Active induction type rehabilitation system and equipment based on MI-BCI

Through the MI-BCI active induction rehabilitation system, signal processing and neural network technology are used to improve the quality of MI-EEG signal, solve the problem of weak signal and interference, realize high-precision dysfunction treatment, and alleviate the problem of insufficient medical resources.

CN118987485AActive Publication Date: 2024-11-22CHINA REHABILITATION RES CENT

Patent Information

Application Number
CN202410986358.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-22
Estimated Expiration
2044-07-23

AI Technical Summary

Technical Problem

The MI-EEG signal is weak, and the signal quality is affected by the subject's concentration and the excitation of motor cortex neurons. There are electroophthalmic, electromyography and volume conduction effects, resulting in a decrease in signal-to-noise ratio. The existing methods cannot fundamentally improve the signal. quality.

Method used

The active induction rehabilitation system adopts MI-BCI, including MI-EEG signal acquisition and preprocessing module, data feature extraction and classification algorithm module and FNS active induction module, through notch filtering, FastICA algorithm, fast Fourier transform, OVR -CSP method and VGG-LSTMnet network structure extract the power spectral density and ERD/ERS characteristics of EEG signals, realize high robustness and high precision feature extraction and classification, adjust the EEG threshold and current intensity, combine vision and voice Text prompts for active induction.

Benefits of technology

It realizes the high robustness and high precision feature extraction and classification of MI-BCI, and actively induces FNS to improve patient dysfunction, enhance the central control over the peripheral, alleviate the problems of insufficient therapists and uneven medical resources, and improves rehabilitation benefits. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118987485A_ABST
    Figure CN118987485A_ABST
Patent Text Reader

Abstract

The invention discloses an active induction type rehabilitation system and device based on MI-BCI, and belongs to the technical field of brain-computer interfaces. The rehabilitation system comprises an MI-EEG signal acquisition and preprocessing module, a data feature extraction and classification algorithm module and an FNS active induction module. The MI-EEG signal acquisition and preprocessing module acquires original EEG electroencephalogram signals by adopting a signal acquisition mode of combining MI with EEG, and performs data preprocessing; the data feature extraction and classification algorithm module extracts ERD / ERS features of rhythms of an alpha frequency band and a beta frequency band of the electroencephalogram signals through a one-to-many common space mode OVR-CSP method, and then classifies the features through a VGG-LSTMnet network; and the FNS active induction module is used for completing active induction according to the classification result of the data feature extraction and classification algorithm module. The device has the advantage of realizing precise rehabilitation under a non-invasive condition, and active induction can be realized on the basis of a neuromuscular feedback pathway so as to improve the dysfunction of a patient caused by the blockage of the neuromuscular pathway.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of brain-computer interface technology, and in particular to an active induction rehabilitation system and equipment based on MI-BCI. Background Art

[0002] Brain-computer interface (BCI) technology is a new type of rehabilitation therapy technology that is controlled by neural activity and can provide real-time feedback output. It can establish a communication and control channel between the brain and the external environment that is independent of peripheral nerve and muscle conduction by decoding individual psychological intentions, converting neural activity into internal or external command signals, thereby realizing direct interaction between the brain and the external environment.

[0003] Electroencephalography (EEG) is easy to collect and has the advantages of being non-destructive, repeatable, and having high temporal and spatial resolution. Therefore, using EEG as a non-invasive way to collect BCI signals can avoid safety issues such as invasive BCI surgical risks and immune responses, and has unique application advantages and value in the field of rehabilitation medicine.

[0004] Motor imagery (MI) is a way to induce EEG signals. It can induce event-related desynchronization of μ rhythm and β rhythm in the primary motor area of ​​the brain without external stimulation and obvious action output. By analyzing the MI tasks corresponding to different feature changes, the user's real movement intention can be understood and specific real-life interactive tasks can be performed spontaneously. Therefore, the use of MI-EEG provides the possibility for the safety and easy promotion of BCI.

