A training method based on motor imagery combined with dream guidance

Abstract

The present invention discloses a training method based on motor imagery combined with dream guidance. The method comprises the following steps: collecting EEG data of motor imagery training offline, performing data preprocessing and feature extraction, and training a motor imagery classification model; extracting EEG data of motor imagery on the affected side online, and performing classification prediction using the previously trained motor imagery classification model to obtain statistical results; performing classification prediction on the nighttime data model obtained through nighttime dream guidance to obtain statistical results; comparing the classification prediction results of the daytime data model with the nighttime data model, screening out the common correct data parts for data superposition, and retraining the classification model using the superimposed data model. If the accuracy rate is higher than that of the daytime data model, the classification model is retained; otherwise, the previous classification model is used. The present invention processes the data generated by dream intervention and combines it with the motor imagery data model to improve the accuracy of the classification model, thereby achieving correct feedback to the patient.

Description

Technical Field

[0001] The present invention belongs to the field of electroencephalogram (EEG) data processing, and in particular relates to a training method based on motor imagery combined with dream guidance. Background Art

[0002] Since its inception, brain-computer interfaces have been widely used in various fields, especially in medicine, where they are used as an effective preventive measure. For example, they can predict diseases such as epilepsy, ADHD, and Alzheimer's disease in advance. They also play a prominent role in postoperative recovery, especially for neurological diseases such as hemiplegia caused by stroke, which can be treated through motor imagery using EEG technology.

[0003] However, the principles of motor imagery require a robust and extensive data set as a foundation. Lack of any of these factors can negatively impact experimental results. This requires long, repetitive training sessions, which are not only time-consuming and labor-intensive but can also lead to negative treatment attitudes. Eliminating this situation is a pressing issue.

[0004] Existing research indicates that a specific EEG signature can be used to identify a patient's state of consciousness: dreaming, deep sleep, wakefulness, and so on. Appropriately utilizing this signature can identify the onset of dreams. Two current dream intervention methods are: one that identifies the onset of dreaming and provides light stimulation; the other that uses wakefulness waves to induce dreaming trends. Using dream states to guide patients through rehabilitation training without disrupting sleep can address issues of both the duration of recovery and the volume of data required. Summary of the Invention

[0005] In order to address the deficiencies in the prior art, the present invention aims to provide a training method based on motor imagery combined with dream guidance.

[0006] To achieve the purpose of the present invention, the technical solution adopted by the present invention is:

[0007] A training method based on motor imagery combined with dream guidance, comprising the following steps:

[0008] (1) Offline collection of 4-second EEG data from motor imagery training until the set number of times is reached, and the data is stored in the daytime data model. The collected EEG data includes information on the affected side and the normal side; data preprocessing and feature extraction are performed to train the motor imagery classification model;

[0009] (2) Online extraction of motor imagery EEG data of the affected side until the set number of times is reached, and the data is stored in the daytime data model. The previously trained motor imagery classification model is used to perform classification prediction to obtain statistical results; if the predicted result is the same as the affected side, it is correct, otherwise it is wrong;

[0010] (3) Classify and predict the nighttime data model obtained through nighttime dream guidance to obtain statistical results; if the predicted result is the same as the affected side, it is correct, otherwise it is wrong;

[0011] (4) Compare the classification prediction results of the daytime data model and the nighttime data model, select the common correct data parts for data superposition, and obtain the superimposed data model; use the superimposed data model to train the classification model again. If the accuracy is higher than that of the daytime data model, retain this classification model; otherwise, use the previous classification model.

[0012] Furthermore, step (3) specifically includes the steps of:

[0013] (3.1) Wear an EEG acquisition device to collect EEG data during sleep and calculate the frequency domain characteristics of EEG waves;

[0014] (3.2) Determine whether the sleep state is deep sleep based on the frequency domain characteristics of brain waves and control the sleep monitoring system to perform different operations;

[0015] (3.3) If the sleep state reaches a deep state, the dream guidance system is controlled to guide the stimulation and conduct the same motor imagery training as during the day to obtain a nighttime data model;

[0016] (3.4) Perform motor imagery training according to the sleep cycle cycle to obtain a night data model of multiple sleep cycles.

