A motion imagination task time recognition system

Abstract

A system for identifying the time of a motor imagery task includes an EEG signal acquisition module, a signal preprocessing module, and an imagery task time identification module. The imagery task time identification module includes brain network construction, global efficiency calculation of the brain network, and the imagery task time identification process. Brain network construction uses an adaptive directional transfer function method to calculate the information flow weight between the α and β frequency bands (8-30 Hz EEG acquisition channels), and represents the brain network graph using a directed adjacency matrix. Global efficiency calculation uses the average of the inverses of the shortest paths between nodes to quantify the global information transmission capacity of the brain network. Imagery task time identification, based on the changing pattern of the global efficiency curve during motor imagery, identifies the start time of ERD, the start time of motor imagery, the end time of motor imagery, and the end time of ERS on both sides of the peak of the global efficiency curve within the time period of the paradigm stimulus, thus identifying the time interval of the strongest motor imagery within a given stimulus time. This invention achieves time identification of motor imagery tasks.

Description

Technical Field

[0001] This invention relates to the fields of neural engineering and brain-computer interface technology in biomedical engineering, and in particular to a time recognition system for motor imagery tasks. Background Technology

[0002] Brain-computer interfaces (BCIs) serve as a direct connection between the human brain and external devices, enabling individuals to control these devices through conscious thought. Motor imagery-based BCI technology can recognize spontaneously generated motor intentions within the brain and thus control external devices, and has already found widespread application in prosthetic limb control and brain-controlled rehabilitation. However, as an endogenous BCI paradigm, motor imagery suffers from several drawbacks. First, each motor imagery task involves a delay and the possibility of premature termination. Second, the activation time between trials is highly variable, significantly impacting the decoding performance of motor imagery-based BCIs and failing to meet the needs of rehabilitation devices requiring sequential control.

[0003] Current research on brain-computer interfaces based on motor imagery largely ignores the delay time and activation duration of motor imagery. During signal decoding, averaged features are extracted from the entire imagery time interval, resulting in weak feature discrimination. Alternatively, they rely on experience to select time intervals with better classification performance, lacking relevant theoretical guidance, and the models' cross-trial generalization ability is also relatively weak. In rehabilitation training devices, some researchers calculate the user's attention value using specific frequency band energy from certain electrodes, and use the attention threshold to determine the degree of motor imagery activation for device control. However, this method firstly utilizes only a limited amount of electrode information, making it susceptible to noise interference, and secondly, it does not utilize the brain activity characteristics of the motor cortex, and is not a direct method for determining the activation time of motor imagery.

[0004] Overall, although some scholars have observed a certain delay in motor imagery trials, there is very little research on time identification for motor imagery tasks both domestically and internationally, and no literature on time identification for motor imagery tasks has been published. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the present invention aims to provide a time identification system for motor imagery tasks. By analyzing the time-varying brain network during motor imagery, the system identifies the start time of ERD, the start time of motor imagery, the end time of motor imagery, the end time of ERS, and the maximum activation intensity of motor imagery.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A time recognition system for motor imagery tasks includes an electroencephalogram (EEG) signal acquisition module, a signal preprocessing module, and an imagery task time recognition module, wherein...

[0008] EEG signal acquisition module: In accordance with the international 10-20 system, measuring electrodes are arranged at the electrode positions numbered C3, C4, CP3, CP4, FC3, FC4, and Cz in the user's head motor area, a reference electrode is arranged at AFz, and a ground electrode is arranged at Fpz. EEG signals are acquired using EEG acquisition equipment.

[0009] Signal preprocessing module: includes filtering and delinearization steps. Butterworth filtering is used for 8-30Hz bandpass filtering to remove irrelevant frequency band EEG signals. The least squares fitted line is then subtracted from the EEG signal to remove linear trends.

[0010] The imagination task time identification module includes brain network construction, global efficiency calculation of brain network, and the imagination task time identification process.

[0011] Brain network construction: The adaptive directional transfer function method was used to calculate the information flow weight between the α and β frequency bands, i.e., the 8-30 Hz EEG acquisition channels, and the brain network graph was represented by a directed adjacency matrix.

[0012] Global efficiency calculation of brain networks: the average of the inverses of the shortest paths between nodes, quantifying the global information transmission capacity of brain networks;

[0013] The task time identification method identifies the starting, starting, ending, and ending times of motor imagination based on the changing pattern of the global efficiency curve during the process of motor imagination. Within the time period of the paradigm stimulus, the time interval of the peak of the global efficiency curve is found to be the start of ERD, the start of motor imagination, the end of motor imagination, and the end of ERS. This allows the identification of the time interval of the most intense motor imagination within a given stimulus time.

