Design methods, devices and equipment for lower limb rehabilitation programs suitable for different individuals

Abstract

The present invention discloses a method, device and equipment for designing a lower limb rehabilitation program suitable for different individuals, which collects multi-source information of subjects and performs preprocessing, extracts motion-related factors in the multi-source information based on a symmetric bilinear model, and extracts individual-irrelevant features; performs feature evaluation based on the Davidson-Botting index and the scattering distribution index, and selects target features; inputs the target features into a convolutional neural network for training, obtains a lower limb rehabilitation program design model, and obtains lower limb rehabilitation programs for different subjects based on the lower limb rehabilitation program design model. The present invention utilizes information decomposition and feature evaluation to obtain robust features that are independent of individuals, thereby improving the versatility of the lower limb rehabilitation program design model; decomposes multi-source information into individual-related factors and motion-related factors, and the motion-related factors are only related to actual rehabilitation training, thereby improving the versatility of the model; and discards redundant information through feature evaluation and feature selection, thereby improving the speed and efficiency of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of rehabilitation program design, and in particular to a method, device and equipment for designing a lower limb rehabilitation program suitable for different individuals. Background Art

[0002] The impaired motor function of the lower limbs of patients or the elderly has brought great inconvenience to their daily lives, and rehabilitation therapy can effectively help patients or the elderly restore their motor function and improve their quality of life. Therefore, the design of a lower limb rehabilitation program is particularly important.

[0003] At present, most lower limb rehabilitation programs are designed for a single individual, that is, each time a new subject is faced, the degree of rehabilitation must be assessed and then a rehabilitation program is formed. There are also some databases and models that can directly select the most appropriate rehabilitation program based on individual information. However, there are many differences between different individuals, especially different lower limb movement methods and kinematic characteristics, which limit the universality of lower limb rehabilitation program design models among different individuals.

[0004] Existing similar published patents:

[0005] A method and system for rehabilitation training management (CN 114242203 A), the steps of the patent are:

[0006] Step 1: Determine the attribute information and initial physiological characteristic information of the target user;

[0007] Step 2: input the attribute information and the initial physiological characteristic information into a pre-trained rehabilitation training model to obtain a rehabilitation training plan for the target user within a training period, wherein the rehabilitation training plan includes at least one training monitoring item;

[0008] Step 3: determining updated physiological characteristic information of the target user under at least one training monitoring item in a previous training cycle;

[0009] Step 4: Input the attribute information and updated physiological characteristic information of the target user into the rehabilitation training model to obtain a rehabilitation training plan for the target user in a subsequent training cycle.

[0010] A method for recommending postoperative rehabilitation programs for orthopedics based on deep learning (CN 110970109A). The patent steps are:

[0011] Step 1: Collect orthopedic postoperative rehabilitation treatment plans, treatment plans, and actual treatment data corresponding to the treatment plans to form a data set;

[0012] Step 2: Classify the treatment plans in the dataset into multiple subcategories based on wound site and wound evaluation;

[0013] Step 3: Take the patient's trauma site and trauma evaluation in the treatment plan of the training set as input, and the treatment plan as output, learn through the neural network, extract features, and form a plan recommendation model;

[0014] Step 4: Input the patient's injured area and trauma evaluation into the treatment plan recommendation model, and the treatment plan recommendation model outputs the recommended treatment plan.

[0015] The existing technology does not take into account the differences between different individuals, and the design method has poor versatility; it does not evaluate and analyze the extracted information, and there is a lot of redundant information.

[0016] Therefore, how to extract robust features with low individual correlation in rehabilitation training scenarios so that the model for lower limb rehabilitation program design is suitable for most individuals is an urgent problem to be solved. Summary of the invention

[0017] The present invention aims to solve the technical problems of ignoring individual differences, poor universality and excessive redundant information in the prior art. The technical solution adopted by the present invention is as follows: a method, device and equipment for designing lower limb rehabilitation programs suitable for different individuals are provided. The technical solution uses information decomposition and feature evaluation to obtain robust features that are independent of individuals, thereby improving the universality of the lower limb rehabilitation program design model.

