle functional connection network by a ordinal partition conversion method;

[0010] Step 3, using a Helmholtz-Hodge-Koide decomposition method to decompose and remove the circulating flow in the cortical muscle functional connection network, and obtaining the remaining gradient flow network;

[0011] Step 4, using node in-strength analysis and comparing the topological features of the cortical muscle functional connection network and the gradient flow network;

[0012] Step 5, performing causal hierarchy analysis on the gradient flow network;

[0013] Step 6, using the node in-strength and the causal hierarchy as classification features, and using a support vector machine as a classifier for classification.

[0014] As a preferred, the step 2 comprises:

[0015] Step 2-1, phase space reconstruction of time series; the time series includes the EEG signal time series and the EMG signal time series;

[0016] Step 2-2, calculating the conditional entropy between the time series of two channels to obtain a conditional entropy matrix of the coupling system composed of a plurality of EEG signal time series and EMG signal time series;

[0017] Step 2-3, constructing a weighted multilayer network containing signal nodes and the conditional entropy matrix elements, removing false connections in the weighted multilayer network; the signal nodes include EEG nodes and EMG nodes;

[0018] Step 2-4, selecting an optimal embedding dimension and time delay parameter for the weighted multilayer network, obtaining the cortical muscle functional connectivity network.

[0019] As preferred, the step 2-1 comprises:

[0020] for the EEG signal time series or the EMG signal time series According to Taknes embedding theorem, it is reconstructed in phase space to obtain an embedding vector:

[0021] v c (t) = {x c (t), x c (t+d), …, x c [t+(M-1)d]},

[0022] where M is the embedding dimension, d is the lag, t = 1, 2, …, T-(M-1)d;

[0023] ν c (t) is mapped to an integer sequence (s0, s1, …, s M-1 ) according to the rank order of its components, each s n is an ordinal pattern; the integer sequence after mapping satisfies:

[0024]

[0025] When the time series is embedded in M dimensions, there are different ordinal patterns, which are represented as π1, π2, …, π M！ .

[0026] As preferred, the step S2-2 comprises:

[0027] for any two channel time series deriving the associated sequence of ordinal patterns of the ordinal patterns under each of the time series and and calculating its conditional co-occurrence frequency

[0028] where τ = 0 represents simultaneous co-occurrence, and τ > 0 represents lag co-occurrence;

[0029] The conditional entropy calculation formula is as follows:

[0030]

[0031] in express and The frequency of simultaneous occurrence; the conditional entropy is used to reflect the causal relationship between any two channel time series.

[0032] Preferably, steps 2-3 include:

[0033] For the coupled system Z = {Z1, Z2, ..., Z...} N}, thus obtaining the conditional entropy matrix:

[0034]

[0035] The main diagonal elements are the Shannon entropy of a single EEG signal time series or EMG signal time series;

[0036] H by hard thresholding τ Thresholding:

[0037]

[0038] Where H max =log2M!, for different lags Through the corresponding matrix The network consists of a weighted multilayer network G = {V, E}, where V = {Z1, Z2, ..., Z}. N} represents the corresponding EEG node or EMG node, where E = {E1, E2, ..., E} J} are the elements of the conditional entropy matrix;

[0039] For each of the following The defined EEG node or EMG node Z m Define its parent node set as:

[0040]

[0041] Where h mjτ Single-level adjacency matrix The element in row m and column j; define Z. m The set of child nodes is:

[0042]

[0043] Define the minimum conditional subset:

[0044]

[0045] Z n Exclude Z m The set of child nodes;

[0046] define parameter ξ n :

[0047]

[0048] set threshold δ, if ξ n < δ, then the relationship between Z n and Z m is regarded as the false connection and removed.

[0049] As preferred, the optimal embedding dimension M = 4.

[0050] As preferred, in step 3, the convergence of decomposition is performed by using optimal gradient flow estimation:

[0051]

[0052] where G ij is the conductance model, and J is the net flow:

[0053]

[0054] J ij is the net flow:

[0055]

[0056] where, and represent the information flow between nodes i and j in the case of bidirectional information flow; if J ij > 0, it means that node i has a stronger influence on j, and information mainly propagates from i to j.

[0057] As preferred, in step 6, Lasso algorithm is used to perform feature screening on the node entry intensity and the causal hierarchy respectively.