[0005] Functional neuromuscular stimulation (FNS) outputs current to stimulate peripheral muscles, and through the model of "stimulated muscle-nerve that controls the muscle-nerve center-efferent nerve that controls the muscle-transmitted to the stimulated muscle again", it promotes nerve remodeling, thereby strengthening the central control over the periphery and promoting the recovery of nerve function. Therefore, FNS technology has been applied in clinical practice, but its usage scenarios are mostly focused on peripheral neuromuscular stimulation, and the central involvement is relatively small, and it is impossible to directly establish a pathway for the central nervous system to control active movement. Therefore, MI-BCI can be selected as the "switch" for active induction to build an active induction rehabilitation system based on MI-BCI, in which active induction is achieved through FNS.

[0006] However, MI-EEG signals are weak. In clinical applications, small sample MI-EEG is usually used. The signal quality is directly affected by the subject's MI concentration and the excitability of the motor cortex neurons. In addition, simple symbols or pictures are often used as prompt interfaces in signal acquisition, which lack relevance to the subject. The intensity and continuity of the subject's active MI are difficult to guarantee, resulting in a decrease in signal quality. In addition, interference such as electrooculography and electromyography and volume conduction effects will also aggravate signal distortion and reduce the MI-EEG signal-to-noise ratio. Although the use of wavelet transform, blind source separation and other methods to remove artifacts can improve the signal-to-noise ratio to a certain extent, it cannot fundamentally improve the quality of MI-EEG signals. Summary of the invention

[0007] In response to the above-mentioned problems, the present invention aims to provide an active induction rehabilitation system and equipment based on MI-BCI, which not only has the advantage of achieving precise rehabilitation under non-invasive conditions, but also can realize active induction on the basis of neuromuscular feedback pathways, so as to improve the functional impairment of patients caused by obstruction of neuromuscular pathways, and further effectively alleviate the problems of insufficient therapists and uneven medical resources.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] The active induction rehabilitation system based on MI-BCI includes MI-EEG signal acquisition and preprocessing module, data feature extraction and classification algorithm module and FNS active induction module;

[0010] The MI-EEG signal acquisition and preprocessing module uses MI combined with EEG signal acquisition to collect raw EEG signals and perform data preprocessing;

[0011] The data feature extraction and classification algorithm module is used to extract features and classify the preprocessed EEG signals;

[0012] The FNS active induction module is used to complete active induction based on the results of data feature extraction and classification algorithm module classification.

[0013] Furthermore, the method for the MI-EEG signal acquisition and preprocessing module to perform data preprocessing on the original EEG brain wave signal includes the following steps:

[0014] (1): Perform notch filtering on the original EEG signal to remove power frequency interference;

[0015] (2): Remove low-frequency drift and high-frequency noise through band-pass filtering from 0.5 Hz to 50 Hz;

[0016] (3): Use FastICA algorithm to eliminate electrooculographic artifacts;

[0017] (4) The frequency was accurate to 1 Hz using the fast Fourier transform method, and the time window of the EEG signal was intercepted from 14 channels;

[0018] (5): Calculate the power spectral density of the EEG signal as the primary feature of the input data feature extraction and classification algorithm module.

[0019] Furthermore, the power spectral density of EEG signals is calculated as follows:

[0020]

[0021] Where P represents the power spectrum density; w k represents discrete frequency components, N represents signal length, y(t) represents time domain signal, and i represents imaginary unit.

[0022] Furthermore, the specific operation of the data feature extraction and classification algorithm module to extract features of the preprocessed EEG signals includes the following steps: the ERD / ERS features of the α-band and β-band rhythms of the EEG signals are extracted from the EEG power spectral density data input by the MI-EEG signal acquisition and preprocessing module through a one-to-many co-spatial pattern OVR-CSP method.