[0014] The specific method of the imagined task time identification module includes the following steps:

[0015] 1) Brain network construction and global efficiency calculation of brain networks:

[0016] Introducing preprocessed EEG signals X(t), an MVAAR model is constructed.

[0017]

[0018] In the formula, X(t) is the multichannel EEG signal at time t, p is the MAVVR model order, A(i,t) is the model parameter to be calculated, and E(t) represents white noise;

[0019] The parameters A(i,t) of the MAVVR model are estimated and calculated using the Kalman filter algorithm.

[0020] The information flow between nodes in the brain network in the 8-30 Hz frequency band is then calculated using the adaptive directional transfer function method.

[0021]

[0022] H(f,t)=A -1 (f,t)

[0023] In the formula, A(f,t) represents the frequency domain transform of A(i,t), and H(f,t) is the adaptive directional transfer function;

[0024] Normalization is performed.

[0025]

[0026] In the formula, This represents the normalized directional transfer function, where n represents the number of nodes;

[0027] The directional transfer function is averaged within the range of 8-30 Hz to obtain the information flow from node j to node i at time t within the studied frequency band (f1, f2).

[0028]

[0029] A time-varying brain network graph is constructed accordingly, represented by a directed adjacency matrix. The global efficiency E of the brain network is then calculated using the global efficiency formula. glob curve;

[0030]

[0031] 2) Identifying the activation time of the motor imagery task: Introducing the global efficiency curve of the brain network calculated in step 1), smoothing is first performed by averaging the global efficiency curve using a time window of length 0.1s and a step size of one sampling interval. Then, the activation time of the motor imagery task is identified:

[0032] Find the minimum value E1 to the left of the peak of the global efficiency curve as the starting point of ERD, and find the minimum value E2 to the right of the peak as the ending point of ERS. p Quantify the activation intensity of motion imagery;

[0033] After the ERD initiation point, using the global efficiency at the start of ERD as a baseline, when the increase in the global efficiency of the brain network first reaches 0.5 times the peak increase in global efficiency, that is, after the ERD initiation point, E... glob The first time it exceeded E1 + 0.5 × (E p-E1) is recorded as the start time of motion visualization; before the end point of ERS, based on the peak global efficiency, when the global efficiency decreases to 0.5 times the decrease at the end of ERS, that is, before the end of ERS, E glob Decreased to E2 + 0.5 × (E p -E2) is recorded as the end time of the motion visualization.

[0034] During the process of motor imagery, the brain network exhibits an Ω-shaped global efficiency curve, starting from the user's ERD, proceeding to the execution of motor imagery, and ending at ERS.

[0035] In multiple imagination tasks for a single user, the time interval with the strongest motor imagination activation in each trial was identified; the experiment was repeated multiple times to statistically analyze the temporal pattern of motor imagination activation, as well as the average intensity and average duration of activation.

[0036] The beneficial effects of this invention are as follows: Since this invention adopts the global efficiency method of brain network, it uses EEG data from at least 8 electrodes to analyze brain activity in the motor area and utilizes the characteristics of changes in brain network attributes during motor imagination. It can accurately identify the activation time of motor imagination task, thus having the advantages of accurate identification, not being easily affected by the activity of other brain areas, and strong anti-interference. Attached Figure Description

[0037] Figure 1 This is a block diagram of the system according to an embodiment of the present invention.

[0038] Figure 2 This is a diagram of a motion imagery-induced experiment according to an embodiment of the present invention.

[0039] Figure 3 This is a flowchart of brain network analysis and global efficiency calculation according to an embodiment of the present invention.

[0040] Figure 4 This is a flowchart of the motion imagination task activation time identification in an embodiment of the present invention.

[0041] Figure 5 This is the activation time identification result of a single motion imagination task in an embodiment of the present invention. Detailed Implementation

[0042] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] Reference Figure 1 A system for time recognition of motor imagery tasks includes an electroencephalogram (EEG) signal acquisition module, a signal preprocessing module, and an imagery task time recognition module.

[0044] EEG signal acquisition module: In accordance with the international 10-20 system, measuring electrodes are arranged at the electrode positions numbered C3, C4, CP3, CP4, FC3, FC4, and Cz in the user's head motor area, a reference electrode is arranged at AFz, and a ground electrode is arranged at Fpz. EEG signals are acquired using EEG acquisition equipment.

[0045] Signal preprocessing module: includes filtering and delinearization steps. Butterworth filtering is used for 8-30Hz bandpass filtering to remove irrelevant frequency band EEG signals. The least squares fitted line is then subtracted from the EEG signal to remove linear trends.

[0046] The visualization task time identification module includes brain network construction, global efficiency calculation of brain network, and visualization task time identification process.