[0018] According to a first aspect of the present invention, a method for designing a lower limb rehabilitation program suitable for different individuals comprises the following steps:

[0019] Collecting and preprocessing multi-source information of the subject, the multi-source information including: surface electromyographic signal, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension;

[0020] Extract motion-related factors from multi-source information based on a symmetric bilinear model, and extract individual-irrelevant features from the motion-related factors;

[0021] Perform feature evaluation on multiple individual irrelevant features and select target features based on the evaluation results;

[0022] The target features are input into the convolutional neural network for training. After the training is completed, a lower limb rehabilitation program design model is obtained. The lower limb rehabilitation programs of different subjects are obtained based on the lower limb rehabilitation program design model.

[0023] Furthermore, the step of extracting motion-related factors from multi-source information based on a symmetric bilinear model and extracting individual-irrelevant features from the motion-related factors includes:

[0024] Assume that multi-source information D = {D1, D2, ..., D nThere are n types of information in total, each type of information D p ∈R K (p=1,2,…N) has K dimensions, and the K value for each type of information is different;

[0025] The p-th type of information is modeled using a symmetric bilinear model, and the model of the p-th type of information is as follows:

[0026]

[0027] Where: X∈R I is the motion-related factor; Y∈R J is the individual correlation factor; X and Y are both orthogonal matrices, that is, XX T =I(YY T =I), I is the unit matrix; W p ∈R I×J (X,Y)→D p The mapping matrix of x i is the column vector of the i-th motion-related factor; y i is the column vector of the jth individual correlation factor; w ij W p Elements in

[0028] Let S subjects at different rehabilitation stages perform the cth (c = 1, 2, ... C) rehabilitation training program and collect data during the training;

[0029] Calculate the objective function and obtain the X, W corresponding to the minimum value of the objective function E p ,Y;

[0030] Have new subjects complete specific training according to instructions and obtain a set of and The value of is the multi-source information collected under this specific training, The movement-related factors for new subjects under this specific training;

[0031] according to and W p Calculate the individual correlation factor Y of the new subject new ; Afterwards, when performing regular training, obtain multi-source data under regular training according to W p and Y new Calculate the exercise-related factor X of the new subject under the conventional training new After obtaining the motion-related factors of each piece of information, new Feature extraction is performed to obtain multiple individual irrelevant features.

[0032] Furthermore, the calculation formula of the objective function is as follows:

[0033]

[0034] Where: X c Y is the exercise-related factor associated with the cth rehabilitation program; s is the individual correlation factor associated with the sth subject.

[0035] Furthermore, the X,W corresponding to the minimum value of the objective function E is obtained. p ,Y steps include:

[0036] Transform the model formula of the p-th type of information to obtain the following formula:

[0037] [D p Y] T =W p T X T

[0038] [D p T X] T =W p Y T

[0039] The first estimate of Y is made through singular value decomposition. The singular value decomposition formula is as follows:

[0040] A=UΣV T

[0041] Where: A is the matrix to be decomposed; U is the left singular matrix; Σ is a matrix in which all elements except the diagonal are 0, and the elements on the diagonal are singular values; V is the right singular matrix;

[0042] Right D p Perform singular value decomposition and take the decomposed V T The first J rows of T The first estimate of

[0043] According to the first estimate right Perform singular value decomposition and take the decomposed V T The first I rows are X T The first estimate of

[0044] According to the first estimate Will Perform singular value decomposition and take the decomposed V T The first J rows of T The new estimate of

[0045] and Each singular value decomposition is completed as one iteration, and after multiple iterations until convergence, the final and After transposition, we get X and Y that minimize the objective function, and W is calculated by the following formula p :

[0046] W p =X T D p Y.

[0047] Furthermore, the individual irrelevant features are features extracted from the motion-related factor X, including: an absolute value mean in the time domain, an autoregressive coefficient, a median frequency in the frequency domain, and a wavelet transform in the time-frequency domain.

