Walking target-oriented lower limb motor imagery brain-computer interface signal acquisition system and interface method

Abstract

This invention relates to a brain-computer interface (BCI) signal acquisition system and method based on walking goal-oriented lower limb motor imagery, aiming to address the low signal quality issue of existing lower limb motor imagery (MI) signals. The system includes an EEG signal acquisition device and a spatial environment for lower limb motor imagery. The spatial environment includes ground start and end lines, footprint markers spaced between the start and end lines, and a display device. The display device shows the subject the accuracy rate of stimulated lower limb motor imagery, which is obtained by combining the decoding result of the decoder based on the subject's EEG signals with the stimulated accuracy rate. The interface method first requires the subject to walk along the ground footprint markers, observe the footprint markers to recall the walking process, and use the walking recall as a template for imagery practice. During actual motor imagery, the subject observes the display device to perform lower limb motor imagery and acquires EEG signals.

Description

Technical Field

[0001] This invention belongs to the field of brain-computer interfaces and relates to a signal acquisition system and interface method for a brain-computer interface for lower limb motor imagination. Background Technology

[0002] Brain-computer interface (BCI) is an application technology that enables direct communication between the human brain and external devices. It detects and decodes neural activity in the brain and transmits it to external devices, thereby enabling external control and interaction. BCI systems utilize electrodes to detect electroencephalogram (EEG) signals in the human brain, analyze user intentions through internal algorithms, and convert these signals into computer-readable instructions. It can be used in industrial fields for thought-controlled machine or program management and efficiency optimization, and in the medical field to assist patients with assistive device control and neurorehabilitation training. Motor imagery (MI) is a BCI-related neural signal pattern, referring to the brain activation generated when people imagine a specific movement. The neural activity induced by MI is very similar to the electrophysiological activity during actual movement, allowing these EEG signals to be interpreted by computers. BCI-MI, as a non-invasive BCI technology highly correlated with user intentions, has great application potential in various fields due to its safety, non-invasiveness, low site dependence, lack of actual movement requirement, and diverse control methods. At the same time, BCI-MI can also provide rehabilitation training for patients with neuromuscular diseases and promote their neuroplasticity.

[0003] The quality of migraine (MI) signals significantly impacts their control and rehabilitation performance. However, MI signals are characterized by low signal-to-noise ratios and significant individual variability, and are highly susceptible to interference from external electromagnetic and physiological signals, as well as internal psychological states, posing challenges to ensuring signal quality. Furthermore, lower limb MI responses are more difficult to detect compared to upper limb MI. Because the corresponding brain regions for the lower limbs are located deep in the sulci, their characteristic responses are weaker, and their classification performance is poorer, limiting their application effectiveness. On the other hand, lower limb MI is more abstract than upper limb MI because, compared to the hands, which are usually within visual range and require fine manipulation, humans rarely focus on leg movements and sensations, making it more difficult to imagine lower limb movements. Some participants with lower limb MI even find the signal enhancement methods commonly used for upper limb MI (such as motion videos and electrical stimulation) confusing, believing that such unnatural stimulation hinders their normal imagination. Most participants struggle to master lower limb MI within a short period and are unable to generate stable EEG signals. These factors all contribute to low signal quality in lower limb MI. These low-quality MI signals cause BCI-MI decoding to be highly dependent on complex decoding algorithms, resulting in high overall system latency, poor real-time performance, and low decoding accuracy, which cannot meet the application requirements of complex equipment or high-precision instruments.

[0004] Improving the MI task paradigm is a key factor in enhancing MI signal quality, even surpassing improvements in feature extraction algorithm performance. Among participants lacking sufficient MI skills, even with strong algorithms, it's difficult to accurately identify a large number of failed MI task signals, resulting in a poorly generalized classifier. Traditional MI experimental paradigms typically use text or simple icons as task cues. However, these abstract cues often lack intuitiveness, requiring participants to immediately associate the cues with imagined actions during the experiment, thus increasing the difficulty of the imaginative process. Summary of the Invention

[0005] This invention aims to address the problem of low signal quality in lower limb MI signals obtained by existing methods.

[0006] A brain-computer interface signal acquisition system for lower limb motor imagination based on walking goal orientation includes at least an electroencephalogram (EEG) signal acquisition device and a spatial environment for lower limb motor imagination.

[0007] The spatial environment includes ground start and end lines and footprint markers spaced between the start and end lines, as well as a display device; the display device displays the subject's accuracy of stimulated lower limb motor imagery during the motor imagery task.

