A rehabilitation training system

Abstract

The application discloses a stroke rehabilitation training system, which comprises a rehabilitation training task recommendation module, a rehabilitation training module and a rehabilitation training task completion evaluation module, wherein the rehabilitation training task recommendation module is used for recommending adaptive rehabilitation training tasks for a patient according to a rehabilitation evaluation result of the patient; the rehabilitation training module is used for performing rehabilitation training on the patient according to the rehabilitation training tasks recommended by the rehabilitation training task recommendation module; the rehabilitation training task completion evaluation module is used for evaluating the rehabilitation training completion degree of the patient and feeding back to the rehabilitation training task recommendation module; and the rehabilitation training task recommendation module comprises a rehabilitation training task recommendation module fusing motor function and brain function.

Description

[0001] This application is a divisional application of the invention patent application with application number "202210812775.3", application date July 12, 2022, and invention title "A Stroke Rehabilitation Training System". Technical Field

[0002] This invention relates to the field of rehabilitation training technology, and in particular to a rehabilitation training system. Background Technology

[0003] Stroke is the leading cause of death and disability among Chinese residents, characterized by high incidence, high disability rate, high mortality rate, and high recurrence rate. Despite access to rehabilitation therapy, approximately 60%-80% of stroke patients still suffer from significant motor dysfunction, imposing heavy care costs on families and society. Therefore, high-quality and efficient rehabilitation training is the most important means to address this problem and an inevitable choice to help patients regain independence and reintegrate into society.

[0004] Studies have shown that stroke patients primarily suffer from damage to neural circuits. Therefore, rehabilitation assessments and training should comprehensively consider information on brain function to assist physicians in providing adaptive rehabilitation training methods for patients. Consequently, recommending rehabilitation training programs that integrate motor and brain function is an important trend in future rehabilitation training.

[0005] In addition, multi-physical stimulation is also an important means of stroke rehabilitation. Non-invasive stimulation of the patient's brain, central nervous system, muscles, etc. through physical stimulation such as sound, light, and electromagnetic stimulation can improve cortical activity after a period of time, thereby promoting the reconstruction of neural circuits and motor functions. Therefore, multi-physical synergistic stimulation is an important direction for the future development of neuromodulation.

[0006] To this end, this invention proposes a training system suitable for stroke rehabilitation. It constructs an optimal rehabilitation training task recommendation model based on limb motor function, based on brain function, and integrating motor function and brain function. It is supplemented by information detection and physical stimulation methods to improve the efficiency of rehabilitation training combining brain-muscle-limb multi-source information and multi-physical stimulation, thereby achieving the best rehabilitation training effect. Summary of the Invention

[0007] To achieve the objectives of this invention, the following technical solution is adopted:

[0008] A stroke rehabilitation training system includes a rehabilitation training task recommendation module, a rehabilitation training module, and a rehabilitation training task completion evaluation module. The rehabilitation training task recommendation module recommends adaptive rehabilitation training tasks to stroke patients based on their rehabilitation assessment results. The rehabilitation training module performs rehabilitation training on stroke patients based on the recommended rehabilitation training tasks. The rehabilitation training task completion evaluation module evaluates the patient's rehabilitation training completion rate and provides feedback to the rehabilitation training task recommendation module.

[0009] The stroke rehabilitation training system includes a rehabilitation training task recommendation module comprising a limb motor function-based module and a brain function-based module. The limb motor function-based module recommends suitable rehabilitation training tasks based on the patient's motor function scale assessment results. The brain function-based module recommends suitable rehabilitation training tasks based on the patient's MRI brain image assessment results.

[0010] The stroke rehabilitation training system described herein includes a rehabilitation training task recommendation module based on limb motor function that recommends suitable rehabilitation training tasks for patients according to the patient's motor function scale assessment results as follows: A rehabilitation training task database based on the stroke patient's motor function scale is constructed, and the scale information and basic information in the database are preprocessed to obtain the scale feature vector L. RK and basic information vector J K The preprocessed scale data and basic information feature vectors are input into a convolutional neural network model for training and model optimization, ultimately resulting in a recommendation model for rehabilitation training tasks based on limb motor function.

[0011] Y RK =[BW i ,TP i HR i ,TM i ,F i ] = G 1CNN (L RK J K )

[0012] Among them, Y RK The recommended rehabilitation training tasks include the type of rehabilitation training task, difficulty level, duration, and frequency for different limb parts; the value of i is [1, 8], representing different limb parts: left foot, left lower limb, left upper limb, left hand, right foot, right lower limb, right upper limb, and right hand; BW i For a certain limb part; TP i Type of rehabilitation training task for a specific body part; HR iThe difficulty level of rehabilitation training tasks for a specific body part; TM i Time allotted for rehabilitation training tasks in a specific area; F i Frequency of rehabilitation training tasks for a specific body part; G 1CNN For a trained convolutional neural network model; L RK J is the preprocessed feature vector of the scale; K This is the patient's basic information after pretreatment.