[0058] Compared with the prior art, the beneficial effects of the present application are embodied in:

[0059] The traditional method based on specific channel pair coupling analysis cannot comprehensively reflect the control mode of the cortical muscle, and the method based on functional connection network is difficult to effectively capture the directionality dynamics of information transmission in the cortical muscle coupling system due to the inherent complexity of the network. In the present application, the network decomposition and causal hierarchy perspective are innovatively introduced into the cortical muscle coupling analysis, and a novel OPTN-HHKD causal analysis method is proposed, which can not only accurately identify the causal relationship between channels and avoid false connections, but also decouple the network into circular flow and gradient flow through information flow decomposition, and the causal hierarchy division can directly reflect the information transmission and causal relationship in the cortical muscle coupling system. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 Flow chart of the method of embodiment 1 of the present application;

[0061] Figure 2 Embedded dimension diagram of embodiment 1 of the present application;

[0062] Figure 3 Cortical muscle function connection network diagram under different frequency bands and grip strength of embodiment 1 of the present application;

[0063] Figure 4 Gradient flow network diagram under different frequency bands and grip strength of embodiment 1 of the present application;

[0064] Figure 5 Node strength diagram of different networks of embodiment 1 of the present application;

[0065] Figure 6 Causal hierarchy diagram under different frequency bands and grip strength of embodiment 1 of the present application. DETAILED DESCRIPTION

[0066] In order to make the technical means, creative features, purposes and effects of the invention easy to understand, specific diagrams are combined to further illustrate. But not limited to the following cases.

[0067] It should be noted that the structure, proportion, size, etc. shown in the drawings attached to the present specification are only used to cooperate with the content disclosed in the specification for understanding and reading by those skilled in the art, and do not define the limiting conditions that can be implemented, so they do not have technical substantive significance. Any modification of structure, change of proportion relationship or adjustment of size, without affecting the effects that can be produced and the purposes that can be achieved, should still fall within the scope of the disclosed technical content.

[0068] Embodiment 1:

[0069] As shown in a cortical muscle coupling analysis method based on information flow decomposition, comprising the following steps: Figure 1

[0070] Step 1, collect and pre-process EEG signals and EMG signals under upper limb grasping paradigm;

[0071] Use the device to synchronously collect 64-channel electroencephalogram (EEG) signals and 6-channel electromyogram (EMG) signals under different grip tasks, with a sampling frequency of 1000Hz;

[0072] ​The EEG signal and the EMG signal are preprocessed, the EEG signal is re-referenced by using average reference, baseline drift is removed, and independent component analysis is applied to eliminate obvious artifacts. The 50Hz power signal is removed by using an adaptive notch filter, and the EEG and EMG signals are subjected to 0.5-75Hz and 0.5-200Hz band-pass filtering by using FIR digital filters, respectively;

[0073] Step 2, a cortical muscle functional connectivity network is constructed by an ordinal partition transition network (hereinafter referred to as "OPTN") method:

[0074] Step 2-1, phase space reconstruction is performed on a single EEG or EMG signal;

[0075] For a given arbitrary EEG or EMG time series The phase space reconstruction can be performed according to the Taknes embedding theorem, and an embedding vector is obtained:

[0076] v c (t)={x c (t),x c (t+d),…,x c [t+(M-1)d]},

[0077] where M is the embedding dimension, d is the lag, and t=1,2,…,T-(M-1)d.

[0078] ν c (t) is mapped into an integer sequence (s0,s1,…,s M-1 ) according to the rank order of its components, and each s n is an ordinal pattern. The mapped sequence satisfies:

[0079]

[0080] When the time series is embedded in M dimensions, there are different ordinal patterns, and the present application uses π1,π2,…,π M！ to represent these patterns.

[0081] Step 2-2, the conditional entropy is used to measure the causal relationship between the EEG signal and the EMG signal;

[0082] For any two-channel time series (including EEG time series and EMG time series, EEG time series and EEG time series, and EMG time series and EMG time series), the present application can derive the associated sequence of the ordinal pattern of each time series and compute their conditional co-occurrence frequencies i.e. when occurs the conditional frequency of occurrence. Where τ = 0 corresponds to simultaneous co-occurrence, while τ > 0 implies lagged co-occurrence. From these co-occurrence frequencies, the present invention can obtain an estimate of the conditional entropy:

[0083]

[0084] where denotes the frequency of simultaneous occurrence. The conditional entropy is able to reflect the causal relationship between any two time series of channels.