[0008] Furthermore, the display device shows the subject a walking animation corresponding to the gait cycle during the motor imagery task, and makes muscle markings on the tibialis anterior and gastrocnemius muscles of the lower leg according to the content corresponding to the gait cycle to prompt the subject.

[0009] Further, the walking animation provides animation cues in a non-immersive third-person perspective.

[0010] Further, the process of presenting the accuracy rate of lower limb motor imagery with incentives includes:

[0011] (1) At t = t1, set the incentive accuracy rate value to 0, start receiving the decoding results obtained by the decoder, and recalculate the display threshold according to the actual threshold :

[0012]

[0013] where the threshold is the classification accuracy rate required for a trigger signal in a single motor imagery experiment; is the incentive value;

[0014] The decoder is used to decode the electroencephalogram (EEG) signals of the subject during the lower limb motor imagery task to obtain the decoding results, that is, the classification results of the EEG signals;

[0015] (2) When t1 < t < t3, the incentive accuracy rate changes according to the time step, and simultaneously loop to receive the current classification accuracy rate , and calculate the display accuracy rate :

[0016]

[0017]

[0018] where represents rounding up upward, ts is the time step for the change of the incentive accuracy rate; t1 is the start time of the lower limb motor imagery, and t3 is the end time of an EEG signal starting from the lower limb motor imagery;

[0019] (3) At t = t3, fix the incentive accuracy rate to , continue to loop to receive the current classification accuracy rate, and calculate the display accuracy rate according to the above formula.

[0020] Preferably, the incentive value is set to 0.5.

[0021] Further, the decoder is obtained through the following steps:

[0022] S201. Calibrate the decoder:

[0023] The specific calibration paradigm is divided into a motor imagery task and a resting task. The number of trials for both tasks is the same. The motor imagery task is completed first, followed by the resting task.

[0024] The motor imagery task consists of: a controlled trial begins, lasting for a period of time for baseline correction; then a motor imagery prompt is given for a duration of t1 to remind the subject to prepare for the mental task; then lower limb motor imagery begins, with the entire task lasting for t5; then a rest period is given to allow the subject to relieve fatigue from the high-intensity mental activity; then the paradigm preparation prompt reappears, and the cycle continues to the next trial.

[0025] The resting task consists of the following steps: After all the motor imagery tasks are completed, participants are required to rest until all resting trials are finished; participants are asked to remain at rest during this period.

[0026] The collected EEG signals are saved and preprocessed to obtain EEG signals for classification, and a classifier is trained. The specific process includes:

[0027] First, remove the data corresponding to the useless channels of the EEG signal acquisition device, and then use a bandpass filter.

[0028] Then, a sliding window with a length of tm and a sliding step of tn is used to segment the EEG data of length tL collected in each trial. For each segment of data, the data of the motor imagery task is marked as "1" and the resting state is marked as "0". This is done for m trials to obtain multiple segments of data, which are used to train the decoder.

[0029] The decoder classifies the EEG signals and establishes a binary classification model for lower limb motor imagery and resting state; it adopts ten-fold cross-validation and takes the weights of the last trained model as the final model weights, which are then used to control the visual feedback lower limb walking online system.

[0030] S202. Evaluate the false positive rate of the trained model online:

[0031] Subjects performed a resting task, and their resting-state EEG signals were collected. The EEG data was segmented using a sliding window and then fed into a decoder. The classification accuracy at time t was measured. Calculated using the following formula:

[0032]

[0033] in, For smoothing coefficients, The accuracy from the previous time step is updated to 0 at the start of each new trial. This is the current classification result;

[0034] Examine the false positive rate of the model and determine whether to retrain the model or recalibrate the subjects by providing visualization guidance.

[0035] Furthermore, the decoder employs the minimum Riemann average distance algorithm.

[0036] The brain-computer interface method for lower limb motor imagery based on walking goal orientation, which is based on the aforementioned brain-computer interface signal acquisition system for lower limb motor imagery based on walking goal orientation, specifically includes: the steps of subject paradigm imagery, and the steps of actual motor imagery.

[0037] The steps of the subject paradigm imagination include:

[0038] Before collecting EEG signals, subjects were asked to walk along the footprints on the ground, perform foot movements and perceive leg activity. After each walk, they returned to their original position, observed the footprints on the ground, and recalled the walking process, using the walking recall as a template for paradigmatic imagination.

[0039] Subjects should perform lower limb motor imagery exercises at the starting point. During the lower limb motor imagery process, subjects are required to observe footprint marks and recall the corresponding walking movements, images, and sensations. Electroencephalogram (EEG) signals are collected during lower limb motor imagery and at rest.

