Wrist joint spasm assessment method and device based on multi-source data fusion

Abstract

The invention discloses a wrist joint spasm assessment method and a wrist joint spasm assessment device based on multi-source data fusion. The rotating speed data, the rotating angular speed data and the angle change data of the wrist joint, the pressure change data of the whole palm, the acceleration change data of the palm, the torque change data of the wrist joint and the electromyographic change data of related muscles in the movement process are measured, and scoring is carried out in combination with an expert system. The evaluation of wrist spasm is realized by combining a multiple linear regression algorithm with a differential evolution algorithm; and based on the multi-source data fusion, establishing an evaluation model by using a Bi-LSTM network so as to realize spasm evaluation. The wrist serves as a spasm detection part, and compared with traditional spasm detection equipment, a more convenient data acquisition mode is achieved while the evaluation precision is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of medical devices, and particularly to a wrist joint spasm evaluation method and device based on multi-source data fusion. Background Art

[0002] Upper limb spasm is a neurological disorder commonly found in patients with diseases such as stroke, brain injury, Parkinson's disease, and multiple sclerosis. This spasm causes involuntary muscle contractions and twitches, severely affecting the patient's daily life and motor ability. According to research, approximately 30% to 40% of stroke patients experience spasm in the upper or lower limbs during the recovery period.

[0003] Compared with other parts of the body, the wrist is a relatively accessible and manipulable part because it is usually at the front end of the patient's arm without excessive obstructions. In contrast, the detection of spasm in other parts may require a more complex sensor installation and data collection process. The advantage of wrist spasm detection is that it can directly focus on the problems affecting hand function and motor ability.

[0004] Currently, the degree and type of spasm are usually judged by observing the patient's symptoms and conducting clinical evaluations, but this method is greatly affected by subjective factors and lacks objectivity and accuracy. Therefore, a new wrist joint spasm evaluation method and device are needed that can objectively and accurately evaluate the degree of wrist spasm in patients. Summary of the Invention

[0005] The purpose of the present invention is to provide a wrist joint spasm evaluation method and device based on multi-source data fusion. The method includes: real-time monitoring of the changes in the angle, angular velocity, acceleration, torque, and electromyogram data of the upper limb wrist joint when the wrist joint is induced to spasm by being pushed at different speeds; using the monitored multi-source data feedback to adjust the wrist joint movement speed in real time, so as to achieve a more accurate spasm level evaluation by combining the data changes with the expert system under the condition of ensuring safety.

[0006] The technical solution for achieving the purpose of the present invention is: a wrist joint spasm evaluation method based on multi-source data fusion, including:

[0007] By promoting the rotational movement of the wrist, real-time collection of multi-source dynamic movement data of the wrist and palm and muscle electromyogram data, and realizing the real-time fusion of multi-source data based on the collected multi-source data combined with the scoring data of the expert system;

[0008] Using the multiple linear regression algorithm combined with the differential evolution algorithm to establish and optimize a linear regression model with the collected multi-source data and the scoring data of the expert system;

[0009] Based on multi-source data fusion, a spasm evaluation model is constructed by using a bidirectional long short-term memory network combined with an attention mechanism to dynamically adjust the weights of multi-source data, so as to realize the evaluation of wrist joint spasm.

[0010] A wrist joint spasm evaluation device based on multi-source data fusion is used to implement the above method. The device includes a front movable support plate and a rear fixed support plate. The opposite ends of the front movable support plate and the rear fixed support plate are staggered and overlapped and hinged by a connecting shaft. One end of the connecting shaft is fixedly connected to an angle acquisition module, and the other end is fixedly connected to a torque acquisition module. The lower end of the front movable support plate is fixedly connected to a pressure acquisition module. The angle acquisition module, the pressure acquisition module, the torque acquisition module, and the electromyogram acquisition module communicate with the microprocessor module, and the microprocessor module communicates with the host computer.

[0011] The front movable support plate is hinged to the upper part of the movable connecting rod. The lower part of the movable connecting rod is connected to the lead screw through a fixed block. The fixed block is driven by a motor to rotate the lead screw to realize movement, and the front movable support plate is pushed by the movable connecting rod to rotate around the connecting shaft.

