A two-way closed-loop brain-computer interaction electroacupuncture parameter intelligent matching system for insomnia

Abstract

The present invention discloses a bidirectional closed-loop brain-computer interaction electroacupuncture parameter intelligent matching system for insomnia, including an electroencephalogram acquisition module, an OpenViBe signal interface and control platform module, a real-time signal processing module, an electroacupuncture parameter intelligent matching module, an electroacupuncture stimulation module, and a real-time display module; the electroencephalogram acquisition module collects the electroencephalogram signals of insomnia patients in real time during the electroacupuncture process; the OpenViBe signal interface and control platform module completes the time synchronization, real-time access, centralized collection and storage requirements in the process of acquiring electroencephalogram signals; the real-time signal processing module performs preprocessing, information decoding and feature screening on electroencephalograms; the electroacupuncture parameter intelligent matching module realizes accurate matching of optimal electroacupuncture stimulation parameters; the electroacupuncture stimulation module performs current stimulation on specific acupuncture points; and the real-time display module visualizes the electroencephalogram information decoding process and the electroacupuncture stimulation parameters. The present invention realizes bidirectional closed-loop interaction by accurately matching electroacupuncture stimulation parameters according to real-time electroencephalogram features, thereby achieving the purpose of optimally treating insomnia.

Description

Technical Field

[0001] The present invention belongs to the technical field of electroacupuncture, and in particular relates to a two-way closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system for insomnia. Background Art

[0002] Insomnia is a state of difficulty falling asleep, usually accompanied by symptoms such as poor spirits, drowsiness, irritability, or depression during the day, which seriously affects the patient's physical and mental health. Electroacupuncture is widely used in clinical practice because it is easy to achieve acupuncture standardization and can improve the efficacy of acupuncture. However, there are few clinical studies on the amount of electroacupuncture stimulation and there are the following shortcomings: First, there are significant differences in the selection of electroacupuncture frequency and waveform, and the intensity of electroacupuncture is still based on patient tolerance, and the quantitative basis lacks objectivity. Second, most scholars study the effect of a single electroacupuncture parameter on the efficacy, and lack clinical research on the optimal combination of multiple electroacupuncture parameters.

[0003] EEG can reflect the internal activity state of the brain, and EEG technology is widely used in the clinical treatment of brain diseases such as insomnia and epilepsy due to its advantages such as non-invasiveness, no radiation, portability and high temporal resolution. There are various types of EEG features, but what kind of EEG features can most accurately characterize the brain state and acupuncture efficacy during electroacupuncture treatment of insomnia remains an unresolved key issue. Summary of the invention

[0004] Purpose of the invention: In order to solve the technical problems mentioned in the above background technology, the present invention proposes a bidirectional closed-loop brain-computer interaction electroacupuncture parameter intelligent matching system for insomnia, and proposes an improved EEG microstate complexity analysis algorithm to establish a bidirectional closed-loop brain-computer interaction system of "electroacupuncture parameters-EEG effects-feedback-adjustment of electroacupuncture parameters" to achieve visualized electroacupuncture precise treatment of insomnia.

[0005] Technical solution: The present invention provides a two-way closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system for insomnia, including an EEG acquisition module, an OpenViBe signal interface and control platform module, a real-time signal processing module, an electroacupuncture parameter intelligent matching module, an electroacupuncture stimulation module, and a real-time display module;

[0006] The EEG acquisition module includes a whole-brain electrode cap and an amplifier, which collects the whole-brain EEG signals of insomnia patients in real time and transmits them to the OpenViBe signal interface and control platform module;

[0007] The OpenViBe signal interface and control platform module completes the time synchronization, real-time access, centralized collection and storage requirements in the process of EEG signal acquisition;

[0008] The real-time signal processing module performs EEG preprocessing, EEG information decoding and EEG feature screening, and adopts an improved microstate complexity analysis method to comprehensively extract microstate discrete time series based on the global field energy peak potential topology information and trough potential topology information, providing objective detection indicators for the electroacupuncture stimulation effect;

[0009] The electroacupuncture parameter intelligent matching module inputs the EEG feature vector into the deep learning model to achieve accurate matching of the optimal electroacupuncture stimulation parameters;

[0010] The electroacupuncture stimulation module includes a control circuit, an output wire, an electroacupuncture clamp, and an acupuncture needle, and automatically performs corresponding current stimulation on specific acupuncture points according to the matched electroacupuncture parameters;

[0011] The real-time display module visualizes the EEG information decoding process and the electroacupuncture stimulation parameters during the electroacupuncture process, and provides a function for the doctor to make manual adjustments.