[0040] The steps of the actual motion visualization include:

[0041] Subjects should observe the display device at the starting point and perform lower limb motor imagery. During the lower limb motor imagery process, subjects are required to recall the images and sensations of the corresponding walking movements. Electroencephalogram (EEG) signals are collected during the actual motor imagery process and at rest. The display device shows the subjects the accuracy of the stimulated lower limb motor imagery.

[0042] Furthermore, in the actual motor imagery step, the subject should observe the display device at the starting point to perform lower limb motor imagery. During this process, the display device is also used to play walking animations.

[0043] Furthermore, in the process of using walking memory as a template for paradigmatic imagination, the template should include the entire gait cycle of walking. A walking gait cycle includes: (1) the entire foot strikes the ground, (2) the heel lifts off the ground, (3) the swing phase, and (4) the heel strikes the ground.

[0044] The advantages and positive effects of this invention are:

[0045] 1. This invention proposes a more intuitive goal-oriented lower limb walking paradigm prompting method, using footprint-shaped markers to guide lower limb motor imagery. Compared to third-person video visual guidance for lower limb motor imagery, this method is less difficult for users to execute and is simpler than immersive visual guidance. Compared to the two conventional visual guidance methods mentioned above, goal-oriented visual prompts require less cognitive ability and concentration from users, reducing the learning and training costs of the lower limb motor imagery paradigm. This allows users to generate clearer MI signals more quickly, compensating to some extent for the low classification accuracy of lower limb motor imagery EEG signals.

[0046] 2. In this invention, a smoothing algorithm is used to convert the single EEG classification result into a classification accuracy with a low false positive rate. The current EEG decoding accuracy is then visually fed back in real time, providing positive progress incentives. Compared to the lower limb MI paradigm without feedback, the intuitive progress bar visual feedback can reduce user anxiety and confusion. Positive progress incentives can boost confidence for users with poor motor imagery abilities. These positive psychological states can, to some extent, compensate for the decline in MI signal quality and stability caused by negative states such as tension and fatigue.

[0047] 3. This invention can be applied to brain-computer interface fields such as engineering mind control and medical rehabilitation, increasing the feasibility of MI-BCI system in controlling external devices related to the lower limbs and walking, and providing strong support for mind-controlled robots and training brain abilities and plasticity. Attached Figure Description

[0048] Figure 1 This is a real-world photograph of the experimental scenario of this invention;

[0049] Figure 2 This is a diagram showing the location of the footprint-shaped markers;

[0050] Figure 3 This is a single-trial paradigm diagram of the present invention;

[0051] Figure 4 This is a diagram of the subject interface for the visual feedback lower limb motor imagery paradigm of this invention;

[0052] Figure 5 This is a waveform diagram showing the classification results and classification accuracy of a single test of lower limb motor imagination by a test subject in this invention. Detailed Implementation

[0053] Goal orientation is an imagination guidance method different from movement observation. It guides the subject to imagine the process of the limb reaching the target position rather than imagining the fixed actions prompted by the video, which is more natural than the latter. Therefore, the present invention proposes a lower-limb walking movement imagination method based on walking goal orientation to more intuitively guide MI, and designs a corresponding feedback result algorithm and a control interaction interface on this basis to construct an online movement imagination system for lower-limb walking based on walking goal orientation. Specific Embodiment 1:

[0055] This embodiment is a system for acquiring brain-computer interface signals for lower-limb movement imagination based on walking goal orientation, which at least includes an electroencephalogram signal acquisition device and a spatial environment for lower-limb movement imagination;

[0056] The spatial environment includes a ground start-stop line and footprint marks arranged at intervals between the start-stop lines, as well as a display device; the display device shows the accuracy rate of lower-limb movement imagination with incentives to the subject during the movement imagination task. The process of showing the accuracy rate of lower-limb movement imagination with incentives includes:

[0057] (1) At t = t1, let the incentive accuracy rate value be 0, start receiving the decoding result obtained by the decoder, and at the same time recalculate the display threshold according to the actual threshold :

[0058]

[0059] Among them, the threshold is the classification accuracy rate required for the trigger signal of a single movement imagination experiment; is the incentive value, and the incentive value is preferably set to 0.5.

[0060] The decoder is used to decode the electroencephalogram signal of the subject during the lower-limb movement imagination task of the subject to obtain a decoding result, that is, the classification result of the electroencephalogram signal; in this embodiment, the decoder adopts the minimum Riemannian average distance algorithm.