[0048] Preferably, the step of performing feature evaluation on a plurality of individual irrelevant features and selecting a target feature according to the evaluation results comprises:

[0049] The Davidson-Boulding Index (DBI) and the Scattering Distribution Index (SCAT) were used for feature evaluation; the definitions of DBI and SCAT are as follows:

[0050]

[0051] The Davidson-Boulding Index (DBI) and the Scattering Distribution Index (SCAT) were used for feature evaluation; the definitions of DBI and SCAT are as follows:

[0052]

[0053] Where: N is the type of rehabilitation training program; is the average Euclidean distance from the i-th feature to its feature center; is the average Euclidean distance from the j-th feature to its feature center; ||ω i -ω j ||2 is the Euclidean distance between the feature centers of the i-th and j-th categories;

[0054]

[0055]

[0056] SCAT=Trace(S w ) / Trace(S B )

[0057] Where: S w is the sum of the covariance matrices of all features; S Bis the covariance matrix between different features; x is the eigenvalue of the i-th feature; μ i is the mean of the i-th feature; m mean is the mean of all features; Trace(·) is the trace of the matrix;

[0058] The STN index is calculated based on the Davidson-Boulding index DBI and the scattering distribution index SCAT. The calculation formula is as follows:

[0059]

[0060] Where: DBI i and SCAT i The DBI index and SCAT index represent the individual-independent characteristics of the i-th rehabilitation program, and They represent the average values ​​of DBI index and SCAT index respectively; STN is the robust performance index;

[0061] The features whose STN index meets the preset conditions are taken as target features.

[0062] Preferably, the convolutional neural network is a multi-input multi-output convolutional neural network, and the shape of the convolution kernel is n×m×h×w, where m is the number of input feature types, n is the number of content items in the rehabilitation training program, and h and w are the length and width.

[0063] According to a second aspect of the present invention, a device for designing lower limb rehabilitation programs suitable for different individuals includes the following modules:

[0064] An acquisition and preprocessing module is used to acquire and preprocess the multi-source information of the subject, wherein the multi-source information includes: surface electromyography signal, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension;

[0065] A motion-related factor extraction module is used to extract motion-related factors from multi-source information based on a symmetric bilinear model, and to extract individual-irrelevant features from the motion-related factors;

[0066] The feature evaluation and selection module is used to evaluate multiple individual irrelevant features and select target features based on the evaluation results;

[0067] The network training and program acquisition module is used to input the target features into the convolutional neural network for training. After the training is completed, a lower limb rehabilitation program design model is obtained, and lower limb rehabilitation programs for different subjects are obtained based on the lower limb rehabilitation program design model.

[0068] According to a third aspect of the present invention, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for designing a lower limb rehabilitation program suitable for different individuals are implemented.

[0069] According to a fourth aspect of the present invention, a storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the method for designing a lower limb rehabilitation program suitable for different individuals are implemented.

[0070] The technical solution provided by the present invention has the following beneficial effects:

[0071] The acquired multi-source information is decomposed into individual-related factors and motion-related factors. The motion-related factors are only related to actual rehabilitation training, which improves the versatility of the lower limb rehabilitation program design model. The extracted features are evaluated, and the features are selected based on the evaluation results, discarding redundant information to improve the speed and efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0073] Figure 1 It is an overall flow chart of a method for designing a lower limb rehabilitation program applicable to different individuals in an embodiment of the present invention;

[0074] Figure 2 X, W in the embodiment of the present invention p ,Y solution flow chart;

[0075] Figure 3 This is a general structural diagram of a lower limb rehabilitation program design device suitable for different individuals in an embodiment of the present invention;

[0076] Figure 4 It is a schematic diagram of the physical structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0078] The present invention is applicable to the lower limb rehabilitation scenario, and uses multi-source information and neural networks to design rehabilitation programs. The overall implementation process of the system is as follows: Figure 1 As shown, the specific steps include:

[0079] S1: Collecting and preprocessing multi-source information of the subject, wherein the multi-source information includes: surface electromyography signal, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension;

[0080] Specifically, in order to obtain the best lower limb rehabilitation training program, the multi-source information that needs to be obtained includes: surface electromyography signals, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength, muscle tension, etc.

[0081] Some information cannot be used directly after being directly obtained through special equipment, and some information cannot even be directly obtained, so certain preprocessing is required. For example, the surface electromyography signal is weak and easily affected by various noises, making it difficult or even impossible to extract effective information from it. The collected original surface electromyography signal must be denoised; the measurement of indicators such as movement speed and joint range of motion requires some specific sensors, and the data obtained also needs to be converted into data form.

[0082] S2: Extract motion-related factors from multi-source information based on a symmetric bilinear model, and extract individual-independent features from the motion-related factors;

[0083] The multi-source information collected and pre-processed in step S1 is decomposed into individual-related factors and exercise-related factors. The individual-related factors are related to the basic condition of the individual, and the exercise-related factors are related to the training content of the rehabilitation training. The exercise-related factors of the same rehabilitation training in different individuals are the same.