[0017] Furthermore, in step (3.1), EEG data is processed by Fourier transform or wavelet transform to calculate the frequency domain characteristics of EEG waves, distinguish different biological signals during sleep, and use them to determine non-rapid eye movement (NREM) and rapid eye movement (REM) periods.

[0018] Furthermore, in step (3.2), the sleep monitoring system includes a sleep aid system, a dream guidance system, and a wake-up system;

[0019] When in the non-rapid eye movement period or preparing to fall asleep, the sleep aid system provides sleep-aid stimulation to enable the patient to fall asleep quickly or enter the next sleep stage, including music sleep aid and electric stimulation massage sleep aid; when in the rapid eye movement period, the dream guidance system provides guidance stimulation; when the time reaches the end, the awakening system provides awakening stimulation.

[0020] Furthermore, in step (3.3), the corresponding part is stimulated according to the rehabilitation part set by the patient, and the same motor imagery training experiment as during the day is conducted during the stimulation process. Each stimulation lasts for 4 seconds until the set number of times is reached, and the EEG data of the imagination training is stored in the night data model.

[0021] Furthermore, label bits are added to the EEG data, and the data is distinguished according to the label bits as daytime training data or nighttime training data.

[0022] Furthermore, stimulation guidance includes electrical stimulation guidance and sound stimulation guidance.

[0023] Furthermore, electrical stimulation is guided. When the sleep state enters the rapid eye movement period, the electrical stimulation device output is turned on, and the current is output to the set part according to the set rehabilitation part. The output current En increases or decreases according to the feedback of each deep sleep state.

[0024] Furthermore, with sound stimulation guidance, when the sleep state enters the rapid eye movement period, the microphone mode is turned on, and the corresponding part or the activities performed on the corresponding part or the words related to the part are spoken according to the set rehabilitation part.

[0025] The beneficial effect of the present invention is that, compared with the prior art, the present invention processes the data generated by dream intervention and combines it with the data model of motor imagery to improve the accuracy of the classification model, thereby achieving correct feedback to the patient.

[0026] The present invention uses EEG acquisition equipment combined with sleep status to determine whether dreams are connected, and then uses external output as intervention guidance conditions, such as electrical, acoustic, light, and magnetic stimulation, to intervene and guide through specific parts or specific output waves, so that the patient can imagine specific parts in the dream for neural repair.

[0027] The present invention uses EEG acquisition equipment combined with the motor imagery paradigm to guide patients to perform corresponding motor imagery, so that patients can re-recognize the movements of the training parts during the training process to achieve rehabilitation effects. In addition, the rehabilitation effect can also be achieved by driving the movement of the affected limb through hardware equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of the training method based on motor imagery combined with dream guidance according to the present invention;

[0029] Figure 2 It is a sleep monitoring diagram;

[0030] Figure 3 is a schematic diagram of a sleep monitoring system;

[0031] Figure 4 This is a schematic diagram of sleep biological signals;

[0032] Figure 5 This is a diagram of the sleep cycle. DETAILED DESCRIPTION

[0033] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and are not intended to limit the scope of protection of this application.

[0034] The training method based on motor imagery combined with dream guidance, described in the present invention, involves performing motor imagery training and training a motor imagery classification model during the day. Using an EEG acquisition device combined with motor imagery, it achieves mirror learning and cognitive learning, thereby achieving brain-neuronal repair. During sleep, the EEG acquisition device collects EEG data during sleep to analyze the user's sleep stage. If the user is in deep sleep, appropriate stimulation is performed to map the areas requiring rehabilitation. The data from the stimulation stage is then saved and provided to the motor imagery classification model, thereby increasing the data volume of the motor imagery classification model and eliminating many environmental interference factors due to the deep sleep state, ensuring data quality.

[0035] like Figure 1 As shown, the training method based on motor imagery combined with dream guidance of the present invention has the following specific steps:

[0036] Step 1, EEG data acquisition and preprocessing;

[0037] Wear an EEG acquisition device to collect EEG data, and pre-process the collected EEG data to remove power frequency interference. Because the EEG data signal characteristic range is within the range of 0.1Hz to 30Hz, the raw data needs to be filtered through a bandpass filter.