[0023] Furthermore, the specific operation of the OVR-CSP method for extracting ERD / ERS features of the α-band and β-band rhythms of EEG signals includes the following steps:

[0024] Step 1: Consider only three channels C3, Cz and C4 for the preprocessed EEG signals;

[0025] Step 2: Calculate the brain area maps of the frequency band (8-14 Hz) and the frequency band (14-30 Hz), mapping ERD in red and ERS in blue.

[0026] Furthermore, the data feature extraction and classification algorithm module adopts the network structure VGG-LSTMnet which is a hybrid of LSTM and VGG to classify the ERD / ERS features of the extracted EEG signal α-band and β-band rhythms.

[0027] Furthermore, the VGG-LSTMnet network model includes three convolutional layers and one 1-D Conv layer. The first convolutional layer is a two-layer stacked structure with 32 convolution kernels of size [3×3], the second convolutional layer is a two-layer stacked structure with 64 convolution kernels of size [3×3], and the third convolutional layer is a single-layer structure with 128 convolution kernels of size [3×3]. Each convolutional layer corresponds to an LSTM moment unit structure. The output result of the 1-D Conv layer is parallelly fused with the LSTM output and input into the fully connected layer together. The fully connected layer still adopts a two-layer stacked structure. The first layer is a tiled structure, which is converted into a 1×512 vector through a fully connected layer to complete the final classification. According to the number of categories required by the sample label, it finally enters the Softmax layer to complete the final solution of the multi-classification problem.

[0028] Furthermore, the FNS active induction module relies on the Pycharm Community Edition software to complete active induction, adjusts the EEG threshold according to the patient's test conditions, adjusts the current intensity according to the patient's tolerance level, collects EEG signals and plays guidance videos for patients at the same time, and prompts patients to perform MI tasks independently through visual stimulation and voice and text. When the EEG device collects MI information and reaches the active induction threshold, it will trigger FNS, stimulate the corresponding nerves and muscles through electrode conduction, and complete active induction.

[0029] Furthermore, the present invention also includes an active induction rehabilitation device based on MI-BCI, comprising at least one processor and a memory communicatively connected to the processor, wherein the memory stores the active induction rehabilitation system based on MI-BCI as described in any one of claims 1 to 8, so that the processor can execute instructions of the active induction rehabilitation system based on MI-BCI.

[0030] The beneficial effects of the present invention are:

[0031] 1. The active induction rehabilitation system based on MI-BCI in the present invention realizes a highly robust and high-precision feature extraction method of MI-BCI, which can meet the general data dimension reduction and accurate classification methods of different types of functional disorders; and uses MI-BCI to control FNS to actively induce neuromuscular stimulation to improve patients' functional disorders, stimulate the peripheral nervous system to generate feedback to reshape brain function, thereby strengthening the central regulation of the periphery.

[0032] 2. The active induction rehabilitation system based on MI-BCI in the present invention can not only realize precise rehabilitation under non-invasive conditions, but also realize active induction based on the neuromuscular feedback pathway to improve the functional impairment caused by the obstruction of the neuromuscular pathway, so as to maximize the rehabilitation benefits of the patients, and further alleviate the problems of insufficient therapists and uneven medical resources.

[0033] 3. The feature extraction and classification method in the present invention is simple and robust, and is suitable for data sets with a small number of subjects and a small amount of data (small sample). BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a brain region diagram showing ERD and ERS phenomena in C3, Cz and C4 channels when imagining hand movement in the present invention;

[0035] Figure 2 Schematic diagram of the VGG-LSTMnet network model structure in the present invention. DETAILED DESCRIPTION

[0036] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0037] Embodiment 1:

[0038] Embodiment 1 provides an active induction rehabilitation system based on MI-BCI, including a MI-EEG signal acquisition and preprocessing module, a data feature extraction and classification algorithm module, and a FNS active induction module.