[0012] Furthermore, the OpenViBE signal receiving and control platform module adopts the open source OpenViBE framework to realize real-time EEG signal acquisition and transmission between different modules, and transmission of EEG characteristics and electroacupuncture parameters, providing an infrastructure for a two-way closed-loop brain-computer interaction system.

[0013] Furthermore, the implementation process of the microstate complexity analysis method is as follows:

[0014] For a whole brain EEG signal with N channels [X1(t),X2(t),X3(t),…,X N (t)], calculate the global field power GFP of brain activity:

[0015]

[0016] A topology-based atomization and aggregation hierarchical clustering algorithm is used as the clustering algorithm for microstate analysis. The potential topology of local peak points and the potential topology of local trough points are clustered separately. The potential topology map of each local peak point is regarded as a family. All potential topology maps are initially regarded as independent groups, and the potential topology operation of local trough points is the same. The spatial correlation of each potential topology map with other topographic maps is calculated separately:

[0017]

[0018] Among them, u and v are two different potential topographic maps. The spatial correlation of each potential topological map with other topological maps is calculated respectively, and the topological map with the lowest total correlation is selected as the worst topological map; the current worst topological map is redistributed to the remaining group with the closest correlation, and the potential topological maps in this group are averaged to obtain a new potential topological distribution; after each iteration, a worst potential topological map is redistributed until the number of clusters reaches the preset number of categories; the above clustering calculations are performed on the potential topological maps of the local peak value point of GFP and the local valley value point, respectively, to obtain the potential topological map based on the local peak value point of GFP. The clustered microstates of the potential topology map of the local trough value points of GFP are compared; the potential topology map of the local peak value points of GFP is clustered into 4 types of microstate topography, and the potential topology map of the local trough value points of GFP is clustered into 2 types of microstate topography; the spatial correlation of the scalp potential topology map at each time point is compared with the standard microstate obtained by clustering, and the scalp potential is matched with the label of the standard microstate; the EEG signal is decoded into a microstate time series in which six microstates alternate. After redefining the microstate calculation process, the improved microstate analysis can obtain a microstate time series with a higher global explained variance GEV:

[0019]

[0020] Among them, m is the number of original scalp potential topological maps, and the new microstate time series can represent the changes in brain states at more moments. After obtaining the discrete sequence of microstates, the multi-scale entropy of the discretized sequence is calculated to obtain the complexity of dynamic switching between brain microstates.

[0021] Furthermore, the EEG feature screening implementation process is as follows:

[0022] The overlapping group sparse LASSO model is introduced into the deep belief network DBN model, and the features are re-encoded based on the neural network; first, the traditional EEG signal features are transformed into new unified modal features, and the optimized weights of each feature dimension are calculated by adopting the structured sparse constraint optimization method; the feature dimensions with higher weights are selected, and the multimodal features are extracted and screened; the objective function used is:

[0023]

[0024] In Group LASSO, the traditional EEG features of different frequencies are divided into G groups, I g is the characteristic subscript of the g group (g=1,…,G), is the weight of each group, λ is the penalty parameter, and β is the parameter to be estimated.

[0025] Furthermore, in the electroacupuncture parameter intelligent matching module, a precise matching model between EEG characteristics and optimal electroacupuncture stimulation parameters is implemented based on deep learning, and transfer learning is used in a real-time closed-loop system to continuously iterate and optimize the model to improve the matching accuracy of electroacupuncture parameters.

[0026] Furthermore, the electroacupuncture stimulation parameters include output waveform, pulse frequency and current intensity.