[0038] Step 2: Perform daytime motor imagery training to obtain a daytime data model for training the motor imagery classification model; compare the daytime data model with the nighttime data model, and superimpose the correct data parts for further training of the motor imagery classification model;

[0039] During the offline phase, motor imagery training was conducted using a motor imagery paradigm. Four seconds of EEG data from these training sessions were stored in the daytime data model. The collected EEG data included both affected and normal-side information. No fewer than 60 EEG data acquisitions were performed under the motor imagery paradigm. The data model was preprocessed, and features were extracted using a common space model algorithm. A motor imagery classification model was trained, and two results were classified using a support vector machine. These results were then compared to determine accuracy.

[0040] During the online phase, several EEG data sets from motor imagery of the affected side are extracted and stored in the daytime data model. Each extraction is then classified using the previously trained motor imagery classification model. If the predicted result matches the affected side, the result is considered correct; otherwise, it is considered incorrect. Furthermore, the nighttime data model, derived through dream guidance, is also used to generate statistical results through classification and prediction.

[0041] After the online phase, the daytime and nighttime data models are compared, and the commonly correct data is selected and superimposed to form a superimposed data model. The superimposed data model is then used to train the classification model again. If the accuracy is higher than that of the daytime data model, the superimposed data model is used and retained. Otherwise, the previous data model and classification model are used.

[0042] The daytime data model refers to the EEG data generated by patients during wakefulness through active motor imagery training, while the nighttime training model refers to the EEG data generated by patients during deep sleep through passive stimulation. Both models are built using data accumulated after several sessions of motor imagery training, both offline and online.

[0043] like Figure 2 As shown, at night, the same motor imagery training as during the day is conducted through dream guidance, and the data is stored in the night data model. EEG data during sleep is collected by EEG acquisition equipment to distinguish sleep states and determine whether the current sleep state is deep sleep. If it is deep sleep, the stimulation source is activated for guidance; if it is not deep sleep, the sleep aid system is activated to accelerate sleep.

[0044] Through stimulation and guidance, according to the part of the patient that needs rehabilitation, the final result of guiding the dream through external intervention after entering the dream is the part that needs rehabilitation; finally, the data generated by stimulation and guidance will be used to create a night data model separately, and compared with the daytime data model, and the correct part of the data will be merged with the motor imagination data to enhance the model effect, thereby increasing the classification accuracy of the motor imagination model and improving the accuracy of the patient's motor imagination.

[0045] The specific steps are as follows:

[0046] Step 1: EEG data collection and processing, calculating the frequency domain characteristics of brain waves;

[0047] Wearing EEG acquisition equipment, Figure 3 In the sleep monitoring system shown, the sleep start time, sleep duration, rehabilitation site (guidance site), and stimulation source type (electricity, light, sound, magnetism, etc.) are set. When the time reaches the preset sleep start time, the EEG acquisition device is turned on and real-time data transmission is performed.

[0048] The sleep monitoring system includes a sleep aid system, a dream guidance system and an awakening system; sleep aid stimulation is provided through the sleep aid system, guidance stimulation is provided through the dream guidance system, and awakening stimulation is provided through the awakening system.

[0049] EEG data processing is to calculate the frequency domain characteristics of brain waves through Fourier transform or wavelet transform, and distinguish different biological signals during sleep, such as Figure 4 It is mainly divided into four parts: non-rapid eye movement stage 1, when the characteristic wave is mainly 8-13Hz alpha wave, with an amplitude of 20-40μV; non-rapid eye movement stage 2, when the characteristic wave is 11-16Hz sleep spindle wave and K complex wave, with an amplitude less than 50μV; non-rapid eye movement stage 3, when the characteristic wave is 0.5-2Hz delta wave, with an amplitude greater than 75μV; rapid eye movement stage, when the characteristic wave is 3-7Hz theta wave, with an amplitude less than 50μV.

[0050] Step 2: judging whether the sleep state is a deep sleep state based on the frequency domain characteristics of the brain waves, and controlling the sleep monitoring system to perform different operations;

[0051] During sleep, there are 4 to 6 sleep cycles, which go from non-rapid eye movement to rapid eye movement and then back to non-rapid eye movement. Based on the real-time filtering processing in step 1, the frequency domain characteristics of the brain waves in each period are obtained, and whether the REM period (deep sleep state) has been entered or exited is determined in real time. The sleep aid system, dream guidance system, and wake-up system need to be switched in real time.