[0061] (2) When t1 < t < t3, the incentive accuracy rate changes according to the time step, and at the same time, the current classification accuracy rate is cyclically received, and the display accuracy rate is calculated:

[0062]

[0063]

[0064] Among them, represents Rounded up, ts represents the time step of the change in stimulus accuracy; t1 represents the start time of lower limb motor imagery; and t3 represents the end time of an EEG signal from the start of lower limb motor imagery. In this embodiment, the EEG signal is divided according to a sliding window with a time window length of tm and a sliding step size of ts. Therefore, in this embodiment, t3 = t1 + tm.

[0065] During the motor imagery task, the display device can selectively show the subject walking animations corresponding to gait cycles. During this process, muscle markers are made on the tibialis anterior and gastrocnemius muscles of the lower leg according to the content corresponding to the gait cycle to prompt the subject. The walking animations are presented in a non-immersive third-person perspective.

[0066] The decoder is obtained through the following steps:

[0067] S201. Calibrate the decoder:

[0068] The specific calibration paradigm is divided into a motor imagery task and a resting task. The number of trials for both tasks is the same. The motor imagery task is completed first, followed by the resting task.

[0069] The motor imagery task consists of: a controlled trial begins, lasting for a period of time for baseline correction; then a motor imagery prompt is given for a duration of t1 to remind the subject to prepare for the mental task; then lower limb motor imagery begins, with the entire task lasting for t5; then a rest period is given to allow the subject to relieve fatigue from the high-intensity mental activity; then the paradigm preparation prompt reappears, and the cycle continues to the next trial.

[0070] The resting task consists of the following steps: After all the motor imagery tasks are completed, participants are required to rest until all resting trials are finished; participants are asked to remain at rest during this period.

[0071] The collected EEG signals are saved and preprocessed to obtain EEG signals for classification, and a classifier is trained. The specific process includes:

[0072] First, remove the data corresponding to the useless channels of the EEG signal acquisition device, and then use a bandpass filter.

[0073] Then, a sliding window with a length of tm and a sliding step of tn is used to segment the EEG data of length tL collected in each trial. For each segment of data, the data of the motor imagery task is marked as "1" and the resting state is marked as "0". This is done for m trials to obtain multiple segments of data, which are used to train the decoder.

[0074] The decoder classifies the EEG signals and establishes a binary classification model for lower limb motor imagery and resting state; it adopts ten-fold cross-validation and takes the weights of the last trained model as the final model weights, which are then used to control the visual feedback lower limb walking online system.

[0075] S202. Evaluate the false positive rate of the trained model online:

[0076] Subjects performed a resting task, and their resting-state EEG signals were collected. The EEG data was segmented using a sliding window and then fed into a decoder. The classification accuracy at time t was measured. Calculated using the following formula:

[0077]

[0078] in, For smoothing coefficients, The accuracy from the previous time step is updated to 0 at the start of each new trial. This is the current classification result;

[0079] Examine the false positive rate of the model and determine whether to retrain the model or recalibrate the subjects by providing visualization guidance. Specific Implementation Method Two:

[0081] This embodiment is a brain-computer interface method for lower limb motor imagery based on walking goal guidance. It is based on the brain-computer interface signal acquisition system for lower limb motor imagery based on walking goal guidance described in Specific Embodiment 1. Specifically, it includes: the step of subject paradigm imagery and the step of actual motor imagery.

[0082] The steps of the subject paradigm imagination include:

[0083] Before collecting EEG signals, the subjects were asked to walk along the footprint marks on the ground, perform foot movements and perceive leg activities. After each walk, they returned to the original position, observed the footprint marks on the ground and recalled the walking process, using the walking recall as a template for paradigm imagination. In the process of using the walking recall as a template for paradigm imagination, the template should include the entire gait cycle of walking. A gait cycle of walking includes: (1) the whole foot touches the ground, (2) the heel leaves the ground, (3) the swing phase, and (4) the heel touches the ground.

[0084] Subjects should perform lower limb motor imagery exercises at the starting point. During the lower limb motor imagery process, subjects are required to observe footprint marks and recall the corresponding walking movements, images, and sensations. Electroencephalogram (EEG) signals are collected during lower limb motor imagery and at rest.

[0085] The steps of the actual motion visualization include:

[0086] Participants should observe the display device at the starting point and perform lower limb motor imagery. During the lower limb motor imagery process, participants are required to recall the images and sensations associated with walking. Electroencephalogram (EEG) signals are collected during the actual lower limb motor imagery process and at rest. The display device shows participants the accuracy rate of stimulated lower limb motor imagery. During this process, the display device also plays walking animations while participants observe the lower limb motor imagery at the starting point.