[0027] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention collects EEG signals of insomnia patients during electroacupuncture in real time, extracts features such as EEG microstates based on an improved microstate algorithm, combines deep learning and other technologies to accurately match the optimal electroacupuncture stimulation amount for insomniacs, constructs a two-way closed-loop brain-computer interaction system, and realizes visual electroacupuncture to accurately treat insomnia. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a block diagram of the bidirectional closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system;

[0029] Figure 2 Schematic diagram of the bidirectional closed-loop brain-computer interaction electroacupuncture parameter intelligent matching system

[0030] Figure 3 A flowchart of the microstate improvement algorithm proposed by the present invention;

[0031] Figure 4 It is a schematic diagram of the working principle of the bidirectional closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system;

[0032] Figure 5 This is a visualization page diagram of electroacupuncture parameters and EEG characteristics in the present invention. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0034] The present invention proposes a two-way closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system for insomnia. Figure 1 As shown, it includes EEG acquisition module, OpenViBe signal interface and control platform module, real-time signal processing module, electroacupuncture parameter intelligent matching module, electroacupuncture stimulation module, and real-time display module. Among them:

[0035] like Figure 2As shown in the figure, the EEG acquisition module includes a whole-brain electrode EEG cap and an amplifier with a sampling rate of 1000Hz. It can collect the whole-brain EEG signals of insomnia patients in real time, and transmit them to the OpenViBe signal receiving and control platform module after amplification by the amplifier. The host includes the OpenViBe signal receiving and control platform module, the real-time signal processing module and the electroacupuncture parameter intelligent matching module. The electroacupuncture instrument contains the control circuit in the electroacupuncture stimulation module. By embedding the host and the electroacupuncture instrument into one, the automatic effect of automatically changing the current output according to the matched electroacupuncture parameters without manual operation can be achieved.

[0036] The OpenViBe signal interface and control platform module completes the time synchronization, real-time access, centralized collection and storage requirements in the process of EEG signal acquisition. The real-time signal processing module performs EEG preprocessing, EEG information decoding and EEG feature screening, and adopts an improved microstate complexity analysis method to comprehensively extract microstate discrete time series based on the global field energy peak potential topology information and trough potential topology information, providing objective detection indicators for the electroacupuncture stimulation effect. The OpenViBe signal reception and control platform module is the main framework of the closed-loop brain-computer interface. OpenViBe is a brain-computer interface software platform for designing, testing and online use, which can be used for real-time neuroscience research and the implementation of bidirectional closed-loop brain-computer interaction technology. OpenViBE provides users with the flexibility to design and change experimental procedures, and provides a powerful interface for applications and processing algorithms written in other languages ​​such as Python and Matlab. The OpenViBe signal receiving and control platform module of the present invention mainly uses the Acquisition Sever application in the OpenViBE application layer. The Acquisition Sever acquires neural signals from the EEG acquisition module through the OpenViBE environment on the PC side, and then sends them to the real-time signal processing module after encapsulation through the LabStreamingLayer (LSL) protocol to complete the real-time transmission of the signal. LSL is mainly composed of the kernel liblsl, data stream outlets (Stream Outlets), data stream inlets (Stream Inlets), resolution functions (Resolve Function) and built-in clocks (Bulit-in Clock), which can meet the requirements of time synchronization, access, centralized collection, storage and visualization in the process of time series signal transmission.

[0037] The electroacupuncture parameter intelligent matching module inputs the EEG feature vector into the deep learning model to achieve accurate matching of the optimal electroacupuncture stimulation parameters. Based on deep learning, an accurate matching model between EEG features and optimal electroacupuncture stimulation parameters is implemented, and transfer learning is used in a real-time closed-loop system to continuously iterate and optimize the model to improve the matching accuracy of electroacupuncture parameters. Electroacupuncture stimulation parameters include output waveform, pulse frequency, and current intensity.

[0038] The electroacupuncture stimulation module includes a control circuit, an output wire, an electroacupuncture clamp, and an acupuncture needle, and automatically performs corresponding current stimulation on specific acupoints according to the matched electroacupuncture parameters.

[0039] The real-time display module visualizes the EEG information decoding process and the electroacupuncture stimulation parameters during electroacupuncture, and provides the doctor with the function of manually adjusting the electroacupuncture parameters and controlling the start and end of electroacupuncture and EEG data collection.

