Motion intention decoding method and system based on slow cortex potential

Abstract

The invention discloses a motion intention decoding method and system based on slow cortical potential, and relates to the technical field of machine-assisted motion, and the method comprises the steps: firstly carrying out the normal form training of a target person, and extracting the slow cortical potential in a resting state and a motion state; preprocessing the slow cortex potential, wherein the preprocessing comprises time domain filtering and frequency domain filtering; feature extraction is carried out on the preprocessed slow cortex potential in the resting state and the motion state, and differential features in the resting state and the motion state are obtained; inputting the difference features in the resting state and the motion state and the corresponding resting label and the motion label into an SVM model for training to obtain a motion intention decoding model; the to-be-decoded slow cortical potential of the target person is obtained in real time, preprocessing and feature extraction are conducted, differential features of the to-be-decoded slow cortical potential are obtained and input into the motion intention decoding model, the motion intention of the target person at the current moment is obtained, and the motion intention decoding accuracy and efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of machine-assisted motion technology, and more specifically to a method and system for decoding motion intention based on slow cortical potentials. Background Art

[0002] Hemiplegia is a motor dysfunction caused by nerve damage. The clinical treatment of hemiplegic patients is mainly one-on-one rehabilitation therapy under the guidance of rehabilitation therapists. This method is time-consuming and costly. The rehabilitation exoskeleton robot combines robotic technology with rehabilitation medicine, which can free rehabilitation therapists from major physical labor and develop better rehabilitation plans for patients to improve rehabilitation effects. There are two types of rehabilitation training: active and passive. Studies have shown that active rehabilitation has more effective rehabilitation effects than passive rehabilitation. Therefore, extracting the patient's movement intention and integrating it into the rehabilitation exoskeleton robot is an effective way to improve the patient's subjective initiative and achieve more efficient rehabilitation effects.

[0003] Brain-computer interface is a discipline that combines hardware, software, and algorithms. It uses hardware equipment to collect cerebral cortex signals when the subject is performing an activity, analyzes the EEG signals, and identifies the subject's true intentions, so as to achieve the purpose of directly obtaining the subject's true thoughts without relying on peripheral nerves and muscle tissue, and realize the interaction or control of the human brain and the external environment. Brain-computer interfaces are divided into invasive and non-invasive types. Although invasive EEG has the advantages of stable data collection, less interference signals, and high recognition accuracy, it is harmful to the subject to a certain extent and has limited application. Non-invasive EEG directly collects the subject's scalp EEG signals, so it does not harm the subject and has a high degree of acceptance.

[0004] Another modality of spontaneous EEG is the slow cortical potential, which is the signal from the lowest frequency band in the spontaneous EEG component. It has a large difference between positive and negative potentials and usually lasts for more than 0.5 seconds, or even several seconds. Negative slow cortical potentials are usually associated with movement-related cortical activity, while positive slow cortical potentials are usually associated with weakened cortical activity. The generation of slow cortical potentials is usually related to movement. Therefore, slow cortical potentials are also called movement-related cortical potentials (MRCP). MRCP represents the brain's preparation process in the early stages before performing a movement. It is the change in activity of the cerebral cortex during movement imagination or execution during the anticipation, preparation, and prediction of the execution of the action, which is associated with the user's movement intention.

[0005] By analyzing and interpreting the characteristics and changes of MRCP, capturing and parsing movement intentions in advance, and applying rehabilitation control strategies within a specific time window, the system can respond more quickly and accurately, reducing the mental load. At the same time, MRCP can also be used as a brain switch to trigger robotic devices or functional electrical stimulation to provide assistance or support to users.

[0006] However, existing technologies for decoding MRCP have problems such as high latency, poor subject adaptability, insufficient decoding accuracy, and low efficiency. Therefore, facing the scene of motion intention recognition in rehabilitation exoskeleton robots, how to use MRCP to decode motion intention, improve the accuracy of decoding MRCP, reduce latency, and improve efficiency are problems that technicians in this field need to solve urgently. Summary of the invention

[0007] In view of this, the present invention provides a method and system for decoding movement intention based on slow cortical potentials, which are used to improve the efficiency and accuracy of movement intention decoding.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A method for decoding movement intention based on slow cortical potentials comprises the following steps:

[0010] Step 1: Conduct paradigm training on the target person to extract the slow cortical potential MRCP in both resting and moving states;

[0011] Step 2, preprocessing the slow cortical potential MRCP, including time domain filtering and frequency domain filtering;

[0012] Step 3, extracting features of the pre-processed slow cortical potential MRCP in the resting and moving states by differential method, and obtaining differential features in the resting and moving states;

[0013] Step 4: Input the differential features in the resting and moving states and the corresponding resting labels and moving labels into the SVM model for training to obtain a motion intention decoding model;

[0014] Step 5: Obtain the target person's slow cortical potential MRCP to be decoded in real time, perform preprocessing and feature extraction according to the methods of steps 2 to 3, obtain the differential features of the slow cortical potential MRCP to be decoded, and input them into the motion intention decoding model to obtain the target person's motion intention at the current moment.

[0015] Optionally, in step 1, the process of performing paradigm training on the target personnel is:

[0016] Step 1.1. Preparation: A touch pad was set at the initial position and the target position. The target person sat on a chair and placed his hand on the contact area of ​​the touch pad as the starting position. The experimental paradigm was trial-based. The prompt information was displayed on the computer screen in front of the subject. At the beginning of the experiment, the target person first received a prompt to inform the number of remaining trials. At the beginning of each trial, the target person was asked to look at the center of the screen.

[0017] Step 1.2: After the experiment begins, the target person receives a resting reminder, and the subject completes the resting state without making any movements, while the single-chip microcomputer system is used to collect data;

[0018] Step 1.3: Prompt slow movement, and the target person completes the finger sliding action as required, that is, completes the straight line action from the initial position to the target position;

[0019] In the above steps, each time the hand is placed in the initial position, a new trial prompt starts. When the hand leaves the initial position, the microcontroller detects the switch signal and sends it to the host computer. The host computer sends the signal to the EEG acquisition supporting software. The software opens a trigger function to synchronize external data with the EEG signal. Once the hand touches the target touch piece, the microcontroller detects the time point of the finger and leaving, that is, the switch signal, and sends it to the host computer to eliminate the erroneous trials in the time interval between the initial position and the target position.

[0020] Optionally, in step 2, the method for preprocessing the slow cortical potential MRCP is:

[0021] Step 2.1, time domain filtering, using Laplace spatial filtering to remove local background activity;

[0022] Step 2.2: Frequency domain filtering, using a 4th-order Butterworth causal filter with a frequency band filtering range of 0.1 to 40 Hz.

[0023] Optionally, in step 2.1, for electrode channel i, the calculation process of the Laplace spatial filter is as follows:

[0024]

[0025]

[0026] in, is the filtered signal, x i is the signal of channel i, x j is the signal of channel j, S i is the set of channels adjacent to channel i, g ij is the weight matrix. The simplified version of the Laplacian spatial filter considers adjacent electrodes to be equidistant, i.e., g ij=1 / N i , where N i is the number of neighboring electrodes of electrode i.

[0027] Optionally, in step 3, the process of extracting features by difference is:

[0028] Let the slow cortical potential MRCP be x[j], the signal length be N, and the sampling frequency be f S , then the calculation expressions of A and B features are as follows:

[0029]

[0030]

[0031] Among them, A is the first-order backward difference of the signal, B is the second-order backward difference of the signal, Δ represents the forward difference, It represents backward difference, N is the signal length after windowing, and the preprocessed EEG signal is divided into windows, and then the A and B features of each window are extracted respectively.

[0032] A motion intention decoding system based on slow cortical potentials, comprising:

[0033] The paradigm training module is used to conduct paradigm training on the target personnel and extract the slow cortical potential MRCP in both resting and moving states;

[0034] A preprocessing module, used for preprocessing the slow cortical potential MRCP, including time domain filtering and frequency domain filtering;

[0035] A feature extraction module is used to extract features of the pre-processed slow cortical potential MRCP in the resting and moving states by differential means, so as to obtain differential features in the resting and moving states;

[0036] The model training module is used to input the differential features in the resting and moving states and the corresponding resting labels and moving labels into the SVM model for training to obtain the movement intention decoding model;

[0037] The intention decoding module is used to obtain the target person's slow cortical potential MRCP to be decoded in real time, perform preprocessing and feature extraction, obtain the differential features of the slow cortical potential MRCP to be decoded, and input them into the movement intention decoding model to obtain the movement intention of the target person at the current moment.