[0085] Step 2-3, removing spurious connections in multi-channel EEG / EMG coupled system

[0086] For a coupled system Z = {Z1, Z2,..., Z N}, the conditional entropy matrix can be obtained by calculating the conditional entropy between each pair of time series:

[0087]

[0088] where the main diagonal elements are the Shannon entropy of individual EEG or EMG time series. The H τ matrix is thresholded by a hard threshold:

[0089]

[0090] where H max = log2M! for different lags A weighted multi-layer network G = {V, E} can be constructed from the corresponding matrix , where V = {Z1, Z2,..., Z N} are the corresponding EEG or EMG nodes, and E = {E1, E2,..., E J} are the corresponding elements in the conditional entropy matrix.

[0091] For each EEG or EMG node Z m defined by , the present invention defines its parent node set as:

[0092]

[0093] where h mjτ is the element in the single-layer adjacency matrix of the mth row and jth column. Similarly, the present invention defines the child node set of Z m as: ​

[0094]

[0095] To represent the causal relationship between any two channels of EEG or EMG nodes Z m and Z n , the present application first defines the minimum conditional subset:

[0096]

[0097] Point Z n exclude the child node set of Z m , if Z m is the only child node of Z n , then At this time If is still empty, let In addition, the conditional set in each case can obtain the corresponding delay {τ'1,…,τ' r}. In order to remove the indirect causal relationship, the present application defines the parameter ξ n :

[0098]

[0099] ξ n can reflect the causal connection between Z m and Z n , the smaller the ξ n , the greater the possibility of Z n being indirectly causal to Z m . By setting a threshold δ, if ξ n < δ, the relationship between Z n and Z m is considered to be a false connection and is removed.

[0100] Step 2-4, find the optimal parameters, and construct the cortical muscle functional connectivity network.

[0101] In order to construct a cortical muscle functional connectivity network with higher reliability and biological interpretation, according to steps 2-1 and 2-2, it is necessary to find the optimal embedding dimension M and time delay parameter τ for the OPTN method.

[0102] In order to select a suitable embedding dimension M, first, the embedding dimension M is tested by using the C3 channel of the primary motor area in electroencephalogram and the finger flexor muscle (Flexor digitorum superficialis, abbreviated as "FDS") related to the grip task. As shown in Figure 2 , when the embedding dimension M is set to 3, the OPTN fails to detect the EEG-EMG coupling, while due to when 10M When much larger than the data length, it can cause many false connections to be identified as causal effects. Therefore, M = 4 is chosen as the optimal embedding dimension to construct the cortical-muscular functional connectivity network;

[0103] The time delay of the bidirectional information transmission between cortex and muscle is mainly in the range of 20-30 ms. Considering the differences between subjects, the delay parameter τ is set to the range of 10-40 ms, and the optimal delay is determined according to the peak coupling strength in this range;

[0104] The coupling relationship between each channel in the β (15-30 Hz) and γ (30-60 Hz) bands is calculated respectively using the optimal parameters, and the cortical-muscular functional connectivity network is constructed. As shown in Figure 3 The functional connectivity between the β band channels is always stronger than that in the γ band, and the cortical-muscular coupling is relatively weak compared with the inter-cortical or inter-muscular coupling.

[0105] Step 3, use the Helmholtz-Hodge-Kodair decomposition (HHKD) method to decompose and remove the circular flow in the cortical-muscular functional connectivity network, and obtain the remaining gradient flow network;

[0106] The obtained cortical-muscular functional connectivity network is decomposed into information flow, and the circular flow in the network is removed while the gradient flow is retained, and the gradient flow network is obtained, as shown in Figure 4 The cortical-muscular coupling has directional differences. In the β band, the descending coupling between the cortex and the superficial flexor muscle of the finger is always strong. With the increase of grip strength, the auxiliary muscles are recruited, and the descending coupling of these muscles is enhanced.