[0013] The stroke rehabilitation training system described herein includes a brain function-based rehabilitation training task recommendation module that recommends suitable rehabilitation training tasks for the patient based on the patient's MRI brain image assessment results, in the following manner:

[0014] A database of rehabilitation training tasks based on MRI brain images of stroke patients was constructed, and the MRI brain images and report data in the database were preprocessed.

[0015] Data features of preprocessed MRI brain images and reports were extracted using a convolutional neural network model;

[0016] H NK =Feedforward(W CNN B CNN JJ CNN CH CNN ;HC,BG;)

[0017] Among them, H NK It is the feature vector extracted by the convolutional neural network model; Feedforward is the feedforward neural network function; W CNN It is the weight matrix of the convolutional neural network model, B CNN It is a bias parameter; JJ CNN It is a convolutional layer of the convolutional module, with a 2x2 kernel and a 2x2 stride; CH CNN This is the pooling layer of the convolutional module, with a 3*3 max pooling kernel; HC and BG are the input MRI brain image data and data report, respectively.

[0018] The extracted features from MRI brain images and report data are input into another convolutional neural network model for training and optimization, ultimately resulting in a brain function rehabilitation training task recommendation model.

[0019] Y NK =[BW i ,TP i HR i ,TM i ,F i ] = G 2CNN (L RK H NK JK )

[0020] Among them, Y NK The recommended rehabilitation training tasks include the type of rehabilitation training task, difficulty level, duration, and frequency for different limb parts; the value of i is [1, 8], representing different limb parts: left foot, left lower limb, left upper limb, left hand, right foot, right lower limb, right upper limb, and right hand; BW i For a certain limb part; TP i Type of rehabilitation training task for a specific body part; HR i The difficulty level of rehabilitation training tasks for a specific body part; TM i Time allotted for rehabilitation training tasks in a specific area; F i Frequency of rehabilitation training tasks for a specific body part; G 2CNN For a trained convolutional neural network model; H NK For the characteristics of MRI brain images and report data; J K This is the patient's basic information after pretreatment.

[0021] The stroke rehabilitation training system further includes a rehabilitation training task recommendation module that integrates motor function and brain function, used to comprehensively recommend suitable rehabilitation training tasks for the patient based on the patient's scale motor function data and MRI brain image data in the following manner:

[0022] The feature vectors of the scale and the features of the nuclear magnetic resonance data are input into the convolutional neural network model for training and model optimization, and finally a rehabilitation training task recommendation model that integrates motor function and brain function is obtained.

[0023] Y NH =[BW i ,TP i HR i ,TM i ,F i ] = G 3CNN (L RK H NK J K )

[0024] Among them, Y NH The recommended rehabilitation training tasks include the type of rehabilitation training task, difficulty level, duration, and frequency for different limb parts; the value of i is [1, 8], representing different limb parts: left foot, left lower limb, left upper limb, left hand, right foot, right lower limb, right upper limb, and right hand; BW i For a certain limb part; TP i Type of rehabilitation training task for a specific body part; HR iThe difficulty coefficient of a rehabilitation training task for a certain body part; TM i Training time for rehabilitation training tasks in a specific area; F i Training frequency for rehabilitation training tasks in a specific area; G 3CNN For a trained convolutional neural network model; L RK H is the feature vector of the scale. NK For the characteristics of MRI brain images and report data; J K This is the patient's basic information after pretreatment.

[0025] The stroke rehabilitation training system further includes an optimal rehabilitation training task recommendation module, which is used to recommend the aforementioned rehabilitation training task Y based on the patient's limb motor function. RK Rehabilitation training task Y based on patient brain function NK Rehabilitation training task Y integrating motor function and brain function NH By performing weight fusion, the optimal rehabilitation training task ZR is obtained. XL :

[0026] ZR XL =(aY RK +bY NK +cY NH )*G WC *m

[0027] ZR XL This is the optimal rehabilitation training task, where A, B, and C are weight matrices: A = [a1, a2, a3, ..., a8], B = [b1, b2, b3, ..., b8], C = [c1, c2, c3, ..., c8], a i b i c i The values ​​of G are all in the range of [0,1], i∈[1,8]. WC G represents the completion rate of the previous rehabilitation training task, and m is the correction coefficient. When the patient is undertaking a rehabilitation training task for the first time or the previous rehabilitation training task had a 100% match rate, G... WC The value is 1; when the patient's previous rehabilitation training task matching degree is less than 100%, G WC The value is less than 1.