[0038] It can be seen from the above technical solutions that the present invention provides a method and system for decoding movement intention based on slow cortical potentials, which has the following beneficial effects compared with the prior art:

[0039] (1) The present invention proposes a scheme framework for motor intention decoding based on slow cortical potential MRCP, which mines the potential information of MRCP in the expression of motor intention and improves the accuracy of motor intention decoding.

[0040] (2) The present invention proposes a training paradigm. Compared with conventional training methods, the present invention uses two single-chip microcomputer-controlled touch panels as trigger signals for motion acquisition. Through paradigm training, it ensures that the MRCP data extracted from the target person in resting and moving states are highly representative and consistent, further improving the reliability of the data.

[0041] (3) The present invention effectively removes background noise and irrelevant signals by performing time domain filtering (Laplacian spatial filtering) and frequency domain filtering (Butterworth causal filter) in the preprocessing step, and highlights the features related to motion intention, thereby improving the signal quality and decoding accuracy.

[0042] (4) The present invention proposes a differential feature and a calculation method thereof. This method is used to extract EEG signal features. Compared with conventional time domain and frequency domain features, it can capture subtle changes in MRCP signals and further improve the decoding accuracy.

[0043] In summary, the present invention has achieved remarkable progress in decoding accuracy, efficiency, real-time performance, etc., and has provided new ideas and methods for the research and application in the field of motion intention recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 It is a schematic diagram of the principle of the method of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The embodiment of the present invention discloses a method for decoding movement intention based on slow cortical potentials, see Figure 1 , including the following steps:

[0048] Step 1: Conduct paradigm training on the target personnel to extract the slow cortical potential MRCP in both resting and moving states. The slow cortical potential MRCP is a special EEG signal. By collecting and analyzing this signal, the decoding and classification of the two signals can be achieved.

[0049] The experimental process of paradigm training for target personnel is as follows:

[0050] Step 1.1. Preparation: The subject (i.e., the target person) sits comfortably in a chair with his hand placed on the contact area of ​​the touch pad as the starting position. The experimental paradigm is trial-based. The prompt information is displayed on the computer screen in front of the subject. At the beginning of the experiment, the subject will first receive a 2-second prompt to inform the number of remaining trials. At the beginning of each trial, the subject is asked to look at the center of the screen and minimize blinking and throat activity.

[0051] Step 1.2: After the experiment begins, the subject will receive a 2-second rest prompt. The subject will rest for 2 seconds without making any movements.

[0052] Step 1.3: Prompt the subject to perform slow movements, and the subject should complete the finger sliding action as required, which should last for more than 2 seconds. The sliding action is to complete the straight line movement from the initial position to the target position, as uniform and stable as possible.

[0053] Among them, a touch piece is set at the initial position and the target position, and a single-chip microcomputer system is used for data collection. Every time the hand is placed in the initial position, a new trial prompt starts, which can ensure a completely autonomous training mode. When the hand leaves the initial position, the single-chip microcomputer detects the switch signal and sends it to the host computer. The host computer receives the signal and sends it to the EEG acquisition supporting software. The software opens a trigger function to synchronize external data with the EEG signal. In addition, once the hand touches the target touch piece, the single-chip microcomputer detects the time point (switch signal) of the finger and leaving, and sends it to the host computer to eliminate the time interval error trials between the initial position and the target position.

[0054] Each subject performed 3 sequences, each of which consisted of 4 rounds, and each round contained 40 trials, including 20 fast and 20 slow movements, which means that each subject had 480 trials. Each round lasted about 15 minutes, and the subject could arrange the rest time between rounds according to the fatigue level of the subject. Each sequence took about 60 minutes. There was a 2-day interval between each sequence.

[0055] Step 2: preprocessing the slow cortical potential MRCP collected in step 1 above, including time domain filtering and frequency domain filtering.