[0052] When the waveform characteristics are those of the non-rapid eye movement (NREM) period or the sleep preparation period, the sleep aid system provides corresponding sleep-aid stimulation to enable the patient to fall asleep quickly or enter the next sleep stage; including: music to aid sleep, which is also the most commonly used way for people to fall asleep, and is very suitable for non-REM 1 or the sleep onset period; electrical stimulation massage to aid sleep, which can release weak electric current through the stimulation electrodes on the scalp to massage the scalp, causing the person to enter a state of fatigue and fall asleep quickly.

[0053] When in the rapid eye movement period, guidance stimulation is provided through the dream guidance system; when the time reaches the end, awakening stimulation is provided through the awakening system.

[0054] Step 3: If the sleep state reaches a deep state, the dream guidance system is controlled to guide and stimulate the sleep state, and the same motor imagery training as during the day is performed to obtain a nighttime data model.

[0055] When the characteristic waveform is the rapid eye movement period (deep sleep state), it enters the guidance mode. Because dreams occur during the rapid eye movement period, the corresponding parts are stimulated according to the rehabilitation parts set by the patient.

[0056] During the guided stimulation process, the same motor imagery training experiment as during the day is conducted, with a stimulation duration of 4 seconds. The 4-second EEG data from the imagery training is stored in the nighttime data model. This EEG data is then labeled and used to distinguish the data phase. The rest period between each stimulation can be adjusted in advance.

[0057] The label bit is used to distinguish daytime training data (active training data set under waking consciousness) from nighttime training data (passive training data set during the sleep stage). Its function is to distinguish the data so as not to confuse the specified model when training the model.

[0058] Guidance includes electrical stimulation guidance and sound stimulation guidance. For electrical stimulation guidance, when the sleep state enters the rapid eye movement period, the output of the electrical stimulation device is turned on, and the current is output to the set part according to the set rehabilitation part. The output current En increases or decreases according to the feedback of each deep sleep state. For example, when the En value output increases to K, the EEG monitor detects that the rapid eye movement period ends and enters the non-rapid eye movement or short-term wakefulness state, and switches to the sleep aid mode and records the current value K. The starting value will be K-Kn when entering the deep sleep period next time, and the operation will be repeated until the time is up or the sleep is awakened. In addition, the patient's feedback on the sleep quality of this time will be recorded. If it is found that it affects the sleep quality, the En value K will be lowered again when the user uses it next time.

[0059] Sound stimulation guidance: when the sleep state enters the rapid eye movement period, turn on the microphone mode, and speak the corresponding part or the activities of the corresponding part or the words related to the part according to the set rehabilitation part. For example, when the rehabilitation part is the nerve that controls the left foot, the words "walking", "running", "left foot" and other guiding words are told to the patient through sound. The volume control is the same as the sound volume in the sleep aid mode, and can be accompanied by music.

[0060] Step 4, performing motor imagery training according to the sleep cycle cycle to obtain a night data model of multiple sleep cycles;

[0061] according to Figure 5 As shown in the sleep cycle, the mode will be switched according to the current state in each sleep cycle until the patient wakes up or the set duration is reached to start the wake-up mode. At this time, the sleep-aiding stimulus or dream-guiding stimulus will be turned off first, and then the wake-up stimulus will be turned on and increased continuously until the patient's EEG signal is detected as being in an awake state, and then the wake-up stimulus will be stopped.

[0062] During the rapid eye movement period (deep sleep state) of each sleep cycle, the dream guidance system is controlled to provide guidance and stimulation, and the same motor imagery training as during the day is carried out to obtain a night data model.

[0063] A set of data will be saved for one deep sleep cycle. Through the creation time, size, quality and real-time stimulation data of the data file, we can get the number of times the patient enters deep sleep, the duration, and the different lengths of deep sleep combined with the intensity of the stimulation source, and infer the best guidance method, duration and stage, so as to gradually improve the guidance method.

[0064] The beneficial effect of the present invention is that, compared with the prior art, the present invention processes the data generated by dream intervention and combines it with the motor imagery data model to improve the accuracy of the classification model, thereby achieving correct feedback to the patient.

[0065] The present invention uses EEG acquisition equipment combined with sleep status to determine whether dreams are connected, and then uses external output as intervention guidance conditions, such as electrical, acoustic, light, and magnetic stimulation, to intervene and guide through specific parts or specific output waves, so that the patient can imagine specific parts in the dream for neural repair.