[0087] Example

[0088] Unless otherwise specified, the hardware and software requirements mentioned in this embodiment are all commercially available. The components of this embodiment can be configured by any suitable facility. The brain-computer interface method for lower limb motor imagery based on walking goal guidance is as follows:

[0089] Step 1: Setting up the experimental environment and instructing subjects:

[0090] The experiment was conducted in a quiet environment, with a ground reference. Figure 1 and Figure 2 Affix labels; inform participants of the experimental procedure and tasks before the experiment; guide participants to follow the instructions. Figure 2 The subject was instructed to perform the movement by walking on either side of the marker and to familiarize themselves with the paradigm. The subject was asked to focus on the process and sensation of leg movement during the walking process and to remember the visual features and kinesthetic sensations during the movement execution process. This was to determine the basic template for subsequent lower limb movement imagination. The basic template included the gait cycle of the entire walking process: (1) the whole foot lands on the ground; (2) the heel leaves the ground; (3) the swing phase; and (4) the heel lands on the ground.

[0091] After the exercise training, participants should stand at the marked starting point and perform lower limb motor imagery exercises. Participants should remain stationary, observe the ground markings, and recall the corresponding images and sensations associated with the movement. Throughout the entire experimental procedure, participants are required to perform only this one type of marker-guided walking lower limb motor imagery.

[0092] After ensuring the subjects understood the basic visualization process and its specific content, the experiment began. A 64-channel EEG acquisition device was used to collect the subjects' EEG signals. The electrodes (silver / silver chloride) were arranged according to the international standard 10-10 system, with a sampling rate of 250Hz. The electrode impedance was kept less than 10kΩ throughout the acquisition process. The EEG acquisition device was used to collect the subjects' EEG signals during imagined walking and resting states.

[0093] Step 2: Calibrate the decoder: The specific calibration paradigm is divided into a motor imagery task and a resting task. The number of trials for both tasks is the same. The motor imagery task is completed first, followed by the resting task.

[0094] The motor imagery task is as follows: The computer screen initially displays a preparatory prompt, "Press the spacebar to begin the trial." The participant presses the button in their hand according to their own situation to initiate the trial. Upon starting the task, a green cross appears on the screen for 3 seconds for baseline correction and to eliminate noise interference. Then, the text prompt "Motor Imagery" appears on the screen for 1 second to remind the participant to prepare for the mental task. Following this, lower limb motor imagery begins, playing a third-person animation of a person walking from the side. Considering that participants focused on motor imagery may not be able to react to the video content and details at a normal speed, the video is slowed down to a 5-second gait cycle animation of a normal person walking. Clear red markings are made for the tibialis anterior and gastrocnemius muscles in the lower leg to guide the participant: bilateral markings for full foot strike, gastrocnemius marking for heel-off strike, no marking for swing phase, and tibialis anterior marking for heel-on strike. During the motor imagery task, participants are required to refer to the non-immersive third-person walking animation prompts to imagine a motor imagery task consistent with the previous requirements. The entire task process is one 5-second video playback time. After the video playback ended, the screen displayed "Rest" for 3 seconds to help participants relieve fatigue from high-intensity mental activity. The paradigm preparation prompt then reappeared, and the cycle continued into the next trial.

[0095] It should be noted that: in step two and subsequent steps three and four, walking animation can be played or not. That is, playing walking animation is not mandatory. In this embodiment, the display threshold and accuracy rate are to be displayed on the monitor, and the progress is also to be displayed. Considering that setting walking animation has the effect of enhancing motion imagination, walking animation is selected to be played in this embodiment.

[0096] The resting task is as follows: After all the motor imagery tasks are completed, the screen will display the prompt "Please prepare for the resting task, press the space bar to start". The subject will press the button according to the actual situation, and a green cross will appear on the screen. Each trial lasts for 5 seconds until all resting trials are completed and the cross disappears. The subject is required to remain in a resting state during this time period.

[0097] The collected EEG signals are saved and preprocessed to obtain clean EEG signals suitable for classification, and a classifier is trained. The specific process includes:

[0098] First, remove peripheral, useless electrodes susceptible to electromyographic interference ('Fpz', 'Fp1', 'Fp2', 'AF7', 'AF8', 'F7', 'F8', 'FT7', 'FT8', 'T7', 'T8', 'TP7', 'TP8', 'P7', 'P8', 'PO7', 'PO8', 'Oz', 'O1', 'O2', 'ECG', 'HEOR', 'HEOL', 'VEOU', The data corresponding to 'VEOL' was collected, and then bandpass filtered using an 8-30Hz filter. Using a whole-brain average reference, EEG signals of frequencies related to lower limb brain regions and motor imagery were obtained. Considering the balance between subject waiting time and the amount of data input to the classifier, a sliding window with a 2s window length and a 1s step size was used to segment the 5s of EEG data collected in each trial, resulting in 4 data segments per trial. For each data segment, data from the motor imagery task was marked as "1", and the resting state was marked as "0". This process was repeated m times to obtain 4m data segments. A 4m*(2s*250Hz)*n three-dimensional matrix dataset was then constructed for training the decoder. 2s*250Hz represents the number of time sampling points, and n represents the number of EEG channels. It should be noted that n represents the number of channels remaining after removing useless electrodes from the EEG signal acquisition device. The number of channels in the EEG signal acquisition device is related to the specific EEG signal acquisition device used.