[0047] Reference Figure 1 A method utilizing a motion imagery task time recognition system includes the following steps:

[0048] Step 1: The EEG signal acquisition module acquires EEG signals: According to the international 10-20 system, measurement electrodes are arranged at the electrode positions numbered C3, C4, CP3, CP4, FC3, FC4, and Cz in the user's head motor area.

[0049] Reference Figure 2 Open the motion visualization task prompt demonstration. The left hand and right hand motion visualization task prompts will appear randomly on the screen. The user performs the corresponding motion visualization according to the task prompts.

[0050] Start the EEG acquisition device. Start the acquisition 1 second before the task prompt begins, continue the task prompt for 4 seconds, and end the acquisition when the task prompt ends. The sampling rate is generally not less than 250 Hz.

[0051] Step 2: The signal preprocessing module preprocesses the EEG signal, including filtering and delinearization. It constructs a Butterworth filter for 8-30Hz bandpass filtering to remove irrelevant frequency band EEG signals, and then subtracts the least squares fitted line from the EEG signal to remove linear trends.

[0052] Step 3, Brain Network Construction and Global Efficiency Calculation:

[0053] Reference Figure 3 Preprocessed EEG signals X(t) are introduced to construct an MVAAR model.

[0054]

[0055] In the formula, X(t) is the multichannel EEG signal at time t, p is the MAVVR model order, A(i,t) is the model parameter to be calculated, and E(t) represents white noise;

[0056] The parameters A(i,t) of the MAVVR model are estimated and calculated using the Kalman filter algorithm.

[0057] The information flow between nodes in the brain network in the 8-30 Hz frequency band is then calculated using the adaptive directional transfer function method.

[0058]

[0059] H(f,t)=A -1 (f,t)

[0060] In the formula, A(f,t) represents the frequency domain transform of A(i,t), and H(f,t) is the adaptive directional transfer function;

[0061] To ensure that the total amount of information flowing into a node from other nodes equals 1, normalization is required.

[0062]

[0063] In the formula, This represents the normalized directional transfer function, where n represents the number of nodes;

[0064] If only a specific frequency band is studied, this embodiment only studies 8-30 Hz. Therefore, the directional transfer function is averaged within 8-30 Hz to obtain the information flow from node j to node i at time t within the studied frequency band (f1, f2).

[0065]

[0066] A time-varying brain network graph is constructed accordingly, represented by a directed adjacency matrix. The global efficiency E of the brain network is then calculated using the global efficiency formula. glob curve;

[0067]

[0068] Step 4, identify the activation time of the motor imagery task: refer to Figure 4 Introducing the global efficiency curve of the brain network calculated in step 3, we first perform smoothing by averaging the global efficiency curve using a time window of length 0.1s and a step size of one sampling interval. Then, we identify the activation time of the motor imagery task.

[0069] Find the minimum value E1 to the left of the peak of the global efficiency curve as the starting point of ERD, and find the minimum value E2 to the right of the peak as the ending point of ERS. p Quantify the activation intensity of motion imagery;

[0070] After the ERD initiation point, using the global efficiency at the start of ERD as a baseline, when the increase in the global efficiency of the brain network first reaches 0.5 times the peak increase in global efficiency, that is, after the ERD initiation point, E... glob The first time it exceeded E1 + 0.5 × (E p -E1) is recorded as the start time of motion visualization; before the end point of ERS, based on the peak global efficiency, when the global efficiency decreases to 0.5 times the decrease at the end of ERS, that is, before the end of ERS, E glob Decreased to E2 + 0.5 × (E p -E2) is recorded as the end point of the motion visualization;

[0071] Reference Figure 5 , Figure 5 This is the result of the division of a single trial. At this point, the activation time interval of a single motor imagery experiment can be identified. By repeating steps 2-4, the activation time sequence of the user's motor imagery can be statistically analyzed, as well as the average intensity and average duration of activation.

[0072] To verify the feasibility of the system of this invention, data were collected from 9 subjects in this embodiment. Each subject underwent 3 rounds of experiments, with 40 motor imagery tasks performed in each round. At the start of each task, motor imagery prompts for the left and right hands were randomly displayed on the screen. The total duration of each experiment was 6 seconds, with a resting time of 2 seconds and a motor imagery time of 4 seconds. The system of this invention was used to identify the task time of all subjects' motor imagery trials. Table 1 shows the statistical results of the activation time identification of the 9 subjects through 120 motor imagery tasks. It can be found that the time dispersion (sample variance) of motor imagery task activation is large among different trials of the same subject, while the dispersion of the maximum global efficiency of the brain network is small. Across subjects, the activation delay time of motor imagery is widespread, and the activation patterns have certain similarities.