[0028] The stroke rehabilitation training system described herein selects rehabilitation training tasks according to the following rules:

[0029] When three rehabilitation training tasks Y RK Y NK Y NH When the types of rehabilitation training tasks for a certain limb are inconsistent, select the rehabilitation training task that is more suitable to achieve the desired results.

[0030] When three rehabilitation training tasks Y RK Y NK Y NH If one rehabilitation training task type for a certain limb is inconsistent with the other two, select the two rehabilitation training tasks with the same type.

[0031] When three rehabilitation training tasks Y RK Y NK Y NH If the rehabilitation training tasks for a certain limb are of the same type, then select three rehabilitation training tasks of the same type.

[0032] The weighting coefficients in the aforementioned stroke rehabilitation training system are calculated as follows:

[0033]

[0034]

[0035]

[0036] Wherein, ρ(Y) RK-BWi Y NK-BWi ) for rehabilitation training task Y RK and Y NK Similarity, ρ(Y) RK-BWi Y NH-BWi ) for rehabilitation training task Y RK and Y NH Similarity, ρ(Y) NK-BWi Y NH-BWi ) for rehabilitation training task Y NH and Y NK Similarity;

[0037]

[0038]

[0039]

[0040] Among them, Y RK-HRi Y NK-HRi Y NH-HRi These are three rehabilitation training tasks Y RK Y NK Y NH The difficulty level, Y RK-TMi Y NK-TMi Y NH-TMi These are three rehabilitation training tasks Y RK Y NK YNH Training time, Y RK-Fi Y NK-Fi Y NH-Fi These are three rehabilitation training tasks Y RK Y NK Y NH The training frequency.

[0041] The stroke rehabilitation training system includes a virtual reality module, an information acquisition module, an information analysis module, and a multi-physical stimulation control module.

[0042] The stroke rehabilitation training system described above includes an information analysis module used to perform time-domain and frequency-domain feature analysis on the electrooculogram (EOG) signal after removing interference noise. The time-domain smoothness feature of the EOG signal is defined as follows:

[0043]

[0044] Among them, Z uv Z represents the smoothness of the electrooculogram signal. iu This refers to the electrooculographic signal from the U electrode (the electrode on the upper eyelid). The mean value of the electrooculographic signal from the u-electrode is... This refers to the electrooculographic signal from the V electrode (the electrode on the lower eyelid). The mean value of the electrooculographic signal at the V electrode;

[0045] The frequency domain characteristics of the average power of the electrooculogram (EOG) signal are defined as follows:

[0046]

[0047] Among them, Z PF Here, f is the average power frequency of the electrooculogram (EOG), f is the frequency of the EOG signal, and f1 and f2 represent the frequency range of the EOG signal.

[0048] Based on the aforementioned time-domain and frequency-domain features, the information analysis module establishes the following attention model indicators:

[0049] Concentration F = A × Z uv +B×Z PF

[0050] Among them, Z uv Z represents the smoothness of the electrooculogram signal. PF Let A and B be the average power frequency of the electrooculogram (EOG), and their respective weights. Experiments have shown that the focus of stroke patients is most strongly correlated with the smoothness of the EOG signal and the average power frequency of the EOG, with the smoothness of the EOG signal accounting for a larger proportion. Therefore, the weights A and B are set as follows: 0.5. <A<1,0<B<0.5,A+B=1。

[0051] The stroke rehabilitation training system according to , wherein the rehabilitation training task completion evaluation module is configured to determine the matching degree of the patient's rehabilitation training task according to the number of multi-physical stimulation调控 times, the concentration duration and the completion degree of the rehabilitation training task in the rehabilitation training module,

[0052] Rehabilitation training task matching degree:

[0053]

[0054] wherein Gc is the rehabilitation training task matching degree, T is the time for completing one rehabilitation training task, t c is the total time of physical stimulation in one rehabilitation training task, Kc is the completion degree of the rehabilitation training task, Z C is the duration for which the patient's concentration reaches the standard in one rehabilitation training task, C1, C2, C3 are weight coefficients, and the values of the weight coefficients C1, C2, C3 are set as follows: 0.1 < C1 < 0.4, 0.6 < C2 ≤ 1, 0.1 < C3 < 0.4, and C1 + C2 + C3 = 1. Description of Drawings

[0055] Figure 1 is a structural schematic diagram of the stroke rehabilitation training system of the present invention;

[0056] Figure 2 is a working flow chart of the stroke rehabilitation training system of the present invention. Detailed Description of Embodiments

[0057] Hereinafter, the specific implementation of the Figure 1-2 present invention will be described in detail with reference to the accompanying drawings.