[0107] Step 3-1:

[0108] The complex cortical-muscular functional connectivity network obtained by the OPTN method is split into gradient components and circular components, where the gradient flow represents the directionality of the causal flow, and the circular flow represents the circular dependence in the network. For a vector field According to the Helmholtz theorem, it can be split into:

[0109] F(r) = G(r) + R(r)

[0110] Where the gradient field is a non-rotational field determined by the potential function Φ(r), and the non-divergence field (circular flow component) R(r) satisfies

[0111] In this embodiment, for a cortical-muscular functional connectivity network, it is assumed that J ijThe information flow between any two nodes i and j, and satisfies:

[0112]

[0113] According to Helmholtz theorem, the network can be uniquely decomposed as:

[0114]

[0115] where is the gradient component of the network, G ij is the weight (default is 1), Φ i is the latent level value of the network. represents the circulation component, for each EEG or EMG node, the sum of the circulating current entering and flowing out is zero.

[0116] Step 3-2: In order to deal with the possible existence of bidirectional information flow in the cortico-muscular coupling network, HHKD introduces the conductivity model to measure the total causal flow between any two EEG or EMG nodes:

[0117]

[0118] This value is symmetric, that is, G ij = C ji , which means that the total causal flow is the same regardless of direction. In addition, define the net flow:

[0119]

[0120] where, and represent the information flow between nodes i and j under the condition of bidirectional information flow; if J ij > 0, it means that node i has a stronger impact on j, and information mainly propagates from i to j. On this basis, the optimal gradient flow estimation is used to converge the decomposition:

[0121]

[0122] In addition, constraints need to be added to Φ to ensure the uniqueness of the decomposition, such as Φ n = 0 or ∑ i Φ i = 0.

[0123] Step 4, analyze and compare the topological characteristics of the cortico-muscular functional connectivity network and the gradient flow network using node in strength;

[0124] Use node in strength to analyze the graph theory parameters of the cortico-muscular functional connectivity network and the gradient flow network:

[0125]

[0126] As Figure 5 shown on the left. For the cortico-muscular functional connectivity network, there is little difference in node strength between each channel type under visual inspection. In contrast, for the gradient flow network on the right, the node strength of the index flexor is significantly higher than the EEG channels. In addition, the node strength of muscles such as the flexor carpi ulnaris and the flexor carpi radialis also increases with increasing grip strength. This pattern indicates that the bidirectional cortico-muscular coupling in the beta band is mainly dominated by the descending coupling. Figure 5

[0127] Step 5, use causal hierarchy analysis to intuitively reflect the causal relationship between the nodes of the cortico-muscular coupling;

[0128] Causal hierarchy analysis of the gradient flow network is shown in Figure 6 In the beta band, the index flexor is always at the lowest level of the causal hierarchy structure, and the causal hierarchy structure of other muscles gradually decreases with increasing grip strength. Compared with the EMG channel, the EEG channel usually occupies a higher position in the causal structure, with the left primary motor area having a higher level. However, in the gamma band, the causal hierarchy of the EMG channel increases with increasing grip strength, and when the grip strength reaches 20 kg, it even exceeds the level of the EEG channel.

[0129] Step 6, use the node strength and causal hierarchy as classification features, and use support vector machine as the classifier to classify different grip strength, the specific steps are as follows:

[0130] Step 6-1: Based on the cortico-muscular functional connectivity network and the gradient flow network, node strength and causal hierarchy are extracted from 36 tasks of all subjects respectively. Considering that high-dimensional features require large amount of calculation and may lead to overfitting, Lasso algorithm is used to select features for the node strength and causal hierarchy of multiple nodes in the network.

[0131] Step 6-2: Select support vector machine as the classifier. 10-fold cross-validation is used to obtain more reliable classification results. Table 1 and Table 2 show the classification results of different frequency band features. The results show that compared with the topological features of the cortico-muscular functional connectivity network, the node strength and causal hierarchy of the decomposed network have better classification accuracy.

[0132] Table 1: Classification results of different features in the beta band

[0133]

[0134] Table 2: Classification results of different features in the gamma band

[0135] ​

Claims

1. A cortical-muscle coupling analysis method based on information flow decomposition, characterized in that, Includes the following steps: Step 1: Acquire and preprocess multi-channel EEG and EMG signals; Step 2: Construct the cortical-muscular functional connectivity network using the ordinal partitioning transformation method; Step 3: Use the Helmholtz-Hodge-Cordella decomposition method to decompose and remove the cyclic flow in the cortical muscle functional connectivity network to obtain the remaining gradient flow network; Step 4: Use node ingress strength analysis to compare the topological features of the cortical muscle functional connectivity network and the gradient flow network; Step 5: Perform causal hierarchy analysis on the gradient flow network; Step 6: Use the node input strength and the causal hierarchy as classification features, and use a support vector machine as a classifier for classification.