[0073] Table 1

[0074]

[0075]

[0076] To verify the performance improvement of the present invention system over traditional decoding algorithms, the original EEG data of the EEG signals collected in the above experiment were truncated from the average MI start time to the average MI end time, and then cosmic pattern feature extraction was performed. Combined with support vector machine classification, the results were compared with the traditional method without time truncation. Table 2 shows the comparison results.

[0077] Table 2

[0078]

[0079]

[0080] In the verification experiment, considering the co-spatial patterns of motor imagery activation time intervals yielded better classification results. These results demonstrate that the system of this invention can identify task activation time by analyzing the activity state of the brain network during a user's motor imagery process. It can assess the activation time and intensity patterns of the user's motor imagery and provide a reference for further processing of motor imagery EEG signals.

Claims

1. A motor imagery task time discrimination system, characterized by: The method comprises a brain electrical signal acquisition module, a signal preprocessing module, and an imagined task time identification module. The brain electrical signal acquisition module comprises a user's head movement area C3, C4, CP3, CP4, FC3, FC4, and Cz numbered electrode position arrangement measurement electrode, AFz arrangement reference electrode, Fpz arrangement ground electrode, and a brain electrical signal acquisition device. The signal preprocessing module comprises a filter, a de-linear trend link, a Butterworth filter for 8-30hz band-pass filtering, and a least square fitting straight line for removing linear trend. The imagined task time identification module comprises brain network construction, brain network global efficiency calculation, and imagined task time identification process. The brain network construction uses an adaptive directional transfer function method to calculate the information flow weight between the alpha and beta frequency bands, i.e., 8-30hz brain electrical signal acquisition channels, and uses a directed adjacency matrix to represent the brain network graph. The brain network global efficiency calculation is the average value of the reciprocal of the shortest path between nodes, which quantifies the global information transmission capacity of the brain network. The imagined task time identification is based on the global efficiency curve change rule in the motor imagination process, and the ERD start, motor imagination start, motor imagination end, and ERS end time points are found on both sides of the global efficiency curve peak value in the time period of the paradigm stimulation, so as to identify the strongest time interval of motor imagination in the given stimulation time. The motor imagination task activation time is identified by introducing the calculated brain network global efficiency curve, smoothing the curve, using a time window with a length of 0.1s and a step of one sampling interval to average the global efficiency curve, and then identifying the motor imagination task activation time. Finding the minimum on the left side of the peak of the global efficiency curve Finding the minimum on the right side of the peak of the corresponding time as the starting point of ERD Finding the minimum on the right side of the peak as the end point of ERS Quantifying the activation strength of motor imagery; After the ERD initiation point, using the global efficiency at the start of ERD as a baseline, the first time the increase in the global efficiency of the brain network reaches 0.5 times the peak increase in global efficiency, i.e., after the ERD initiation point, First time exceeding The time is recorded as the start of the motion visualization; before the end of the ERS, based on the peak global efficiency, when the global efficiency decreases to 0.5 times the decrease at the end of the ERS, that is, before the end of the ERS, Descending to The time is recorded as the end point of the motion visualization.

2. The motor imagery task time discrimination system of claim 1, wherein: The specific method of the imagined task time identification module comprises the following steps: 1) Brain network construction and brain network global efficiency calculation: Introducing pre-processed electroencephalography signals , constructing a MAVVR model, In the formula, is the multi-channel electroencephalogram signal at time t, p is the order of the MAVVR model, is the model parameter to be calculated, represents white noise; Estimating the parameters of the MAVVR model using Kalman filter algorithm calculations; The adaptive directional transfer function method is used to calculate the information flow between the nodes of the brain network in the 8-30hz frequency band. wherein represents a frequency domain transform of is an adaptive directivity transfer function; The normalized processing is performed, wherein denotes the normalized directional transfer function, n denotes the number of nodes; The directional transfer function is averaged over 8-30 Hz to obtain the information flow flux from node j to node i at time t in the studied frequency band (fmin, fmax) ;​ The time-varying brain network graph is constructed, the constructed brain network graph is represented by a directed adjacency matrix, and the global efficiency of the brain network is calculated according to a calculation formula of global efficiency Curve; 。 3. The system of claim 1, wherein: The brain network global efficiency curve presents an Ω shape from the user's ERD start, motor imagination execution, to the ERS end.

4. The system of claim 1, wherein: In multiple imagined tasks of a single user, the time interval with the strongest motor imagination activation degree in each trial is identified. Multiple experiments are repeated, and the user's motor imagination activation timing rule, average strength, and average duration are statistically obtained.

Images