[0058] As Figure 1 shown, the stroke rehabilitation training system comprises a rehabilitation training task recommendation module, a rehabilitation training module and a rehabilitation training task completion evaluation module. Wherein: the rehabilitation training task recommendation module is configured to recommend adaptive rehabilitation training tasks for the patient according to the rehabilitation evaluation result of the stroke patient; the rehabilitation training module is configured to conduct rehabilitation training for the stroke patient according to the rehabilitation training tasks recommended by the rehabilitation training task recommendation module; the rehabilitation training task completion evaluation module is configured to evaluate the rehabilitation training completion degree of the patient and feed back the result to the rehabilitation training task recommendation module.

[0059] The rehabilitation training task recommendation module comprises a rehabilitation training task recommendation module based on limb motor function, a rehabilitation training task recommendation module based on brain function, a rehabilitation training task recommendation module integrating motor function and brain function, and an optimal rehabilitation training task recommendation module. The detailed description is as follows:

[0060] The rehabilitation training task recommendation module based on limb motor function is used to recommend suitable rehabilitation training tasks for patients according to the patient's motor function scale assessment results, as follows:

[0061] A rehabilitation training task database based on motor function scales for stroke patients was constructed, and the scale information and basic information in the database were preprocessed: the Fugl-Meyer Rating Scale, Barthel Index, Brunnstrom Rating Scale, Berg Balance Scale, and basic information (age, gender, education, occupation, history of hypertension, etc.) were normalized and preprocessed to obtain the scale feature vectors L. RK and basic information vector J K .

[0062] The preprocessed scale data and basic information feature vectors are input into a convolutional neural network model for training and model optimization. The model parameters are continuously optimized to finally obtain a rehabilitation training task recommendation model based on limb motor function.

[0063] Y RK =[BW i ,TP i HR i ,TM i ,F i ] = G 1CNN (L RK J K )

[0064] Among them, Y RK The recommended rehabilitation training tasks include the type of rehabilitation training task, difficulty level, duration, and frequency for different limb parts; the value of i is [1, 8], representing different limb parts: left foot, left lower limb, left upper limb, left hand, right foot, right lower limb, right upper limb, and right hand; BW i For a certain limb part; TP i Type of rehabilitation training task for a specific body part; HR i The difficulty level of rehabilitation training tasks for a specific body part; TM i Time allotted for rehabilitation training tasks in a specific area; F i Frequency of rehabilitation training tasks for a specific body part; G 1CNN For a trained convolutional neural network model; L RK J is the preprocessed feature vector of the scale; K This is the patient's basic information after pretreatment.

[0065] The convolutional neural network model consists of 3 convolutional layers and 3 pooling layers. Each convolutional layer is connected to a pooling layer. The convolutional kernel of each convolutional layer is 2*2 and the convolutional stride is 1*1. The pooling layer is a 2*2 max pooling layer.

[0066] Preferably, the patient scale data is updated every month, and rehabilitation training tasks are recommended to the patient again through the rehabilitation training task recommendation module based on limb motor function.

[0067] A rehabilitation training task recommendation module based on brain function was constructed to recommend suitable rehabilitation training tasks for patients based on their MRI brain image assessment results. Details are as follows:

[0068] A rehabilitation training task database based on MRI brain images of stroke patients was constructed, and the MRI brain images and report data in the database were preprocessed: Gaussian filtering and median filtering were used to filter and remove noise from the MRI brain images; word2vec word vectorization was used to vectorize the MRI data report results.

[0069] Data features of preprocessed MRI brain images and reports were extracted using a convolutional neural network model;

[0070] H NK =Feedforward(W CNN B CNN JJ CNN CH CNN ;HC,BG;)

[0071] Among them, H NK It is the feature vector extracted by the convolutional neural network model; Feedforward is the feedforward neural network function; W CNN It is the weight matrix of the convolutional neural network model, B CNN It is a bias parameter; JJ CNN It is a convolutional layer of the convolutional module, with a 2x2 kernel and a 2x2 stride; CH CNN This is the pooling layer of the convolutional module, with a 3*3 max pooling kernel; HC and BG are the input MRI brain image data and data report, respectively.

[0072] The extracted MRI brain images and report data features are input into another convolutional neural network model for training and model optimization. The model parameters are continuously optimized to finally obtain a brain function rehabilitation training task recommendation model.

[0073] Y NK =[BW i ,TP i HR i ,TMi ,F i ] = G 2CNN (L RK H NK J K )

[0074] The other convolutional neural network model consists of 3 convolutional layers and 3 pooling layers. Each convolutional layer is connected to a pooling layer. The first convolutional layer has a 2*2 kernel and a 2*1 stride. The second convolutional layer has a 4*2 kernel and a 3*1 stride. The first convolutional layer has an 8*4 kernel and a 4*1 stride. All pooling layers are 2*2 max pooling.