[0012] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it realizes the above-mentioned wrist joint spasm evaluation method based on multi-source data fusion.

[0013] Compared with the prior art, the advantages of the present invention are as follows:

[0014] (1) The present invention takes the wrist as the spasm detection site. Compared with traditional spasm detection devices, the present invention has a more convenient data acquisition method while ensuring the evaluation accuracy.

[0015] (2) Based on the multi-source data provided by multiple sensors, the present invention constructs a regression model by using a linear regression algorithm and a differential evolution algorithm, and constructs an evaluation model by using a bidirectional long short-term memory network combined with an attention mechanism. Compared with traditional spasm detection methods, it can give higher evaluation accuracy in a shorter detection time.

[0016] (3) Compared with the large and expensive detection devices, the device proposed by the present invention is small in size, high in structural integration, fully automated in the operation process, and has a supporting host computer software, which better meets the needs of clinical auxiliary diagnosis.

[0017] (4) The device proposed by the present invention realizes the automatic control and feedback of spasm detection. It realizes the automatic rotational movement of the wrist joint through the motor-driven lead screw and link mechanism, and combines the torque, electromyogram, and pressure data collected in real time to dynamically adjust the motor speed, ensuring the safety and accuracy of the evaluation process. At the same time, based on multi-sensors, the synchronous acquisition of the rotational speed, angle, pressure, torque, and electromyogram data of the wrist joint is realized, improving the comprehensiveness and accuracy of data acquisition. Description of the Drawings

[0018] Figure 1 It is the schematic diagram of the principle of the wrist joint spasm evaluation method based on multi-source data fusion of the present invention.

[0019] Figure 2 It is the schematic diagram of the structure of the wrist joint spasm evaluation device of the present invention.

[0020] Figure 3 It is the schematic diagram of the power mechanism and the electromyogram acquisition module.

[0021] Figure 4 It is the schematic diagram of the upper computer interface of the present invention. Detailed Implementation Modes

[0022] In order to make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] The present invention proposes a method and device for wrist spasm assessment based on multi-source data fusion. The method measures the rotational speed data, rotational angular velocity data, angular change data of the wrist joint, overall pressure change data of the palm, acceleration change data of the palm, torque change data of the wrist joint, and electromyogram change data of related muscles during the rotational movement of the human wrist by pushing it. Combining with the scoring of an expert system and using a multiple linear regression algorithm combined with a differential evolution algorithm to achieve the assessment of wrist spasm; based on the above multi-source data fusion and using a Bi-LSTM network to establish an assessment model to achieve the assessment of spasm. To ensure safety, limit the motor speed and pushing distance; set an emergency stop button for the overall device; through the overall processing of the wrist joint torque data, electromyogram data, and overall palm pressure data measured during the movement process, and adjust the motor speed in real time through feedback. Thus, it is possible to objectively and accurately assess wrist spasm on the premise of ensuring safety, and to address the problem that current clinical spasm detection is relatively vague and highly dependent on doctors' subjective experience, and a method and device for wrist spasm assessment with low cost, high safety, and high precision are proposed.

[0024] Combined with Figure 1 , a method for wrist spasm assessment based on multi-source data fusion according to the present invention includes the following steps:

[0025] By pushing the rotational movement of the human wrist, real-time collect multi-source dynamic movement data of the wrist and palm and muscle electromyogram data, and realize the real-time fusion of multi-source data based on the collected multi-source data combined with the scoring data of the expert system;

[0026] Use a multiple linear regression algorithm combined with a differential evolution algorithm to establish and optimize a linear regression model with the collected multi-source data and the scoring data of the expert system;

[0027] Based on multi-source data fusion, use a bidirectional long short-term memory network (Bi-LSTM) combined with an attention mechanism to dynamically adjust the weights of multi-source data, construct a spasm assessment model, and achieve precise assessment of wrist spasm;

[0028] During the assessment process, real-time monitor multi-source data, adjust the motor speed through a feedback mechanism to ensure the safety of the assessment process, and present the spasm assessment results in real time on the data display module.