[0040] like Figure 3 As shown in the figure, the EEG microstate calculation method used in the real-time signal processing module provides an objective detection index for the electroacupuncture stimulation effect on insomnia. The specific implementation process is as follows:

[0041] (1) For a whole-brain EEG signal with N channels [X1(t),X2(t),X3(t),…,X N (t)], calculate the global field power (GFP) of brain activity:

[0042]

[0043] Traditional microstate analysis assumes that the local maximum of the GFP curve represents the moment of highest field strength and maximum signal-to-noise ratio, so the EEG potential topology at the local maximum of GFP is extracted for the next step of microstate clustering analysis. However, its limitations are: ① The potential topology at the GFP peak point only accounts for a very small part of all time points, so a large amount of relevant information will be lost. ② In addition, the peak value range of GFP spans a large range, and the GFP peak point may include a very small GFP value, which is contrary to the assumption that it extracts the EEG potential at the point with the maximum signal-to-noise ratio, and the impact caused has not been verified by any article. Therefore, in the improved algorithm proposed in this application, the extraction of microstates is redefined, and the potential topology is extracted from the local peak value point and the local trough value point, respectively, and the next step of microstate clustering is performed respectively.

[0044] (2) Using the topology-based atomization and aggregation hierarchical clustering algorithm (TAAHC) as the clustering algorithm for microstate analysis, the potential topology of the local peak value point and the potential topology of the local trough value point are clustered and analyzed respectively. The steps of the clustering algorithm are:

[0045] 1) The potential topology of each local peak point is regarded as a family. All potential topologies are initially treated as independent groups, and the potential topology operations of local trough points are also the same.

[0046] 2) Calculate the spatial correlation between each potential topological map and other topographic maps:

[0047]

[0048] Among them, u and v are two different potential topographic maps. The spatial correlation between each potential topological map and other topological maps is calculated respectively, and the topological map with the lowest total correlation is selected as the worst topological map.

[0049] 3) Redistribute the current worst topology map to the remaining group with the closest correlation, and average the potential topology maps in this group to obtain a new potential topology distribution;

[0050] 4) Repeat steps 2) and 3), and reallocate a worst potential topology map in each iteration until the number of clusters reaches the preset number of categories.

[0051] 5) Perform the above clustering calculations on the potential topology of the local peak value point of GFP and the potential topology of the local trough value point, respectively, to obtain clustered microstates based on the potential topology of the local peak value point of GFP and clustered microstates based on the potential topology of the local trough value point of GFP. Generally, the traditional microstate analysis is to cluster and calculate 4 types of standard microstate topographic maps of the potential topology of the GFP peak position. In the improved method of the present application, the potential topology of the local peak value point of GFP is clustered into 4 types of microstate topography, and the potential topology of the local trough value point of GFP is clustered into 2 types of microstate topography.

[0052] 6) Compare the spatial correlation of the scalp potential topology at each time point with the standard microstate obtained by clustering, and match the scalp potential with the label of the standard microstate. After calculation, the EEG signal is decoded into a microstate time series in which six microstates appear alternately. After redefining the microstate calculation process, the improved microstate analysis can obtain a microstate time series with a higher global explained variance (GEV):

[0053]

[0054] Among them, m is the number of original scalp potential topological maps, and the new microstate time series can represent the changes in brain states at more moments.

[0055] 7) After obtaining the discrete sequence of microstates, the multi-scale entropy of the discretized sequence is calculated to obtain the complexity of the dynamic switching between brain microstates.