[0075] Among them, Y NK The recommended rehabilitation training tasks include the type of rehabilitation training task, difficulty level, duration, and frequency for different limb parts; the value of i is [1, 8], representing different limb parts: left foot, left lower limb, left upper limb, left hand, right foot, right lower limb, right upper limb, and right hand; BW i For a certain limb part; TP i Type of rehabilitation training task for a specific body part; HR i The difficulty level of rehabilitation training tasks for a specific body part; TM i Time allotted for rehabilitation training tasks in a specific area; F i Frequency of rehabilitation training tasks for a specific body part; G 2CNN For a trained convolutional neural network model; H NK For the characteristics of MRI brain images and report data; J K This is the patient's basic information after pretreatment.

[0076] Preferably, the patient's MRI brain images and report information are updated every two months, and rehabilitation training tasks are recommended to the patient again through the brain function-based rehabilitation training task recommendation module.

[0077] A rehabilitation training task recommendation module integrating motor function and brain function was constructed to recommend suitable rehabilitation training tasks for patients based on pre-processed scale motor function data and MRI brain image data. The details are as follows:

[0078] The aforementioned scale feature vectors and MRI data features are input into a convolutional neural network model for training and model optimization. The model parameters are continuously optimized to ultimately obtain a recommended model for rehabilitation training tasks that combine motor function and brain function.

[0079] Y NH =[BW i ,TP i HR i ,TMi ,F i ] = G 3CNN (L RK H NK J K )

[0080] The convolutional neural network model consists of 4 convolutional layers and 4 pooling layers. Each convolutional layer is connected to a pooling layer. The first convolutional layer has a 2*2 kernel and a 2*1 stride; the second convolutional layer has a 3*3 kernel and a 3*1 stride; the third convolutional layer has a 6*4 kernel and a 4*1 stride; and the fourth convolutional layer has an 8*4 kernel and a 4*1 stride. All pooling layers use 3*3 max pooling.

[0081] Among them, Y NH The recommended rehabilitation training tasks include the type of rehabilitation training task, difficulty level, duration, and frequency for different limb parts; the value of i is [1, 8], representing different limb parts: left foot, left lower limb, left upper limb, left hand, right foot, right lower limb, right upper limb, and right hand; BW i For a certain limb part; TP i Rehabilitation training task type for a specific body part (passive, assisted, active, resistance); HR i The difficulty coefficient of a rehabilitation training task for a certain body part; TM i Training time for rehabilitation training tasks in a specific area; F i Training frequency for rehabilitation training tasks in a specific area; G 3CNN For a trained convolutional neural network model; L RK H is the feature vector of the scale. NK For the characteristics of MRI brain images and report data; J K This is the patient's basic information after pretreatment.

[0082] Construct an optimal rehabilitation training task recommendation module to recommend the aforementioned rehabilitation training task Y based on the patient's limb motor function. RK Rehabilitation training task Y based on patient brain function NK Rehabilitation training task Y integrating motor function and brain function NH By performing weight fusion, the optimal rehabilitation training task ZR is obtained. XL .

[0083] ZR XL =(aY RK +bY NK +cY NH )*G WC *m

[0084] ZR XLThis is the optimal rehabilitation training task, where A, B, and C are weight matrices: A = [a1, a2, a3, ..., a8], B = [b1, b2, b3, ..., b8], C = [c1, c2, c3, ..., c8], a i b i c i The values ​​of G are all in the range of [0,1], i∈[1,8]. WC G represents the completion rate of the previous rehabilitation training task, and m is the correction coefficient. When the patient is undertaking a rehabilitation training task for the first time or the previous rehabilitation training task had a 100% match rate, G... WC The value is 1; when the patient's previous rehabilitation training task matching degree is less than 100%, G WC The value is less than 1.

[0085] Specifically, A, B, and C are weight matrices, analyzed as follows:

[0086] When three rehabilitation training tasks Y RK Y NK Y NH When the types of rehabilitation training tasks for a specific limb are inconsistent, select the rehabilitation training task that is most suitable for achieving the desired results. For example, when there are three rehabilitation training tasks Y... RK Y NK Y NH BW targeting a specific limb area i The rehabilitation training tasks are divided into passive, assisted, and active tasks. Therefore, the "passive" rehabilitation training task is selected, with a weighting coefficient 'a'. i =1, b i =c i =0.

[0087] When three rehabilitation training tasks Y RK Y NK Y NH If one of the rehabilitation training tasks for a specific limb is inconsistent with the other two, select two rehabilitation training tasks of the same type. For example: when there are three rehabilitation training tasks Y... RK Y NK Y NH BW targeting a specific limb area i The rehabilitation training tasks are: passive, assisted, and assisted. Therefore, the rehabilitation training task with the "assisted" component is selected, with a weighting coefficient 'a'. i =0, b i =c i =0.5.