[0084] Assume that multi-source information D = {D1, D2, ..., D n There are n types of information in total, each type of information D p ∈R K (p=1,2,…N) has K dimensions, and the K value of each type of information is different. For example, D1 represents the surface electromyography signal, and the K value represents the number of detection channels; D2 represents the movement speed, and the K value represents the number of selected joints.

[0085] The p-th type of information is modeled using a symmetric bilinear model:

[0086]

[0087] Where: X∈R I represents the motion-related factor; Y∈R J represents individual correlation factor; W p ∈R I×J Represents (X,Y)→D p The mapping matrix of x i represents the column vector of the i-th motion-related factor; y i represents the column vector of the jth individual correlation factor; w ij W p The elements in .

[0088] In formula 1, both X and Y are orthogonal matrices, that is, XXT =I(YY T =I), I is the unit matrix.

[0089] To determine formula 1, we need to find S subjects at different stages of rehabilitation, let them perform the cth (c = 1, 2, ... C) rehabilitation training program, and collect data during training. The objective function can be set as:

[0090]

[0091] Where:

[0092] X c represents the exercise-related factor associated with the cth rehabilitation program; Y s represents the individual correlation factor associated with the sth subject.

[0093] The purpose of formula 2 is to find the appropriate X for all s, c, p c ,W p ,Y s , so that the value of the objective function E is minimized. In order to derive the objective function E, formula 1 can be transformed to a certain extent:

[0094] [D p Y] T =W p T X T (Formula 3)

[0095] [D p T X] T =W p Y T (Formula 4)

[0096] First, we can make a first estimate of Y through singular value decomposition. The singular value decomposition formula is as follows:

[0097] A=UΣV T (Formula 5)

[0098] In the formula: A represents the matrix to be decomposed; U represents the left singular matrix; Σ is a matrix in which all elements except the diagonal are 0, and the elements on the diagonal are singular values; V represents the right singular matrix.

[0099] Solving for X,W p , the specific steps of Y are as follows Figure 2 First, D p Perform singular value decomposition and take the decomposed V T The first J rows of T The first estimate of According to formula 3 and the estimated value Will Perform singular value decomposition and take the decomposed V T The first I rows are X T The first estimate of According to formula 4 and the estimated value Will Perform singular value decomposition and take the decomposed V T The first J rows of T The new estimate of The last two processes are one iteration. After repeated iterations and convergence, the final and After transposition, we can get the X and Y we are looking for. Finally, we can get the W we are looking for through formula 6. p :

[0100] W p =X T D p Y (Formula 6)

[0101] Although the mapping matrix W is obtained p , but for new subjects, both X and Y are unknown. Therefore, before using the model directly, let the new subjects complete specific training according to the instructions. Specific training means letting the new subjects perform the most commonly used and less difficult rehabilitation training movements. Since the movements are relatively simple and the relevant data are sufficient, it is relatively easy to determine the movement-related factors of the specific movements. In the scenario of specific training, a set of and The value of is the multi-source information collected by the new subject under this specific training, is the exercise-related factor for the new subject under this specific training. and W p Substituting them into Formula 4, we can obtain the individual correlation factor Y of the new subject. new After obtaining the individual correlation factors of the new subjects, conventional training is carried out. Conventional training refers to some rehabilitation training that the new subjects have recently carried out. Obtain multi-source data under conventional training Will W p and Y new Substituting them into Formula 3, we can obtain the exercise-related factor X of the new subject under the conventional training: new After obtaining the motion-related factors of each piece of information, new Perform feature extraction. It can extract multiple individual-independent features at the same time, such as the absolute value mean and autoregressive coefficient in the time domain, the median frequency in the frequency domain, and the wavelet transform in the time-frequency domain. According to different features, a reasonable lower limb rehabilitation program can be selected, and the influence of individual differences on the results can be avoided.

[0102] S3: Perform feature evaluation on multiple individual-independent features and select target features based on the evaluation results;

[0103] Not all individual-independent features extracted in step S2 are suitable for obtaining the final rehabilitation plan, so the extracted features need to be evaluated.