[0039] The MI-EEG signal acquisition and preprocessing module uses MI combined with EEG signal acquisition to collect the original EEG brain wave signals;

[0040] The operations of preprocessing the collected raw EEG signals include: (1) performing a 50Hz notch filter on the raw EEG signals to remove power frequency interference; (2) performing a bandpass filter from 0.5Hz to 50Hz to remove low-frequency drift and high-frequency noise; (3) eliminating electrooculogram artifacts using the FastICA algorithm; (4) using the fast Fourier transform method to accurately calculate the frequency of the EEG signals to 1Hz, and intercepting the time window of the EEG signals from 14 channels; (5) calculating the power spectral density (PSD) of the EEG signals, and inputting it as a primary feature into the data feature extraction and classification algorithm module for further feature extraction and classification. The power spectral density extracted in the present invention is the average power at unit frequency resolution, and its calculation formula is as follows:

[0041]

[0042] Where P represents the power spectrum density; w k represents discrete frequency components, N represents signal length, y(t) represents time domain signal, and i represents imaginary unit.

[0043] Furthermore, the data feature extraction and classification algorithm module is used to extract features and classify the EEG signals preprocessed by the MI-EEG signal acquisition and preprocessing module.

[0044] Specifically, data feature extraction is to extract ERD / ERS features of the α-band and β-band rhythms of the EEG signal from the EEG power spectral density data input by the MI-EEG signal acquisition and preprocessing module using a one-versus-rest commonspatial pattern (OVR-CSP) method.

[0045] Data classification is achieved by introducing generative adversarial networks (GANs) into the convolutional neural network (CNN) structure and optimizing network weights to build a highly robust semi-supervised learning model to accurately classify MI-EEG brain electrical signals. The feature extraction and classification methods in the present invention are simple and robust, and are suitable for data sets with a small number of subjects and a small amount of data (small sample).

[0046] The FNS active induction module is used to complete active induction. After the original MI-EEG signal is processed by the MI-EEG signal acquisition and preprocessing module, the data feature extraction and classification algorithm module, it is converted into motor imagery classification data that can be recognized by the BCI device control terminal, and then the BCI controls the functional neuromuscular electrical stimulation (FNS) to generate impulses to act on peripheral nerves and muscles, thereby achieving the purpose of active induction. It should be noted that the FNS active induction module is an existing technology and has been used clinically. The functional neuromuscular electrical stimulation (FNS) outputs current to stimulate peripheral muscles. Through the mode of "stimulated muscle-nerve that controls muscle-nerve center-efferent nerve that controls muscle-transmitted to the stimulated muscle again", it promotes nerve remodeling, thereby strengthening the control of the center over the periphery and playing a role in promoting the recovery of nerve function.

[0047] The active induction rehabilitation system based on MI-BCI in the present invention relies on Pycharm Community Edition software, adjusts the EEG threshold according to the patient's test situation, adjusts the current intensity according to the patient's tolerance, collects EEG signals and performs corresponding data processing. At the same time, a guidance video is played for the patient. The FNS active induction module converts the patient's EEG signals into motor imagery classification data that can be recognized by the BCI device control end, and prompts the patient to perform MI tasks independently through visual stimulation and voice and text. When the EEG device collects MI information and reaches the active induction threshold, it will trigger FNS, stimulate the corresponding nerves and muscles through electrode conduction, and complete active induction.

[0048] Embodiment 2:

[0049] Embodiment 2 Based on Embodiment 1, a specific implementation of a data feature extraction and classification algorithm module is provided.