2. The cortical-muscle coupling analysis method based on information flow decomposition according to claim 1, characterized in that, Step 2 includes: Step 2-1, performing phase space reconstruction on the time series; the time series includes the EEG signal time series and the EMG signal time series; Step 2-2: Calculate the conditional entropy between the time series of the two channels to obtain the conditional entropy matrix of the coupled system consisting of multiple EEG signal time series and EMG signal time series; Steps 2-3: Construct a weighted multilayer network containing signal nodes and the elements of the conditional entropy matrix, and remove spurious connections in the weighted multilayer network; the signal nodes include EEG nodes and EMG nodes; Steps 2-4: Select the optimal embedding dimension and time delay parameters for the weighted multilayer network to obtain the cortical muscle function connectivity network.

3. The cortical-muscle coupling analysis method based on information flow decomposition according to claim 2, characterized in that, Step 2-1 includes: processing the EEG signal time series or the EMG signal time series. Reconstructing the phase space using Taknes' embedding theorem yields the embedding vector: v c (t)={x c (t),x c (t+d),…,x c [t+(M-1)d]}, Where M is the embedding dimension, d is the lag, and t = 1, 2, ..., T-(M-1)d; ν c (t) is mapped to an integer sequence (s0, s1, ..., s) according to the rank order of its components. M-1 ), each s n It is an ordinal pattern; the mapped integer sequence satisfies: When the time series is embedded in M ​​dimensions, there are different ordinal patterns, using π1, π2, ..., π M！ These represent the ordinal patterns.

4. The cortical-muscle coupling analysis method based on information flow decomposition according to claim 3, characterized in that, Step S2-2 includes: For any two-channel time series Derive the associated sequences of the ordinal patterns for each of the time series. and And calculate its conditional co-occurrence frequency. Where τ = 0 indicates simultaneous co-occurrence, while τ > 0 indicates delayed co-occurrence; The formula for calculating the conditional entropy is as follows: in express and The frequency of simultaneous occurrence; the conditional entropy is used to reflect the causal relationship between any two channel time series.

5. The cortical-muscle coupling analysis method based on information flow decomposition according to claim 4, characterized in that, Steps 2-3 include: For the coupled system Z = {Z1, Z2, ..., Z...} N }, thus obtaining the conditional entropy matrix: The main diagonal elements are the Shannon entropy of a single EEG signal time series or EMG signal time series; H by hard thresholding τ Thresholding: Where H max =log2M!, for different lags Through the corresponding matrix The network consists of a weighted multilayer network G = {V, E}, where V = {Z1, Z2, ..., Z}. N } represents the corresponding EEG node or EMG node, where E = {E1, E2, ..., E} J } are the elements of the conditional entropy matrix; For each of the following The defined EEG node or EMG node Z m Define its parent node set as: Where h mjτ Single-level adjacency matrix The element in row m and column j; define Z. m The set of child nodes is: Define the minimum conditional subset: Z n Exclude Z m The set of child nodes; Define parameter ξ n : Set a threshold δ, if ξ n <δ, then Z n With Z m The relationship between them is considered a spurious connection and is removed.

6. The cortical-muscle coupling analysis method based on information flow decomposition according to claim 2, characterized in that, The optimal embedding dimension M = 4.

7. The cortical-muscle coupling analysis method based on information flow decomposition according to claim 1, characterized in that, In step 3, the optimal gradient flow estimation is used to converge the decomposition: Among them, through the conductivity model G ij Used to measure the total causal flow between node i and node j: J ij Net flow: in, and This represents the amount of information flow between node i and node j in the case of bidirectional information flow; if J ij A value greater than 0 indicates that node i has a stronger influence on j, and information mainly propagates from i to j.

8. The cortical-muscle coupling analysis method based on information flow decomposition according to claim 1, characterized in that, In step 6, the Lasso algorithm is used to perform feature filtering on the node ingress strength and the causal hierarchy, respectively.

Images