[0104] In this embodiment, the Davidson-Boulding Index (DBI Index) and the Scattering Distribution Index (SCAT Index) are preferably used to evaluate the features. The specific definitions of the DBI Index and the SCAT Index are as follows:

[0105]

[0106] Where:

[0107] N represents the type of rehabilitation training program; Represents the average Euclidean distance from the i-th feature to its feature center; is the average Euclidean distance from the j-th feature to its feature center; ||ω i -ω j ||2 represents the Euclidean distance between the feature centers of the i-th and j-th categories.

[0108]

[0109] Where: S w represents the sum of the covariance matrices of all features; S B represents the covariance matrix between different features; x represents the eigenvalue of the i-th feature; μ i represents the mean of the i-th category feature; m mean is the mean of all features; Trace(·) represents the trace of the matrix.

[0110] The DBI index represents the degree of overlap between adjacent features. A lower DBI index represents better feature separability. The SCAT index uses the covariance matrix between different features and the trace relationship of the covariance matrix to represent the overall situation of the sample space. A lower SCAT index represents a better distribution of samples. In order to better calculate the robust performance index (STN) of the feature, it is necessary to obtain the standard deviation of the DBI index and the SCAT index:

[0111]

[0112] Where: DBI i and SCAT i The DBI index and SCAT index represent the individual-independent characteristics of the i-th rehabilitation program, and They represent the average values ​​of DBI index and SCAT index respectively; STN is the robust performance index.

[0113] The features whose STN index meets the preset conditions are selected as the target features. The smaller the STN index is, the better the robust performance of the representative feature is. The preset condition is generally to select several features with the smallest STN index or features with STN index less than 1.

[0114] In other embodiments, in order to obtain the target feature, the Relief feature selection algorithm can also be selected. The basic principle is to randomly select a sample in a certain category (that is, a feature in a certain rehabilitation training mode), and search for the same number of samples with the closest distance from all samples. The weight of the feature is updated according to the principle that the value of the same feature in the same type of samples should be as consistent as possible and the value in different types of samples should be as inconsistent as possible, so as to obtain the optimal subset of features with higher weights.

[0115] S4: Input the target features into the convolutional neural network for training to obtain a lower limb rehabilitation program design model, and obtain lower limb rehabilitation programs for different subjects based on the lower limb rehabilitation program design model.

[0116] The lower limb rehabilitation program generally includes training content, number of trainings, training time, training cycle, etc., wherein the training content may be muscle training, joint range of motion training, motor function training, etc. Since the input of the neural network is a variety of features, and the final output rehabilitation program includes multiple contents, a multi-input and multi-output convolutional neural network is used to obtain the rehabilitation program. Assuming that the input feature types are m and the content of the rehabilitation training program is n, the shape of the convolution kernel of the convolutional neural network is n×m×h×w, where h and w are length and width. The target feature obtained in step S3 is input into the convolutional neural network for training. The specific training process includes: the convolutional neural network first initializes the weights, the input target feature is passed through the convolution layer, the pooling layer, and the fully connected layer to obtain the output value, the error between the output value of the network and the target value is calculated, the weight is updated according to the error, and the above steps are repeated using the updated weight until the error is equal to or less than the set expected value. The training is terminated, and the lower limb rehabilitation program design model is obtained after the training is completed.

[0117] For new subjects, they only need to take a simple test to determine their own individual related factors. The multi-source information collected during rehabilitation training is input into the lower limb rehabilitation program design model to obtain a rehabilitation program suitable for the new subjects.

[0118] In other embodiments, other types of convolutional neural network (CNN) architectures can also be selected for model training, such as multi-path-based CNN, feature map-based CNN, etc. In addition to CNN, deep learning network structures also include deep neural networks (DNN), recurrent neural networks (RNN), etc., which can be used as replaceable networks for the lower limb rehabilitation program design method described in the present invention.

[0119] The following is a description of a lower limb rehabilitation program design device suitable for different individuals provided by the present invention. The lower limb rehabilitation program design device described below and the lower limb rehabilitation program design method described above can be referenced to each other.