[0056] The method for preprocessing slow cortical potential MRCP is:

[0057] Step 2.1, time domain filtering, using Laplace spatial filtering to remove local background activity. For electrode channel i, the calculation process of the Laplace spatial filter is as follows:

[0058]

[0059]

[0060] in, is the filtered signal, x i is the signal of channel i, x j is the signal of channel j, S i is the set of channels adjacent to channel i, g ij is the weight matrix. The simplified version of the Laplacian spatial filter considers adjacent electrodes to be equidistant, i.e., g ij =1 / N i , where N i is the number of neighboring electrodes of electrode i.

[0061] Step 2.2: Frequency domain filtering, using a 4th-order Butterworth causal filter. Considering that the active EEG signals of the brain are mainly concentrated in other frequency bands such as theta (4-7 Hz), alpha (8-13 Hz) and beta (13-30 Hz), the frequency band filtering range is 0.1 to 40 Hz.

[0062] Step 3: Extract features of the preprocessed slow cortical potential MRCP in the resting and moving states by differential method to obtain differential features in the resting and moving states.

[0063] Assume that the discrete system f(n) satisfies the N-order linear constant coefficient difference equation and has a unique solution. When N sequential output values ​​are given, there are generally two cases: forward difference and backward difference. Assume that the forward difference is represented by Δ and the backward difference is represented by Represented. The first-order forward difference equation and the first-order backward difference equation are:

[0064] Δf(n)=f(n+1)-f(n);

[0065]

[0066] The second-order forward difference equation and the second-order backward difference equation are:

[0067] Δ 2 f(n)=Δf(n+1)-Δf(n)=f(n+2)-2f(n+1)+f(n);

[0068]

[0069] For discrete equations, difference is an effective method to achieve differentiation. The first-order difference can be approximated as the first-order derivative, and the second-order difference can be approximated as the second-order derivative.

[0070] For this embodiment, the EEG signal, namely the slow cortical potential MRCP, is preprocessed to extract the A and B features of the signal. Let the slow cortical potential MRCP be x[j], the signal length be N, and the sampling frequency be f S , then the calculation expressions of A and B features are as follows:

[0071]

[0072]

[0073] Among them, A is the first-order backward difference of the signal, B is the second-order backward difference of the signal, Δ represents the forward difference, It represents backward difference, N is the signal length after windowing, and the preprocessed EEG signal is divided into windows, and then the A and B features of each window are extracted respectively.

[0074] Step 4: Use the differential features in the resting and moving states as input features, and the resting label and the moving label as corresponding output labels, input them into the SVM model (support vector machine) for training, and obtain the motion intention decoding model.

[0075] Step 5: Obtain the target person's slow cortical potential MRCP to be decoded in real time, perform preprocessing and feature extraction according to the methods of steps 2 to 3, obtain the differential features of the slow cortical potential MRCP to be decoded, and input them into the motion intention decoding model to obtain the target person's motion intention at the current moment.

[0076] In this embodiment, the EEG sensor uses a NeuSen W series 64-channel wireless acquisition circuit, and the EEG signal is synchronized through a multi-parameter synchronization trigger box of Neuracle and collected at a sampling frequency of 1000 Hz.

[0077] Another embodiment of the present invention further provides a motor intention decoding system based on slow cortical potentials, corresponding to the above method, comprising:

[0078] The paradigm training module is used to conduct paradigm training on the target personnel and extract the slow cortical potential MRCP in both resting and moving states;

[0079] A preprocessing module, used for preprocessing the slow cortical potential MRCP, including time domain filtering and frequency domain filtering;

[0080] A feature extraction module is used to extract features of the pre-processed slow cortical potential MRCP in the resting and moving states by differential means, so as to obtain differential features in the resting and moving states;

[0081] The model training module is used to input the differential features in the resting and moving states and the corresponding resting labels and moving labels into the SVM model for training to obtain the movement intention decoding model;

[0082] The intention decoding module is used to obtain the target person's slow cortical potential MRCP to be decoded in real time, perform preprocessing and feature extraction, obtain the differential features of the slow cortical potential MRCP to be decoded, and input them into the movement intention decoding model to obtain the target person's movement intention at the current moment.