[0050] The specific operation of extracting ERD / ERS features of EEG signal α-band and β-band rhythms from the preprocessed EEG power spectral density data using the OVR-CSP method includes the following steps: Analyze the features extracted by CSP. When performing two types of motor imagery, all subjects will show corresponding ERD and ERS phenomena in the cerebral cortex. For example, two different motor imagery tasks are taken from dataset one, namely imagining hand and foot lifting movements. The goal is to generate ERDS mapping for each of the two tasks. First, import 5s data. The dataset contains multiple channels, and only three channels C3, Cz, and C4 are considered. Then, the brain area maps of the frequency band (8-14Hz) and the frequency band (14-30Hz) are calculated, and the ERD is mapped in red and the ERS is mapped in blue. When imagining hand movement, the ERD and ERS phenomena displayed by the C3, Cz, and C4 channels are shown in the attached figure. Figure 1 Finally, a test corrected for multiple comparisons within channels was performed to estimate significant ERDS values.

[0051] Furthermore, in this embodiment, by introducing generative adversarial networks (GANs) into the convolutional neural network (CNN) structure and optimizing the network weights, a highly robust semi-supervised learning model is built to achieve the accuracy of MI-EEG EEG signals; specifically, the accuracy of MI-EEG EEG signals is achieved through the VGG-LSTMnet neural network structure.

[0052] The advantage of the VGG network architecture is that it contains a regularization term, which is to decompose a 5×5 or 7×7 convolution into a stack of multiple 3×3 convolutions. The fully connected layer is replaced by a 1×1 convolution kernel. By flexibly modifying the network structure and pre-initializing some network layers, the model training is faster and more accurate than the general network. The number of convolution layers and convolution kernels of the VGG network structure is adjusted to obtain a convolutional neural network model with a smaller number of network layers, named VGG-LSTMnet. The model framework of the VGG network structure is shown in Table 1 below.

[0053] Table 1 VGG network model framework

[0054]

[0055] By adjusting the number of convolutional layers and convolutional kernels of the VGG network structure, a convolutional neural network model with a smaller number of network layers is obtained, named VGG-LSTMnet, and its model framework is shown in Table 2 below.

[0056] Table 2 VGG-LSTMnet network model framework

[0057]

[0058] The VGG-LSTMnet neural network structure of the present invention is shown in the attached figure. Figure 2As shown in the figure, it specifically includes three convolution layers. The first layer is a two-layer stacked structure with 32 convolution kernels of size [3×3], denoted as C1; the second layer is a two-layer stacked structure with 64 convolution kernels of size [3×3], denoted as C2; the third layer is a single-layer structure with 128 convolution kernels of size [3×3], denoted as C3. The long short-term memory network LSTM needs to capture different temporal pattern features across multiple frames, corresponding to the three LSTM moment unit structures. The inputs at time t-1, time t, and time t+1 come from the features of the shallow, deeper, and deep layers of the structure respectively (the transformation of spatial features and temporal features, mapping high-dimensional feature vectors to low-dimensional space). After passing through the maximum pooling layer in each layer, the vector is expanded into a vector through the Reshape layer as the input structure of the LSTM network, realizing serial fusion. Finally, the LSTM layer and the layer composed of 128 cells with different numbers of storage units obtain the best results. It should be noted that in this model, a 1-D Conv layer, i.e., a 1×1 convolution layer, with 64 convolution kernels, is added after the output. 1×1 convolution is actually a linear combination of each pixel on different channels. It is used to filter adjacent elements of the one-dimensional input and retain the original planar structure of the feature map, regulate the depth, and reduce the dimension of the feature map as much as possible. Then the output of the 1-D Conv layer is fused in parallel with the LSTM output. The Reshape layer changes the dimension of the input data, excluding the dimension of the number of samples (batch size), and keeps the content unchanged. After fusion, it is input into the fully connected layer together. The fully connected layer still uses a two-layer stacked structure. The first layer is a tiled structure, which is converted into a 1×512 vector through a fully connected layer. The purpose is to complete the final classification. According to the number of categories required by the sample label, it finally enters the Softmax layer to complete the final solution of the multi-classification problem.