[0120] like Figure 3 As shown, a lower limb rehabilitation program design device suitable for different individuals includes the following modules:

[0121] The acquisition and preprocessing module 1 is used to acquire and preprocess the multi-source information of the subject, wherein the multi-source information includes: surface electromyography signal, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension;

[0122] A motion-related factor extraction module 2 is used to extract motion-related factors from multi-source information based on a symmetric bilinear model, and to extract individual-irrelevant features from the motion-related factors;

[0123] The feature evaluation and selection module 3 is used to evaluate multiple individual irrelevant features and select target features according to the evaluation results;

[0124] The network training and program acquisition module 4 is used to input the target features into the convolutional neural network for training. After the training is completed, a lower limb rehabilitation program design model is obtained, and lower limb rehabilitation programs for different subjects are obtained based on the lower limb rehabilitation program design model.

[0125] The various modules of the lower limb rehabilitation program design device suitable for different individuals correspond one-to-one to the various steps of the lower limb rehabilitation program design method suitable for different individuals, and are respectively used to implement the various steps of the above method, which will not be repeated here.

[0126] like Figure 4As shown, a schematic diagram of the physical structure of an electronic device is exemplified, and the electronic device may include: a processor (processor) 610, a communication interface (Communications Interface) 620, a memory (memory) 630 and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call the logic instructions in the memory 630 to execute the steps of the lower limb rehabilitation program design method applicable to different individuals, specifically including: collecting and preprocessing the multi-source information of the subject, the multi-source information includes: surface electromyographic signal, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension; extracting the movement-related factors in the multi-source information based on the symmetric bilinear model, and extracting individual-independent features from the movement-related factors; evaluating the characteristics of multiple individual-independent features, and selecting the target features according to the evaluation results; inputting the target features into the convolutional neural network for training, and obtaining the lower limb rehabilitation program design model after the training is completed, and obtaining the lower limb rehabilitation programs of different subjects based on the lower limb rehabilitation program design model.

[0127] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random15 Access Memory), disk or optical disk and other media that can store program codes.

[0128] On the other hand, an embodiment of the present invention further provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for designing a lower limb rehabilitation program suitable for different individuals, specifically including: collecting and preprocessing multi-source information of the subject, the multi-source information including: surface electromyography signals, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension; extracting motion-related factors from the multi-source information based on a symmetric bilinear model, and extracting individual-irrelevant features from the motion-related factors; performing feature evaluation on multiple individual-irrelevant features, and selecting target features based on the evaluation results; inputting the target features into a convolutional neural network for training, and obtaining a lower limb rehabilitation program design model after the training is completed, and obtaining lower limb rehabilitation programs for different subjects based on the lower limb rehabilitation program design model.

[0129] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0130] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order and these words may be interpreted as identifiers.

[0131] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for designing a lower limb rehabilitation program suitable for different individuals, characterized in that: The following steps are involved: Collecting and preprocessing multi-source information of the subject, the multi-source information including: surface electromyographic signal, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension; Extract motion-related factors from multi-source information based on a symmetric bilinear model, and extract individual-irrelevant features from the motion-related factors; Perform feature evaluation on multiple individual irrelevant features and select target features based on the evaluation results; The target features are input into the convolutional neural network for training. After the training is completed, a lower limb rehabilitation program design model is obtained. The lower limb rehabilitation programs of different subjects are obtained based on the lower limb rehabilitation program design model. The step of extracting motion-related factors from multi-source information based on a symmetric bilinear model and extracting individual-irrelevant features from the motion-related factors comprises: Assume that multi-source information D = {D1, D2, ..., D n There are n types of information in total, each type of information D p ∈R K (p=1,2,…N) has K dimensions, and the K value for each type of information is different; The p-th type of information is modeled using a symmetric bilinear model, and the model of the p-th type of information is as follows: Where: X∈R I is the motion-related factor; Y∈R J is the individual correlation factor; X and Y are both orthogonal matrices, that is, XX T =I(YY T =I), I is the unit matrix; W p ∈R I×J (X,Y)→D p The mapping matrix of x i is the column vector of the i-th motion-related factor; y i is the column vector of the jth individual correlation factor; w ij W p Elements in Let S subjects at different rehabilitation stages perform the cth (c = 1, 2, ... C) rehabilitation training program and collect data during the training; Calculate the objective function and obtain the X, W corresponding to the minimum value of the objective function E p ,Y; Have new subjects complete specific training according to instructions and obtain a set of and The value of is the multi-source information collected under this specific training, The movement-related factors for new subjects under this specific training; according to and W p Calculate the individual correlation factor Y of the new subject new ; Afterwards, during regular training, multi-source data D under regular training is obtained pnew According to D pnew , W p and Y new Calculate the exercise-related factor X of the new subject under the conventional training new After obtaining the motion-related factors of each piece of information, new Perform feature extraction to obtain multiple individual irrelevant features; The calculation formula of the objective function is as follows: Where: X c Y is the exercise-related factor associated with the cth rehabilitation program; s is the individual correlation factor associated with the sth subject; The X,W corresponding to the minimum value of the objective function E is obtained p ,Y steps include: Transform the model formula of the p-th type of information to obtain the following formula: The first estimate of Y is made through singular value decomposition. The singular value decomposition formula is as follows: A=UΣV T Where: A is the matrix to be decomposed; U is the left singular matrix; Σ is a matrix in which all elements except the diagonal are 0, and the elements on the diagonal are singular values; V is the right singular matrix; Right D p Perform singular value decomposition and take the decomposed V T The first J rows of T The first estimate of According to the first estimate right Perform singular value decomposition and take the decomposed V T The first I rows are X T The first estimate of According to the first estimate Will Perform singular value decomposition and take the decomposed V T The first J rows of T The new estimate of and Each singular value decomposition is completed as one iteration, and after multiple iterations until convergence, the final and After transposition, we get X and Y that minimize the objective function, and W is calculated by the following formula p : W p =X T D p Y; The individual irrelevant features are features extracted from the motion-related factor X, including: absolute value mean in the time domain, autoregressive coefficient, median frequency in the frequency domain, and wavelet transform in the time-frequency domain.