[0099] The decoder uses the least Riemann average distance algorithm to classify EEG signals, establishing a binary classification model for lower limb motor imagery and resting state. A ten-fold cross-validation method is employed, dividing all EEG data into ten equal parts. One part is used as the test set, and the others are used to train the model, repeated ten times. A confusion matrix is ​​built for each trained model against the test set, and the average accuracy of correct classification over ten training iterations is calculated. This metric is used to evaluate model performance. To avoid overfitting, the weights of the last trained model are used as the final model weights, which are then used for control of the visual feedback lower limb walking online system.

[0100] Step 3: Evaluate the false positive rate of the trained model online.

[0101] Based on the online transmission interface provided by the selected EEG acquisition device's software, communication between the online EEG acquisition device and the classification feedback program is established. The subject stands at a comfortable distance and looks at the computer screen, which displays a green cross, prompting the subject to perform a resting task for 2 minutes. The subject's resting-state EEG signal is acquired, decoded by a decoder, and processed using a smoothing algorithm to obtain the real-time classification accuracy, which is then transmitted to the display interface. The subject's resting-state classification result waveform is obtained from the interactive interface, and the model's false positive rate is checked. Since motor imagery is greatly influenced by the subject's psychological state, observing a good resting-state result waveform can enhance the subject's trust and confidence in motor imagery training to a certain extent, which has a positive effect on subsequent imagery. Each signal segment (in this embodiment, a signal with a time window of 2 seconds) corresponds to a classifier's classification result, which is converted into a label: 1 represents motor imagery; 0 represents resting state. Considering the transmission rate, the online classification time window is 2 seconds, and the step size is 0.2 seconds. The classification accuracy at time t is... Calculated using the following smoothing algorithm:

[0102]

[0103] in, The smoothing coefficient is set to 0.05 by default. The accuracy from the previous time step is updated to 0 at the start of each new trial. This is the current classification result.

[0104] As can be seen from the formula and definition, when the false positive rate is low, the classification results of resting trials should be mostly 0 (resting state), with almost no classification results of 1 (motor imagery state), and the classification accuracy waveform should be a curve with small fluctuations close to 0. When the model's false positive rate is too high, retraining the model or recalibrating the subjects' imagery should be considered based on the actual results.

[0105] Step Four: Officially begin the motor imagery experiment:

[0106] Reference for subject standing position and monitor placement Figure 1 The subject places both feet on the starting point of the footprint-shaped marker. The monitor is positioned so as not to obstruct the subject's standing position after the marker ends, with a distance of approximately 2.5 meters between them. The experimenter inquires about the subject's condition, and the subject agrees to begin the first trial. The task procedure is as follows: Figure 3 As shown.

[0107] Once the subject is ready, the experimenter presses the start button on the interactive interface, denoted as t=0s. The experimenter then plays the preparation audio "Please prepare for gait motor imagery" and "Begin imagery," initiating the lower limb motor imagery task. Simultaneously, the experimenter begins collecting the subject's EEG signals for this trial and transmits them to the decoder. The display screen during the task phase is as follows... Figure 4As shown. The subject performs a mental task according to the footprints on the ground or according to the footprints on the ground and the screen animation prompts, with a maximum duration of 10 s. During this period, the subject can notice the classification accuracy rate and the threshold position displayed on the screen, determine the decoding situation at that time, and appropriately adjust the imagination method and intensity. Among them, the threshold is the classification accuracy rate required for the trigger signal of a single motor imagery experiment, which can be set in real time by the experimenter (it can be set in fact, but it will take effect in the next trial), with a default value of 0.65, and adjusted as appropriate according to the subject's performance. Regardless of the subject's imagination situation, the paradigm display interface will give the subject a certain degree of progress incentive over time, and its incentive accuracy rate that increases with time is . The incentive accuracy rate fills the blank period between the start of the task and the appearance of the first classification accuracy rate, which eliminates negative psychological states such as frustration or confusion that may occur when the subject waits for the feedback result. At the same time, the provided smooth and positively increasing incentive accuracy rate can encourage the subject to continue with motor imagery and increase the overall result of motor imagery. Therefore, the display values (display accuracy rate and display threshold ) seen by the subject are the normalized results of the sum of the actual values (actual accuracy rate and actual threshold ) and the incentive accuracy rate . The specific calculation process of the display accuracy rate is as follows:

[0108] (1) When t = 1 s, set the value of the incentive accuracy rate to 0, start receiving the decoding results transmitted by the decoder, and recalculate the display threshold as:

[0109]

[0110] Among them is the incentive value. If this value is too small, the early incentive is too slow and too small to have an encouraging effect. If this value is too large, the actual accuracy rate will be overwhelmed by the incentive and there will be no normal result feedback function. In this example, is set to 0.5.