[0029] Furthermore, the multi-source data includes the rotational speed data of the wrist joint, rotational angular velocity data, angular change data of the wrist joint, overall pressure change data of the palm, acceleration change data of the palm, torque change data of the wrist joint, electromyogram data of five muscles including the flexor carpi radialis, flexor carpi ulnaris, extensor carpi radialis longus, extensor carpi radialis brevis, and extensor carpi ulnaris, and the scoring data of the expert system.

[0030] The wrist joint rotation speed data, rotation angular velocity data, wrist joint angle change data, and palm acceleration change data are completed by the angle acquisition module. An incremental encoder is used to detect and count the position changes in real time, collect the angle and position data of the rotating support plate, and calculate the speed and acceleration information during the movement process. The overall palm pressure change data is completed by the pressure acquisition module, and a combination of a foil strain gauge and an alloy steel elastic body is used to record the pressure change information during the movement of the support plate. The wrist joint torque change data is completed by the torque acquisition module. One side of the uniaxial torque sensor is fixed to the bearing seat through a connecting piece, and the other side is connected and fixed to the connecting piece that fixes the support plate rotating shaft to collect the torque information during the movement process. The electromyography data is completed by the electromyography acquisition module, and an electromyography sensor is used to collect the electromyography data of five muscles: the flexor carpi radialis, flexor carpi ulnaris, extensor carpi radialis longus, extensor carpi radialis brevis, and extensor carpi ulnaris.

[0031] Further, the linear regression model is established through multi-source sensor data and scoring data. First, a linear regression equation is established. The linear regression equation is a regression analysis that models the relationship between one or more independent variables using the least squares function, and the function is a linear combination of the model parameters of one or more regression coefficients.

[0032] Further, in linear regression, the data is modeled using a linear prediction function, and the unknown function model parameters are also estimated through the data. The independent variables of the regression model are electromyography data, pressure data, and torque data, and the dependent variable is scoring data.

[0033] Further, the collected data is used to perform parameter fitting on the established multiple linear regression model through the method of machine regression, and finally a preliminary regression model can be obtained.

[0034] Further, the multiple linear regression algorithm is based on the collected multi-source data and expert scoring data to establish a multiple linear regression equation. Let the electromyography data be x1, the pressure data be x2, the torque data be x3, and the expert system scoring data be y. Then the multiple linear regression equation can be expressed as:

[0035] y = m0 + m1x1 + m2x2 + m3x3

[0036] Among them, m0 is the constant term, and m1, m2, and m3 are the partial regression coefficients of the electromyography data, pressure data, and torque data, respectively.

[0037] Further, the multiple linear regression can realize the establishment of a preliminary regression model, but only using the multiple linear regression will cause some errors, and further algorithms need to be combined to perform parameter fitting and optimization on the preliminary regression model.

[0038] Further, the differential evolution algorithm is used to perform parameter fitting and optimization on the obtained preliminary regression model.

[0039] The differential evolution algorithm is a kind of genetic evolution algorithm and an evolutionary algorithm for solving optimization problems. It mutates, crosses, and selects a randomly initialized set to generate a better set than before.

[0040] First, initialize the population. The overall population is represented as a two-dimensional matrix. Each individual is represented by a group of real number row vectors in the matrix. Therefore, the number of rows of the two-dimensional matrix is the total number (NP). The number of columns represents the number of dimensions (D), and each dimension is constrained within an interval inside. Use a uniform random number generator to randomly generate each real number parameter as follows:

[0041]

[0042] where is the generated real number parameter, rand ji (0, 1) is the random number generator, and are the maximum and minimum values of parameter j.

[0043] In the establishment of the regression model, the population consists of four parameters: a0, a1, a2, and a3. Each individual in the population is a solution composed of four parameters, representing a potential solution. Then, use a uniform random number generator to randomly generate a certain number of solutions within the search space for population initialization. Among them, define the value range of variables:

[0044] bounds = [(0, 1), (0, 1), (0, 1), (0, 1)]

[0045] Exemplarily, define the population size (popsize) as 20.

[0046] After initializing a population, randomly perform differential mutation on the individuals in it to obtain a mutant vector as follows:

[0047]

[0048] where g is the number of iterations, is the mutation vector of the i-th individual in the g-th generation, is the base vector, are two different individuals in the population, and F = [0, 1] is the scaling factor used to scale the differences between parameters.