2. The method for designing a lower limb rehabilitation program according to claim 1, characterized in that: The step of evaluating multiple individual irrelevant features and selecting a target feature according to the evaluation result comprises: The Davidson-Boulding Index (DBI) and the Scattering Distribution Index (SCAT) were used for feature evaluation; the definitions of DBI and SCAT are as follows: Where: N is the type of rehabilitation training program; is the average Euclidean distance from the i-th feature to its feature center; is the average Euclidean distance from the j-th feature to its feature center; ||ω i -ω j ||2 is the Euclidean distance between the feature centers of the i-th and j-th categories; SCAT=Trace(S w ) / Trace(S B ) Where: S w is the sum of the covariance matrices of all features; S B is the covariance matrix between different features; x is the eigenvalue of the i-th feature; μ i is the mean of the i-th feature; m mean is the mean of all features; Trace(·) is the trace of the matrix; The STN index is calculated based on the Davidson-Boulding index DBI and the scattering distribution index SCAT. The calculation formula is as follows: Where: DBI i and SCAT i The DBI index and SCAT index represent the individual-independent characteristics of the i-th rehabilitation program, and They represent the average values ​​of DBI index and SCAT index respectively; STN is the robust performance index; The features whose STN index meets the preset conditions are taken as target features.

3. The method for designing a lower limb rehabilitation program according to claim 1, characterized in that: The convolutional neural network is a multi-input multi-output convolutional neural network, and the shape of the convolution kernel is n×m×h×w, where m is the number of input feature types, n is the number of content items of the rehabilitation training program, and h and w are the length and width.

4. A device for designing lower limb rehabilitation programs suitable for different individuals, used to implement a method for designing lower limb rehabilitation programs suitable for different individuals as described in any one of claims 1 to 3, characterized in that: Includes the following modules: An acquisition and preprocessing module is used to acquire and preprocess the multi-source information of the subject, wherein the multi-source information includes: surface electromyography signal, heart rate, blood pressure, movement speed, movement acceleration, joint range of motion, muscle strength and muscle tension; A motion-related factor extraction module is used to extract motion-related factors from multi-source information based on a symmetric bilinear model, and to extract individual-irrelevant features from the motion-related factors; The feature evaluation and selection module is used to evaluate multiple individual irrelevant features and select target features based on the evaluation results; The network training and program acquisition module is used to input the target features into the convolutional neural network for training. After the training is completed, a lower limb rehabilitation program design model is obtained, and lower limb rehabilitation programs for different subjects are obtained based on the lower limb rehabilitation program design model.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for designing a lower limb rehabilitation program suitable for different individuals as described in any one of claims 1-3 are implemented.

6. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for designing a lower limb rehabilitation program suitable for different individuals as described in any one of claims 1 to 3 are implemented.

Images