[0088] When three rehabilitation training tasks Y RK Y NK Y NHWhen the rehabilitation training tasks for a specific limb are of the same type, then select three rehabilitation training tasks of the same type. For example: when the three rehabilitation training tasks Y... RK Y NK Y NH BW targeting a specific limb area i If all rehabilitation training tasks are "passive," then select the "passive" rehabilitation training task and calculate the Y value for the three rehabilitation training tasks. RK Y NK Y NH BW targeting a specific limb area i The similarity of the rehabilitation training tasks. The weighting coefficients are calculated as follows:

[0089]

[0090]

[0091]

[0092] Wherein, ρ(Y) RK-BWi Y NK-BWi ) for rehabilitation training task Y RK and Y NK Similarity, ρ(Y) RK-BWi Y NH-BWi ) for rehabilitation training task Y RK and Y NH Similarity, ρ(Y) NK-BWi Y NH-BWi ) for rehabilitation training task Y NH and Y NK The similarity.

[0093]

[0094]

[0095]

[0096] Among them, Y RK-HRi Y NK-HRi Y NH-HRi These are three rehabilitation training tasks Y RK Y NK Y NH The difficulty level, Y RK-TMi Y NK-TMi Y NH-TMi These are three rehabilitation training tasks Y RK Y NK Y NH Training time, Y RK-Fi YNK-Fi Y NH-Fi These are three rehabilitation training tasks Y RK Y NK Y NH The training frequency.

[0097] Furthermore, when all a are calculated... i b i c i By calculating the values, we can obtain A, B, and C as weight matrices.

[0098] The rehabilitation training module includes a virtual reality module, an information acquisition module, an information analysis module, and a multi-physical stimulation modulation module (including a transcranial magnetic stimulation module, a peripheral electrical stimulation module, an auditory stimulation module, etc.).

[0099] The virtual reality module is used to display the rehabilitation training tasks recommended by the rehabilitation training task recommendation module on the display screen, providing VR-based rehabilitation training tasks.

[0100] The information acquisition module includes: a near-infrared brain function device for acquiring near-infrared brain blood oxygenation signals of the patient during rehabilitation training; a surface electromyography device for acquiring electromyography signals of the patient during rehabilitation training; and an eye tracker for acquiring electrooculography signals of the patient during rehabilitation training.

[0101] The information analysis module is used to perform filtering preprocessing and feature extraction analysis on the near-infrared cerebral oxygenation signals, electromyography signals, and eye movement signals acquired by the information acquisition module, and to calculate the patient's brain activation level, brain-muscle coherence, and attention span. Specifically:

[0102] The information analysis module performs wavelet transform and complex transform on near-infrared brain blood oxygenation signals to analyze and calculate brain activation, which reflects the activity level of brain function.

[0103]

[0104] Where WO represents brain activation level, J represents the number of functionally connected pathways in the healthy brain region, and WA... i The amplitude of the cerebral oxygenation signal represents the number of functional connectivity channels present in the healthy brain region; H represents the number of functional connectivity channels present in the affected brain region; WA k The brain blood oxygenation signal amplitude of the functional connectivity channel exists in the affected brain region.

[0105] The information analysis module calculates the coherence of cerebral blood oxygenation signal and electromyography signal based on the power spectrum calculation method. This correlation reflects the correlation between the two signals in the spectrum.

[0106]

[0107] Where MP represents the coherence between cerebral oxygenation signals and electromyographic signals, and CP represents the correlation coefficient between these signals. XJY (f) shows the cross-power spectrum of the contralateral cerebral oxygenation signal and electromyographic signal, CP XHY (f) shows the cross-power spectrum of cerebral oxygenation signal and electromyographic signal on the affected side, CP XJ (f) is the autopower spectrum of the contralateral cerebral oxygenation signal, CP XH (f) is the autopower spectrum of cerebral oxygenation signal on the affected side, CP Y (f) represents the autopower spectrum of the electromyographic signal, XJ represents the oxygenation signal of the healthy side of the brain, XH represents the oxygenation signal of the affected side of the brain, and Y represents the electromyographic signal.

[0108] The information analysis module performs time-domain and frequency-domain feature analysis on the electrooculogram (EOG) signal after removing interference noise. The time-domain smoothness feature of the EOG signal is defined as follows:

[0109]

[0110] Among them, Z uv Z represents the smoothness of the electrooculogram signal. iu This refers to the electrooculographic signal from the U electrode (the electrode on the upper eyelid). Z represents the mean value of the electrooculogram signal from the u-electrode. iv This refers to the electrooculographic signal from the V electrode (the electrode on the lower eyelid). The mean value of the electrooculographic signal at the V electrode.