[0111] (2) When 1 s < t < 3 s, the incentive accuracy rate increases by every 0.2 s according to the step size, and at the same time, continuously receive the current classification accuracy rate and calculate the display accuracy rate :

[0112]

[0113]

[0114] Among them,​ Indicates to Round up.

[0115] (3) t=3s, the excitation accuracy will be... Fixed as It continues to receive the current classification accuracy in a loop and calculates and displays the accuracy according to the formula above.

[0116] When the subjects successfully achieved the desired results in the motor imagery task Alternatively, the experimenter can manually trigger the trigger, and the computer stores a trigger signal, ending the trial after 3 seconds. The trigger signal can be used to activate external devices such as exoskeletons, or as the starting signal for lower limb motor rehabilitation training. In this example, it is only used to provide feedback on successful triggering; the computer announces "Successfully triggered" and displays "Trial successful" on the screen, indicating that the subject's lower limb motor imagery result for this trial is good. If the subject fails to trigger, the computer announces "Please rest" and ends the trial after 3 seconds. After the trial ends, the experimenter asks the subject if they are ready to proceed to the next trial. If the subject fails to trigger the trigger for two consecutive trials, the experimenter should manually trigger it in the third trial, adjusting the trigger threshold appropriately so that the subject can trigger it independently, to prevent the subject from losing confidence after consecutive failures and affecting subsequent experimental results. After all trials are completed, the EEG and classification results are saved for subsequent analysis to obtain relevant characteristics of the subject's lower limb motor imagery.

[0117] This invention proposes and constructs a system for online lower limb motor imagery based on visual feedback. This system intuitively guides subjects through lower limb imagery tasks, collects electroencephalogram (EEG) signals, and achieves effective online recognition of lower limb motor imagery. For example... Figure 5 As shown, the recognition results of a single motor imagery process by the test subject are selected. The horizontal axis represents the subject's classification accuracy for that trial, and the vertical axis represents time in seconds. The orange line represents the classification result at each moment, the blue line represents the classification accuracy at each moment, and the gray line represents the trigger threshold set for that trial. The subject maintained the motor imagery judgment for 4.0 seconds and successfully triggered it at 5.6 seconds. This signal can be used for the control of other devices or as a prompt to start other brain function training.

[0118] The above examples of the present invention are merely illustrative of the computational model and process of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is impossible to exhaustively list all possible implementations here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A lower limb motor imagery brain-computer interface signal acquisition system based on walking target orientation, characterized in that, It includes at least an EEG signal acquisition device and a spatial environment for lower limb motor imagery; The spatial environment includes ground start and end lines and footprint markers spaced between the start and end lines, as well as a display device; the display device displays the subject's stimulated lower limb motor imagery accuracy during the motor imagery task. The process of demonstrating the accuracy of stimulated lower limb motor imagery includes: (1) When t = t1, let the excitation accuracy rate be 0, start receiving the decoding result obtained by the decoder, and simultaneously recalculate the display threshold value according to the actual threshold value : Among them, threshold The classification accuracy required to trigger a motion imagery experiment; This is the incentive value; The decoder is used to decode the EEG signals of the subject in the lower limb motor imagery task to obtain the decoding result, that is, the classification result of the EEG signal; (2) When t1 < t < t3, the excitation accuracy rate changes according to the time step, and the current classification accuracy rate is received in a loop , the display accuracy rate is calculated : in, Indicates to Rounded up, ts is the time step of the change in the accuracy of the stimulus; t1 is the moment when the lower limb motor imagery begins; t3 is the moment when the EEG signal ends at the beginning of the lower limb motor imagery; t represents time t. (3) t = t3, which will affect the accuracy of the stimulus. Fixed as It continues to receive the current classification accuracy in a loop and calculates and displays the accuracy according to the formula above.

2. The walking target-oriented lower limb motor imagery brain-computer interface signal acquisition system according to claim 1, wherein, The display device shows the subject a walking animation corresponding to the gait cycle during the motor imagery task. During this process, muscle markings are made on the tibialis anterior and gastrocnemius muscles of the lower leg according to the content corresponding to the gait cycle to prompt the subject.