[0059] In the present invention, the ERD / ERS features of the α-band and β-band rhythms of the EEG signal are input into the VGG-LSTMnet neural network model. The network model decomposes the 5×5 or 7×7 convolution into a stack of multiple 3×3 convolutions. The fully connected layer is replaced by a 1×1 convolution kernel. The network structure can be flexibly modified and the network layer can be pre-initialized, so that the model training is faster than the general network, with excellent generalization ability and higher accuracy. The classification accuracy is as high as 90.38% after experimental comparison, which can be used in high-performance motor imagery multi-task classification and recognition systems. The interface realizes accurate classification of MI-EEG for limbs, tongue, and eyes while ensuring small samples and high robustness. The classification results reach the active induction threshold, induce the patient's active movement intention, and realize active rehabilitation treatment for patients with motor, visual, language and cognitive dysfunction.

[0060] The classification performance of the VGG-LSTMnet neural network model in this embodiment is verified. The specific parameters of the average classification performance obtained by 5-fold cross validation are shown in Table 3. The average accuracy of different categories obtained from different classification algorithms and experiments of each subject are shown in Table 3. The experimental results prove the effectiveness of the VGG-LSTMnet neural network model in the present invention.

[0061] Table 3 Classification performance of dataset 1

[0062]

[0063] Application examples:

[0064] The active induction rehabilitation system based on MI-BCI in the present invention is created based on Python language, and is carried out based on the deep learning framework Keras+Tensorlow of Theano, and the environment is Intel 8th generation processor (6 cores 5.2GHz), memory 32.0GByte DDR4, hard disk 500GB SSD+4TB SATA, GPU is GeForce RTX2080 video memory 16GByte. The data set adopts the international general GDF format, including left (right) hand (foot) movement, eye rotation (four directions), tongue MI EEG signal, sampling frequency 125Hz, sampling channel adopts 16 leads, and is collected by wearing an electrode cap according to the international 10-20 standard electrode placement method. The signals of the three channels of C3, C4, and Cz are closely related to the motor imagination of the brain. As the main reference electrode of the neural network classifier, it is supplemented by three midline electrodes (Fz, Oz, Pz) and two ear electrodes (A1, A2).

[0065] The original EEG signal was notched at 50Hz to remove power frequency interference; low-frequency drift and high-frequency noise were removed by bandpass filtering from 0.5Hz to 50Hz. The FastICA algorithm with fixed-point iteration was introduced to eliminate electrooculographic artifacts and obtain the non-Gaussian maximum value of WTX. The fast Fourier transform method was accurate to 1Hz in frequency, and the time window of the EEG signal was intercepted from the 14 channels, and the required frequency band was intercepted by bandpass filtering.

[0066] The original data set is preprocessed to obtain the EEG signal after noise is filtered out, and then the ERD / ERS features of the α-band and β-band rhythms of the EEG signal are extracted using the OVR-CSP method. Finally, GANs are introduced into the CNN structure and the network weights are optimized to build a highly robust semi-supervised learning model to complete the accurate classification of MI-EEG.

[0067] The process of using this rehabilitation system to conduct active induced rehabilitation training for patients is as follows: determine the patient's body position, identify the target nerves and muscles for active induction, fully expose the skin, stick the electrode patch to the target point, set the FNS stimulation parameters, and select the appropriate stimulation mode, stimulation intensity, frequency, pulse width, treatment time and rehabilitation treatment cycle according to the patient's function and tolerance. The patient wears an EEG device, watches the MI task guidance video, and implements MI-BCI active induced training through visual, auditory and FNS multi-sensory input interaction. The specific implementation process includes the following steps:

[0068] (1) Comprehensively evaluate the patient's motor, sensory, cognitive and other functional states, determine the target neuromuscular system for active induction, and then determine the rehabilitation treatment prescription. The prescription should include but is not limited to: stimulation mode, stimulation intensity, frequency, pulse width, treatment time and rehabilitation treatment cycle.