[0111] The frequency domain characteristics of the average power of the electrooculogram (EOG) signal are defined as follows:

[0112]

[0113] Among them, Z PF f is the average power frequency of the electrooculogram (EOG), f is the frequency of the EOG signal, and f1 and f2 represent the frequency range of the EOG signal.

[0114] Based on the aforementioned time-domain and frequency-domain features, the information analysis module establishes the following attention model indicators:

[0115] Concentration F = A × Z uv +B×Z PF

[0116] Among them, Z uv Z represents the smoothness of the electrooculogram signal. PF Let A and B be the average power frequency of the electrooculogram (EOG), and their respective weights. Experiments have shown that the focus of stroke patients is most strongly correlated with the smoothness of the EOG signal and the average power frequency of the EOG, with the smoothness of the EOG signal accounting for a larger proportion. Therefore, the weights A and B are set as follows: 0.5. <A<1,0<B<0.5,A+B=1。

[0117] The multi-physical stimulation modulation module is used to regulate transcranial magnetic stimulation, peripheral electrical stimulation, auditory stimulation, and other methods based on data such as the patient's brain activation level, brain-muscle coherence, and attention span analyzed by the information analysis module. The specific modulation scheme is as follows:

[0118] When the patient's brain activation level is lower than the set threshold, transcranial magnetic stimulation is used for brain neuromodulation; when the patient's brain-muscle correlation is lower than the set threshold, peripheral electrical stimulation is used for neuromuscular modulation; when the patient's concentration is lower than the set threshold, auditory stimulation is used to regulate concentration, such as playing rhythmic music.

[0119] Furthermore, transcranial magnetic stimulation, peripheral electrical stimulation, and auditory stimulation can be performed individually, in pairs, or simultaneously, depending on the patient's brain activation level, brain-muscle coherence, concentration, and other factors.

[0120] Preferably, when the analysis finds that the patient's brain activation and brain-muscle coherence are lower than their respective set thresholds, transcranial magnetic stimulation and peripheral electrical stimulation can be performed simultaneously; when the analysis finds that brain activation and attention are lower than their respective set thresholds, transcranial magnetic stimulation and auditory stimulation can be performed simultaneously; when the analysis finds that brain-muscle coherence and attention are lower than their respective set thresholds, peripheral electrical stimulation and auditory stimulation can be performed simultaneously; when the analysis finds that brain activation, brain-muscle coherence, and attention are lower than their respective set thresholds, transcranial magnetic stimulation, peripheral electrical stimulation, and auditory stimulation can be performed simultaneously.

[0121] Preferably, the stimulation site, frequency, duration, and amplitude of specific transcranial magnetic stimulation (TMS) can be precisely controlled by analyzing data indicators such as brain lateralization, functional connectivity, and effect connectivity obtained from brain oxygenation signals. The specific precise control scheme is as follows: A TMS parameter adjustment database is established by continuously collecting data on patient brain activation, lateralization, functional connectivity, basic clinical information, and optimal TMS stimulation site, frequency, duration, and amplitude. An adaptive TMS parameter model based on deep learning is then established, continuously trained and tested to optimize the neural network model parameters, forming an adaptive TMS parameter model that intelligently recommends personalized adaptive TMS parameters and makes fine adjustments based on the patient's real-time rehabilitation training, reducing the number of manual parameter adjustments.

[0122] Preferably, data indicators such as muscle strength and muscle fatigue can be obtained through further calculation and analysis of cerebral blood oxygen signals and electromyography signals, so as to regulate parameters such as stimulation frequency, stimulation duration and stimulation pulse of specific peripheral electrical stimulation, thereby achieving precise regulation. The specific precise regulation scheme is as follows: a database for adjusting peripheral electrical stimulation parameters is established by continuously collecting data indicators such as patient's brain-muscle coherence, muscle strength, muscle fatigue, optimal electrical stimulation frequency, stimulation duration and stimulation pulse, an adaptive peripheral electrical stimulation parameter model based on deep learning is established, continuous training and testing are performed, neural network model parameters are optimized, and the adaptive peripheral electrical stimulation parameter model is formed, which can intelligently recommend personalized adaptive peripheral electrical stimulation parameters, and perform further fine adjustment according to the patient's real-time rehabilitation training situation, reducing the number of manual parameter adjustments.

[0123] The rehabilitation training task completion evaluation module is configured to determine the patient's rehabilitation training task matching degree according to the number of multi-physical stimulation regulaitons, the concentration duration and the completion degree of the rehabilitation training task in the rehabilitation training module.