[0056] In the process of EEG feature screening, in order to mine the key EEG features that characterize the effect of electroacupuncture, the present invention introduces the overlapping group sparse LASSO model into the deep belief network DBN model and re-encodes the features based on the neural network. The traditional LASSO uses L1 regularization penalty to implement a sparse model to learn useful low-level feature representations and suppress overfitting of the model. However, this model cannot take into account whether there is an inter-group relationship between different types of features. The model sparsity caused by the overlapping group LASSO model is at the feature group level. Through the overlapping structure between groups, similar features between groups are linked together, reducing the simultaneous activation of multiple groups due to the discrete existence of a large number of similar features in multiple groups. It can handle the problem of overlapping features between groups, which helps to improve the recognition rate of the model and make up for the shortcomings of the traditional sparse DBN network. Specifically:

[0057] First, the traditional EEG signal features are transformed into new unified modal features. By using the structured sparse constraint optimization method, the optimized weight of each feature dimension is calculated. The feature dimensions with higher weights are finally selected and become the multimodal features that are finally extracted and screened. The objective function used here is:

[0058]

[0059] In Group LASSO, the traditional EEG features of different frequencies are divided into G groups, I g is the characteristic subscript of the g group (g=1,…,G), is the weight of each group, λ is the penalty parameter, and β is the parameter to be estimated. Unlike the Lasso method, which sums up the absolute values ​​of the coefficients of each feature, what is summed up here is the L2 norm of the coefficients of each group. During the optimization process, this structure tries to select fewer groups (sparse between groups). This method produces a sparse effect both between feature groups and within groups. Some feature dimensions within a group and some similar features between groups are eliminated at the same time. After the screening, the key EEG features are input into the intelligent matching model of electroacupuncture parameters.

[0060] The intelligent matching module for electroacupuncture parameters mainly outputs the best electroacupuncture parameters based on the previously trained model and EEG features as input. Through a large amount of experimental data, different electroacupuncture parameters correspond to different EEG features, and the best treatment score for insomnia is the goal. The supervised classification model is trained through the fast compressed residual convolutional neural network FCRes-CNNs. The trained classification model can be directly applied to calculate the best electroacupuncture parameters in real time based on EEG features. In addition, transfer learning is used in the real-time closed-loop system to continuously iterate and optimize the model to improve the matching accuracy of electroacupuncture parameters.

[0061] The electroacupuncture stimulation module automatically changes the output waveform, pulse frequency and current intensity for stimulation according to the matched electroacupuncture parameters.

[0062] like Figure 5 As shown in the figure, the real-time display module visualizes the calculated EEG characteristics and electroacupuncture parameters to help doctors better understand the current patient conditions. At the same time, it provides doctors with the function of manually adjusting electroacupuncture parameters and controlling the start and end of electroacupuncture and EEG data collection.

[0063] The real-time display interface is implemented by Python+Pyqt architecture, and various human-computer interaction functions are realized through the PC monitor, including the setting of acquisition and experimental parameters, real-time EEG information decoding feedback display and optimal electroacupuncture stimulation parameter display. This solution can effectively perform online experimental verification of the algorithm, and the modular programming method has great potential for expanding the system functions.

[0064] The working principle of the present invention is as follows: Figure 4 As shown in the figure, the whole brain EEG signals of insomnia patients were collected during electroacupuncture. Figure 3 The proposed method calculates the patient's EEG microstate sequence and other features, and uses the deep learning training model to accurately match the electroacupuncture parameters based on the microstate and other feature database of existing patients before and after treatment, thereby changing the electroacupuncture stimulation parameters for electroacupuncture treatment; extract new microstates and other features based on the EEG signal after treatment, thereby forming a continuous two-way closed-loop interaction. The two-way here means that the EEG signal is extracted and matched with the model to the electroacupuncture parameters, and then the electroacupuncture performs current stimulation according to the electroacupuncture parameters to affect the EEG signal.