[0069] (2) Select a quiet room as the treatment room to ensure that the patient can concentrate. The patient sits with his arms relaxed and his forearms fully exposed. The electrodes are placed 3 finger widths proximal to the radial end of the wrist and above the ulnar styloid process, and 3 finger widths distal to the lateral epicondyle of the wrist extensor muscles. Put the EEG cap on the patient's head and lay it flat. The electrodes are placed according to the 10-20 international standard lead system, with a total of 16 leads, namely FP1, FP2, F3, F4, F7, F8, C3, C4, T3, T4, T5, T6, P3, P4, O1, and O2. It can be seen from the EEG power spectrum that the power spectrum density of MI is mainly concentrated in the range of 0-30Hz. The frequency range of oscillatory cortical activity corresponding to the MI EEG signal is mainly reflected in the two frequency bands of α wave (8-13Hz) and β wave (13-30Hz). The required frequency band can be intercepted by bandpass filtering, and the EEG threshold can be calculated based on the ratio of α wave to β wave.

[0070] (3) Open Pycharm Community Edition software, adjust the EEG threshold according to the patient's test results, adjust the current intensity according to the patient's tolerance, and use the standard of being able to induce the patient's wrist dorsiflexion movement without causing pain. The frequency is 40Hz, the pulse width is 200-400μs, and the treatment lasts for 15 minutes, 3 times a week, for 4 weeks. Right-click and select "Run'BCI-hub'", select "COM5" as the serial port, and then click "Start Acquisition". While collecting EEG signals, play a guiding video for the patient, and prompt the patient to perform the motor imagery task of wrist dorsiflexion autonomously through visual stimulation and voice text. When the EEG device collects MI information, it triggers electrical stimulation, thereby inducing the patient's wrist dorsiflexion movement. When the patient does not perform the MI task or does not reach the EEG threshold, FNS active induction cannot be induced. The EEG threshold is a test conducted before treatment. Let the patient perform the MI task, observe the threshold that can stimulate FES, and set it as the training threshold. The threshold can be adjusted according to the patient's treatment situation.

[0071] (4) Ask the patient to watch the video and imagine the actions he or she is going to perform according to the video guidance. There is no need to actually move. When MI is successfully performed, the electrodes at the target position will be electrically stimulated. If there is any discomfort during the stimulation, please inform the therapist immediately and the stimulator will be turned off immediately to avoid discomfort to the patient.

[0072] Embodiment three:

[0073] Embodiment three provides an active induction rehabilitation device based on MI-BCI, comprising at least one processor and a memory communicatively connected to the processor, wherein the memory stores the active induction rehabilitation system based on MI-BCI described in Embodiment one, so that the processor can execute instructions of the active induction rehabilitation system based on MI-BCI.

[0074] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. Active induction rehabilitation system based on MI-BCI, characterized by: It includes MI-EEG signal acquisition and preprocessing module, data feature extraction and classification algorithm module and FNS active induction module; The MI-EEG signal acquisition and preprocessing module uses MI combined with EEG signal acquisition to collect raw EEG signals and perform data preprocessing; The data feature extraction and classification algorithm module is used to extract features and classify the preprocessed EEG signals; The FNS active induction module is used to complete active induction based on the results of data feature extraction and classification algorithm module classification.

2. The active induction rehabilitation system based on MI-BCI according to claim 1, characterized in that: The method for the MI-EEG signal acquisition and preprocessing module to perform data preprocessing on the original EEG brain electrical signal includes the following steps: (1): Perform notch filtering on the original EEG signal to remove power frequency interference; (2): Remove low-frequency drift and high-frequency noise through bandpass filtering from 0.5 Hz to 50 Hz; (3): Use FastICA algorithm to eliminate electrooculographic artifacts; (4) The frequency was accurate to 1 Hz using the fast Fourier transform method, and the time window of the EEG signal was intercepted from 14 channels; (5): Calculate the power spectral density of the EEG signal as the primary feature of the input data feature extraction and classification algorithm module.