[0083] As for the system device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0084] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0085] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for decoding movement intention based on slow cortical potentials, characterized in that: The following steps are involved: Step 1: Conduct paradigm training on the target person to extract the slow cortical potential MRCP in both resting and moving states; Step 2, preprocessing the slow cortical potential MRCP, including time domain filtering and frequency domain filtering; Step 3, extracting features of the pre-processed slow cortical potential MRCP in the resting and moving states by differential method, and obtaining differential features in the resting and moving states; Step 4: Input the differential features in the resting and moving states and the corresponding resting labels and moving labels into the SVM model for training to obtain a motion intention decoding model; Step 5: Obtain the target person's slow cortical potential MRCP to be decoded in real time, perform preprocessing and feature extraction according to the methods of steps 2 to 3, obtain the differential features of the slow cortical potential MRCP to be decoded, and input them into the motion intention decoding model to obtain the target person's motion intention at the current moment.

2. The method for decoding movement intention based on slow cortical potentials according to claim 1, characterized in that: In step 1, the process of conducting paradigm training on the target personnel is as follows: Step 1.

1. Preparation: A touch pad was set at the initial position and the target position. The target person sat on a chair and placed his hand on the contact area of ​​the touch pad as the starting position. The experimental paradigm was trial-based. The prompt information was displayed on the computer screen in front of the subject. At the beginning of the experiment, the target person first received a prompt to inform the number of remaining trials. At the beginning of each trial, the target person was asked to look at the center of the screen. Step 1.2: After the experiment begins, the target person receives a resting reminder, and the subject completes the resting state without making any movements, while the single-chip microcomputer system is used to collect data; Step 1.3: Prompt slow movement, and the target person completes the finger sliding action as required, that is, completes the straight line action from the initial position to the target position; In the above steps, each time the hand is placed in the initial position, a new trial prompt starts. When the hand leaves the initial position, the microcontroller detects the switch signal and sends it to the host computer. The host computer sends the signal to the EEG acquisition supporting software. The software opens a trigger function to synchronize external data with the EEG signal. Once the hand touches the target touch piece, the microcontroller detects the time point of the finger and leaving, that is, the switch signal, and sends it to the host computer to eliminate the erroneous trials in the time interval between the initial position and the target position.

3. The method for decoding movement intention based on slow cortical potentials according to claim 1, characterized in that: In step 2, the method for preprocessing the slow cortical potential MRCP is: Step 2.1, time domain filtering, using Laplace spatial filtering to remove local background activity; Step 2.2: Frequency domain filtering, using a 4th-order Butterworth causal filter with a frequency band filtering range of 0.1 to 40 Hz.

4. The method for decoding movement intention based on slow cortical potentials according to claim 3, characterized in that: In step 2.1, for electrode channel i, the calculation process of the Laplace spatial filter is as follows: in, is the filtered signal, x i is the signal of channel i, x j is the signal of channel j, S i is the set of channels adjacent to channel i, g ij is the weight matrix. The simplified version of the Laplacian spatial filter considers adjacent electrodes to be equidistant, i.e., g ij =1 / N i , where N i is the number of neighboring electrodes of electrode i.

5. The method for decoding movement intention based on slow cortical potentials according to claim 1, characterized in that: In step 3, the process of feature extraction by difference is as follows: Let the slow cortical potential MRCP be x[j], the signal length be N, and the sampling frequency be f S , then the calculation expressions of A and B features are as follows: Among them, A is the first-order backward difference of the signal, B is the second-order backward difference of the signal, Δ represents the forward difference, It represents backward difference, N is the signal length after windowing, and the preprocessed EEG signal is divided into windows, and then the A and B features of each window are extracted respectively.

6. A motor intention decoding system based on slow cortical potentials, characterized in that: include: The paradigm training module is used to conduct paradigm training on the target personnel and extract the slow cortical potential MRCP in both resting and moving states; A preprocessing module, used for preprocessing the slow cortical potential MRCP, including time domain filtering and frequency domain filtering; A feature extraction module is used to extract features of the pre-processed slow cortical potential MRCP in the resting and moving states by differential means, so as to obtain differential features in the resting and moving states; The model training module is used to input the differential features in the resting and moving states and the corresponding resting labels and moving labels into the SVM model for training to obtain the movement intention decoding model; The intention decoding module is used to obtain the target person's slow cortical potential MRCP to be decoded in real time, perform preprocessing and feature extraction, obtain the differential features of the slow cortical potential MRCP to be decoded, and input them into the movement intention decoding model to obtain the movement intention of the target person at the current moment.

Images