3. The walking target-oriented lower limb motor imagery brain-computer interface signal acquisition system according to claim 2, wherein, The walking animation is presented in a non-immersive third-person perspective.

4. The walking target-oriented lower limb motor imagery brain-computer interface signal acquisition system according to claim 1, wherein, The excitation value is set to 0.

5.

5. The walking target-oriented lower limb motor imagery brain-computer interface signal acquisition system according to claim 1, wherein, The decoder is obtained through the following steps: S201. Calibrate the decoder: The specific calibration paradigm is divided into a motor imagery task and a resting task. The number of trials for both tasks is the same. The motor imagery task is completed first, followed by the resting task. The motor imagery task consists of: a controlled trial begins, lasting for a period of time for baseline correction; then a motor imagery prompt is given for a duration of t1 to remind the subject to prepare for the mental task; then lower limb motor imagery begins, with the entire task lasting for t5; then a rest period is given to allow the subject to relieve fatigue from the high-intensity mental activity; then the paradigm preparation prompt reappears, and the cycle continues to the next trial. The resting task consists of the following steps: After all the motor imagery tasks are completed, the participants will rest until all resting trials are finished; the participants are required to remain in a resting state during the period from the start of the resting trial to the end of the resting trial. The collected EEG signals are saved and preprocessed to obtain EEG signals for classification, and a classifier is trained. The specific process includes: First, remove the data corresponding to the useless channels of the EEG signal acquisition device, and then use a bandpass filter. Then, a sliding window with a length of tm and a sliding step of tn is used to segment the EEG data of length tL collected in each trial. For each segment of data, the data of the motor imagery task is marked as "1" and the resting state is marked as "0". This is done for m trials to obtain multiple segments of data, which are used to train the decoder. The decoder classifies the EEG signals and establishes a binary classification model for lower limb motor imagery and resting state; it adopts ten-fold cross-validation and takes the weights of the last trained model as the final model weights, which are then used to control the visual feedback lower limb walking online system. S202. Evaluate the false positive rate of the trained model online: The subject performs a resting task, and the resting state electroencephalogram of the subject is collected, the electroencephalogram data is segmented by using a sliding window, is sent into a decoder for decoding, and the classification accuracy at t time is calculated by the following formula: in, For smoothing coefficients, The accuracy from the previous time step is updated to 0 at the start of each new trial. This is the current classification result; Examine the false positive rate of the model and determine whether to retrain the model or recalibrate the subjects by providing visualization guidance.

6. The walking target-oriented lower limb motor imagery brain-computer interface signal acquisition system according to claim 4, wherein, The decoder uses the minimum Riemann average distance algorithm.

7. A lower limb motor imagery brain-computer interface method based on walking target orientation, characterized in that, The brain-computer interface based on the walking goal-oriented lower limb motor imagery brain-computer interface signal acquisition system according to any one of claims 1 to 6 specifically includes: the subject's paradigm imagination steps, and the actual motor imagery steps. The steps of the subject paradigm imagination include: Before collecting EEG signals, subjects were asked to walk along the footprints on the ground, perform foot movements and perceive leg activity. After each walk, they returned to their original position, observed the footprints on the ground, and recalled the walking process, using the walking recall as a template for paradigmatic imagination. Subjects should perform lower limb motor imagery exercises at the starting point. During the lower limb motor imagery process, subjects are required to observe footprint marks and recall the corresponding walking movements, images, and sensations. Electroencephalogram (EEG) signals are collected during lower limb motor imagery and at rest. The steps of the actual motion visualization include: Subjects should observe the display device at the starting point and perform lower limb motor imagery. During the lower limb motor imagery process, subjects are required to recall the images and sensations of the corresponding walking movements. Electroencephalogram (EEG) signals are collected during the actual motor imagery process and at rest. The display device shows the subjects the accuracy of the stimulated lower limb motor imagery.

8. The walking target-oriented lower limb motor imagery brain-computer interface method according to claim 7, characterized in that, In the actual motor imagery step, the subject should observe the display device at the starting point to perform lower limb motor imagery. The display device is also used to play walking animations.

9. The walking target-oriented lower limb motor imagery brain-computer interface method according to claim 7 or 8, characterized in that, In the process of using walking memory as a template for paradigmatic imagination, the template should include the entire gait cycle of walking. A walking gait cycle includes: (1) the whole foot strikes the ground, (2) the heel lifts off the ground, (3) the swing phase, and (4) the heel strikes the ground.

Images