3. The active induction rehabilitation system based on MI-BCI according to claim 2, characterized in that: The power spectral density calculation method of EEG signal is: Where P represents the power spectral density; w k represents discrete frequency components, N represents signal length, y(t) represents time domain signal, and i represents imaginary unit.

4. The active induction rehabilitation system based on MI-BCI according to claim 2, characterized in that: The specific operation of the data feature extraction and classification algorithm module to extract features of the preprocessed EEG signals includes the following steps: the ERD / ERS features of the α-band and β-band rhythms of the EEG signals are extracted from the EEG power spectral density data input by the MI-EEG signal acquisition and preprocessing module through a one-to-many co-spatial pattern OVR-CSP method.

5. The active induction rehabilitation system based on MI-BCI according to claim 4, characterized in that: The specific operation of OVR-CSP method for extracting ERD / ERS features of α-band and β-band rhythms of EEG signals includes the following steps: Step 1: Consider only three channels C3, Cz and C4 for the preprocessed EEG signals; Step 2: Calculate the brain area maps of the frequency band (8-14 Hz) and the frequency band (14-30 Hz), mapping ERD in red and ERS in blue.

6. The active induction rehabilitation system based on MI-BCI according to claim 4, characterized in that: The data feature extraction and classification algorithm module uses a hybrid LSTM and VGG network structure VGG-LSTMnet to classify the ERD / ERS features of the extracted EEG signal α-band and β-band rhythms.

7. The active induction rehabilitation system based on MI-BCI according to claim 6, characterized in that: The VGG-LSTMnet network model includes three convolutional layers and one 1-D Conv layer. The first convolutional layer is a two-layer stacked structure with 32 convolution kernels of size [3×3], the second convolutional layer is a two-layer stacked structure with 64 convolution kernels of size [3×3], and the third convolutional layer is a single-layer structure with 128 convolution kernels of size [3×3]. Each convolutional layer corresponds to an LSTM moment unit structure. The output result of the 1-D Conv layer is parallelly fused with the LSTM output and input into the fully connected layer together. The fully connected layer still adopts a two-layer stacked structure. The first layer is a tiled structure, which is converted into a 1×512 vector through a fully connected layer to complete the final classification. According to the number of categories required by the sample label, it finally enters the Softmax layer to complete the final solution of the multi-classification problem.

8. The active induction rehabilitation system based on MI-BCI according to claim 7, characterized in that: The FNS active induction module relies on the Pycharm Community Edition software to complete active induction, adjusts the EEG threshold according to the patient's test conditions, adjusts the current intensity according to the patient's tolerance level, collects EEG signals and plays guidance videos for patients at the same time, and prompts patients to perform MI tasks independently through visual stimulation and voice and text. When the EEG device collects MI information and reaches the active induction threshold, it will trigger FNS, stimulate the corresponding nerves and muscles through electrode conduction, and complete active induction.

9. An active induction rehabilitation device based on MI-BCI, characterized in that: It includes at least one processor and a memory communicatively connected to the processor, wherein the memory stores the active induction rehabilitation system based on MI-BCI as described in any one of claims 1 to 8, so that the processor can execute instructions of the active induction rehabilitation system based on MI-BCI.

Citation Information

Patent Citations

  • Common spatial pattern and deep learning method based on brain-computer interface auxiliary rehabilitation therapy

    CN107958213A

  • Key technology method for brain-computer interface of strokerehabilitation system

    CN109620223A

  • Visual image reconstruction system based on brain-computer interface

    CN111539331A

  • Brain-computer interface method for cerebral apoplexy rehabilitation system based on generative adversarial network

    CN112370066A

  • An electroencephalogram emotion classification method based on time-space-frequency joint features

    CN113288146A

Cited By

  • Complex action-based space-time frequency domain feature fusion action recognition method and system

    CN121327782A