[0066] The present invention uses EEG acquisition equipment combined with the motor imagery paradigm to guide patients to perform corresponding motor imagery, so that patients can re-recognize the movements of the training parts during the training process to achieve rehabilitation effects. In addition, the rehabilitation effect can also be achieved by driving the movement of the affected limb through hardware equipment.

[0067] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation plans of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, and is not a limitation on the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

Claims

1. A training method based on motor imagery combined with dream guidance, characterized in that: Including steps: (1) Offline collection of 4-second EEG data from motor imagery training until the set number of times is reached, and the data is stored in the daytime data model. The collected EEG data includes information on the affected side and the normal side; data preprocessing and feature extraction are performed to train the motor imagery classification model; (2) Online extraction of motor imagery EEG data of the affected side until the set number of times is reached, and the data is stored in the daytime data model. The previously trained motor imagery classification model is used for classification prediction to obtain statistical results; if the predicted result is the same as the affected side, it is correct, otherwise it is wrong; (3) Classify and predict the nighttime data model obtained through nighttime dream guidance to obtain statistical results; if the predicted result is the same as the affected side, it is correct, otherwise it is wrong; The specific steps include: (3.1) Wear an EEG acquisition device to collect EEG data during sleep and calculate the frequency domain characteristics of the EEG waves; (3.2) Determine whether the sleep state is deep sleep based on the frequency domain characteristics of brain waves and control the sleep monitoring system to perform different operations; (3.3) If the sleep state reaches a deep state, the dream guidance system is controlled to guide and stimulate the sleep state, and the same motor imagery training as during the day is performed to obtain a nighttime data model. (3.4) Performing motor imagery training according to the sleep cycle to obtain a nightly data model of multiple sleep cycles; (4) Compare the classification prediction results of the daytime data model and the nighttime data model, select the common correct data parts for data superposition, and obtain the superimposed data model; use the superimposed data model to train the classification model again. If the accuracy is higher than that of the daytime data model, retain this classification model; otherwise, use the previous classification model.

2. The training method based on motor imagery combined with dream guidance according to claim 1, characterized in that: Step (3.1) processes EEG data through Fourier transform or wavelet transform, calculates the frequency domain characteristics of EEG waves, distinguishes different biological signals during sleep, and is used to determine non-rapid eye movement (NREM) and rapid eye movement (REM) periods.

3. The training method based on motor imagery combined with dream guidance according to claim 1, characterized in that: Step (3.2), the sleep monitoring system includes a sleep aid system, a dream guidance system, and a wake-up system; When in the non-rapid eye movement period or preparing to fall asleep, the sleep aid system provides sleep-aid stimulation to enable the patient to fall asleep quickly or enter the next sleep stage, including music sleep aid and electric stimulation massage sleep aid; when in the rapid eye movement period, the dream guidance system provides guidance stimulation; when the time reaches the end, the awakening system provides awakening stimulation.

4. The training method based on motor imagery combined with dream guidance according to claim 1, characterized in that: In step (3.3), the corresponding part of the body is stimulated according to the rehabilitation part set by the patient. During the stimulation process, the same motor imagery training experiment as during the day is conducted. Each stimulation lasts for 4 seconds until the set number of times is reached. The EEG data of the imagery training is stored in the night data model.

5. The training method based on motor imagery combined with dream guidance according to claim 4, characterized in that: Add label bits to the EEG data and use the label bits to distinguish whether the data is in the daytime training data or nighttime training data.

6. The training method based on motor imagery combined with dream guidance according to claim 4, characterized in that: Stimulation guidance includes electrical stimulation guidance and sound stimulation guidance.

7. The training method based on motor imagery combined with dream guidance according to claim 6, characterized in that: Electrical stimulation guidance: when the sleep state enters the rapid eye movement period, the electrical stimulation device output is turned on, and the current is output to the set part according to the set rehabilitation part. The output current En increases or decreases according to the feedback of each deep sleep state.

8. The training method based on motor imagery combined with dream guidance according to claim 6, characterized in that: Sound stimulation guidance: when the sleep state enters the rapid eye movement period, turn on the microphone mode and speak the corresponding part or the activities performed on the corresponding part or the words related to the part according to the set rehabilitation part.

Images