Claims

1. A two-way closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system for insomnia, characterized in that: It includes EEG acquisition module, OpenViBE signal interface and control platform module, real-time signal processing module, electroacupuncture parameter intelligent matching module, electroacupuncture stimulation module, and real-time display module; The EEG acquisition module includes a whole-brain electrode cap and an amplifier, which collects the whole-brain EEG signals of insomnia patients in real time and transmits them to the OpenViBE signal interface and control platform module; The OpenViBE signal interface and control platform module completes the time synchronization, real-time access, centralized collection and storage requirements in the process of EEG signal acquisition; The real-time signal processing module performs EEG preprocessing, EEG information decoding and EEG feature screening, and adopts an improved microstate complexity analysis method to comprehensively extract microstate discrete time series based on the global field energy peak potential topology information and trough potential topology information, providing objective detection indicators for the electroacupuncture stimulation effect; The electroacupuncture parameter intelligent matching module inputs the EEG feature vector into the deep learning model to achieve accurate matching of the optimal electroacupuncture stimulation parameters; The electroacupuncture stimulation module includes a control circuit, an output wire, an electroacupuncture clamp, and an acupuncture needle, and automatically performs corresponding current stimulation on specific acupuncture points according to the matched electroacupuncture parameters; The real-time display module visualizes the EEG information decoding process and the electroacupuncture stimulation parameters during the electroacupuncture process and provides a function for the doctor to make manual adjustments; The improved microstate complexity analysis method is implemented as follows: For a whole brain EEG signal with N channels [X1(t),X2(t),X3(t),…,X N (t)], calculate the global field power GFP of brain activity: A topology-based atomization and aggregation hierarchical clustering algorithm is used as the clustering algorithm for microstate analysis. The potential topology of local peak points and the potential topology of local trough points are clustered separately. The potential topology map of each local peak point is regarded as a family. All potential topology maps are initially regarded as independent groups, and the potential topology operation of local trough points is the same. The spatial correlation of each potential topology map with other topographic maps is calculated separately: Among them, u and v are two different potential topographic maps. The spatial correlation of each potential topological map with other topological maps is calculated respectively, and the topological map with the lowest total correlation is selected as the worst topological map; the current worst topological map is redistributed to the remaining group with the closest correlation, and the potential topological maps in this group are averaged to obtain a new potential topological distribution; after each iteration, a worst potential topological map is redistributed until the number of clusters reaches the preset number of categories; the above clustering calculations are performed on the potential topological maps of the local peak value point of GFP and the local valley value point, respectively, to obtain the potential topological map based on the local peak value point of GFP. The clustered microstates of the potential topology map of the local trough value points of GFP are compared; the potential topology map of the local peak value points of GFP is clustered into 4 types of microstate topography, and the potential topology map of the local trough value points of GFP is clustered into 2 types of microstate topography; the spatial correlation of the scalp potential topology map at each time point is compared with the standard microstate obtained by clustering, and the scalp potential is matched with the label of the standard microstate; the EEG signal is decoded into a microstate time series in which six microstates alternate. After redefining the microstate calculation process, the improved microstate analysis can obtain a microstate time series with a higher global explained variance GEV: Where m is the number of original scalp potential topological maps. The new microstate time series can represent the changes in brain states at more moments. After obtaining the microstate discrete sequence, the multi-scale entropy is calculated for the discretized sequence to obtain the complexity of the dynamic switching between brain microstates. The EEG feature screening implementation process is as follows: The overlapping group sparse LASSO model is introduced into the deep belief network DBN model, and the features are re-encoded based on the neural network; first, the traditional EEG signal features are transformed into new unified modal features, and the optimized weights of each feature dimension are calculated by adopting the structured sparse constraint optimization method; the feature dimensions with higher weights are selected, and the multimodal features are extracted and screened; the objective function used is: In Group LASSO, the traditional EEG features of different frequencies are divided into G groups, I g is the characteristic of group g, subscript g = 1,…,G, is the weight of each group, λ is the penalty parameter, and β is the parameter to be estimated.

2. A two-way closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system for insomnia according to claim 1, characterized in that: The OpenViBE signal interface and control platform module adopts the open source OpenViBE framework to realize real-time EEG signal acquisition and transmission between different modules, and transmission of EEG characteristics and electroacupuncture parameters, providing an infrastructure for a two-way closed-loop brain-computer interaction system.

3. A two-way closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system for insomnia according to claim 1, characterized in that: In the electroacupuncture parameter intelligent matching module, a precise matching model between EEG characteristics and optimal electroacupuncture stimulation parameters is implemented based on deep learning, and transfer learning is used in a real-time closed-loop system to continuously iterate and optimize the model to improve the matching accuracy of electroacupuncture parameters.

4. A two-way closed-loop brain-computer interactive electroacupuncture parameter intelligent matching system for insomnia according to claim 1, characterized in that: The electroacupuncture stimulation parameters include output waveform, pulse frequency and current intensity.

Images