[0124] Rehabilitation training task matching degree:

[0125]

[0126] wherein Gc is the rehabilitation training task matching degree, T is the time for completing one rehabilitation training task, t c is the total time of physical stimulation in one rehabilitation training task, Kc is the completion degree of the rehabilitation training task, Z C is the duration for which the patient's concentration reaches the standard in one rehabilitation training task, and C1, C2, C3 are weight coefficients. Experiments show that the completion degree of the rehabilitation training task accounts for a relatively large proportion in the rehabilitation training task matching degree, therefore the values of the weight coefficients C1, C2, C3 are set as follows: 0.1 < C1 < 0.4, 0.6 < C2 ≤ 1, 0.1 < C3 < 0.4, and C1 + C2 + C3 = 1.

[0127] Compared with the prior art, the beneficial effects of the present invention are:

[0128] (1) Patient rehabilitation training tasks are recommended comprehensively through three aspects of indicators, namely limb motor function data indicators, brain function data indicators and fused motor function and brain function data indicators of stroke patients, and the adaptation degree of multi-source information such as the patient's limb, brain, and brain-limb coordination to rehabilitation training tasks is fully considered, so as to realize accurate recommendation of rehabilitation training tasks for patients.

[0129] (2) In combination with the brain activation degree, brain-muscle coherence, concentration and the like during the patient's rehabilitation training, multiple physical stimulation modes including transcranial magnetic stimulation, peripheral electrical stimulation and auditory stimulation are used for regulation, so as to promote the collaborative optimization and real-time feedback of brain-limb and physical stimulation data.

[0130] (3) Using this system, patients can be provided with personalized and adaptive rehabilitation training task recommendations, and multiple physical stimulation rehabilitation training can be used to maximize the benefit effect and improve the efficiency and effectiveness of rehabilitation training.

[0131] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A stroke rehabilitation training system, comprising a rehabilitation training task recommendation module, a rehabilitation training module, and a rehabilitation training task completion evaluation module, characterized in that: The rehabilitation training task recommendation module is used to recommend adaptive rehabilitation training tasks to stroke patients based on their rehabilitation assessment results; the rehabilitation training module is used to conduct rehabilitation training for stroke patients based on the rehabilitation training tasks recommended by the rehabilitation training task recommendation module. The rehabilitation training task completion evaluation module is used to evaluate the patient's rehabilitation training completion rate and provide feedback to the rehabilitation training task recommendation module; the rehabilitation training task recommendation module includes a rehabilitation training task recommendation module that integrates motor function and brain function; in the stroke rehabilitation training system, the rehabilitation training task recommendation module recommends suitable rehabilitation training tasks for the patient based on the patient's MRI brain image assessment results in the following manner: A database of rehabilitation training tasks based on MRI brain images of stroke patients was constructed, and the MRI brain images and report data in the database were preprocessed. Data features of preprocessed MRI brain images and reports were extracted using a convolutional neural network model; Among them, H NK It is the feature vector extracted by the convolutional neural network model; Feedforward is the feedforward neural network function; W CNN It is the weight matrix of the convolutional neural network model, B CNN It is a bias parameter; JJ CNN It is a convolutional layer of the convolutional module, with a 2x2 kernel and a 2x2 stride; CH CNN This is the pooling layer of the convolutional module, with a 3*3 max pooling kernel; HC and BG are the input MRI brain image data and data report, respectively. The extracted features from MRI brain images and report data are input into another convolutional neural network model for training and optimization, ultimately resulting in a brain function rehabilitation training task recommendation model. Among them, Y NK The recommended rehabilitation training tasks include the type of rehabilitation training task, difficulty level, duration, and frequency for different limb parts; the value of i is [1, 8], representing different limb parts: left foot, left lower limb, left upper limb, left hand, right foot, right lower limb, right upper limb, and right hand; BW i For a certain limb part; TP i Type of rehabilitation training task for a specific body part; HR i The difficulty level of rehabilitation training tasks for a specific body part; TM i Time allotted for rehabilitation training tasks in a specific area; F i Frequency of rehabilitation training tasks for a specific body part; G 2CNN For a trained convolutional neural network model; H NK It is the feature vector extracted by the convolutional neural network model; J K Basic patient information after pretreatment; L RK This is the feature vector of the scale.

2. The stroke rehabilitation training system according to claim 1, characterized in that... The rehabilitation training task completion evaluation module is used to determine the patient's rehabilitation training task matching degree based on the number of times the multi-physical stimulation is modulated, the duration of focus, and the completion degree of the rehabilitation training task in the rehabilitation training module.

3. The stroke rehabilitation training system according to claim 1, characterized in that: The rehabilitation training task recommendation module recommends suitable rehabilitation training tasks for patients based on their MRI brain image assessment results, as follows: A database of rehabilitation training tasks based on MRI brain images of stroke patients was constructed, and the MRI brain images and report data in the database were preprocessed. We used a convolutional neural network model to extract data features from preprocessed MRI brain images and reports.

Images