[0049] Exemplarily, a scaling factor is defined as 0.5 in the establishment of the regression model because a high mutation rate has a greater impact on the population but may also cause the algorithm to fall into a local optimal solution.

[0050] Perform crossover processing. The crossover process usually uses binomial crossover and exponential crossover, with binomial crossover being more commonly used. Cr = [0, 1] is defined as the crossover rate. If the value generated by the random number generator for the j-th parameter of an individual is less than or equal to C, the original carrier will be replaced by the mutant carrier; otherwise, it remains unchanged. The formula is as follows:

[0051]

[0052] Exemplarily, a crossover rate is defined as 0.7 in the establishment of the regression model. Increasing the diversity of the population can also prevent the algorithm from falling into a local optimal solution.

[0053] Since the objective function is a function optimized to the minimum value, in the next generation, when the function value of the mutant carrier is less than that of the original carrier, the mutant carrier will replace the original carrier; otherwise, the original carrier is retained.

[0054]

[0055] Perform iteration. Iteration means continuously repeating the mutation, crossover, and selection operations until a preset termination condition is reached. In the regression model, the optimization process terminates according to the set number of iterations. In each iteration, by calling the objective function (target_func):

[0056] y = a0 + a1x1 + a2x2 + a3x3

[0057] And return the error value. Here, a0, a1, a2, and a3 are population parameters.

[0058] Calculate the fitness of each individual, and the callback function (record_params) for recording parameter changes is used for subsequent analysis. The final solution will be the individual with the best fitness in the population.

[0059] Furthermore, the spasm evaluation model is constructed based on the regression model by using a bidirectional long short-term memory network combined with an attention mechanism.

[0060] The construction of a wrist joint spasm detection regression model based on multi-source data is divided into four stages: data collection, data analysis and preprocessing, data fusion, and model training and spasm detection. First, multi-sensor data is collected, including the rotational speed data of the wrist joint, the rotational angular velocity data, the angular change data of the wrist joint, the pressure change data of the overall palm, the acceleration change data of the palm, the torque change data of the wrist joint, and the electromyogram change data of the relevant muscles. Then, the collected raw data is preprocessed using methods such as filtering to remove high-frequency noise and artifacts, and feature selection and extraction are completed through the analysis of the wrist spasm process. Feature splicing is used to achieve the fusion of various data. Finally, the spliced feature sequence is used to complete the training of the time series model, and spasm detection and evaluation are completed.

[0061] Furthermore, the Bi-LSTM is a Bidirectional Long Short-Term Memory Network and can be well applied to data classification tasks. By combining it with the Attention mechanism, the performance of the model can be further improved.

[0062] The Attention mechanism can help the model pay more attention to the important parts in the input sequence. By dynamically allocating weights to different positions in the sequence, the Attention mechanism can make the model focus more on the key information related to the classification task, thereby improving the classification performance of the model.

[0063] The combination of the Attention mechanism and Bi-LSTM to build a spasm evaluation network model based on deep learning mainly consists of five layers. The input layer is used to input the processed sensor feature vectors into the network. The Bi-LSTM+Attention layer first uses the bidirectional long short-term memory network to further learn the temporal information existing in the feature vectors, and then uses the Attention mechanism to synthesize the data features at different times and outputs a feature matrix for data classification. The linear layer + Softmax module takes the output matrix of the Attention layer as input, completes the mapping from features to data categories, and outputs the recognition result.

[0064] Based on multi-source data fusion, a spasm evaluation model is constructed by using a bidirectional long short-term memory network combined with an attention mechanism to dynamically adjust the weights of multi-source data, so as to realize the evaluation of wrist joint spasm. Specifically:

[0065] First, data preprocessing and feature extraction are carried out. The multi-source data collected is preprocessed, including denoising, normalization, and feature extraction. The multi-source data includes the rotational speed of the wrist joint, the angular velocity of rotation, the angular change, the change in palm pressure, the change in palm acceleration, the change in wrist joint torque, the electromyogram change of related muscles, and the scoring data of the expert system. Through filtering and feature selection, the key features related to spasm are extracted to form a feature vector. Then, a bidirectional long short-term memory network (Bi-LSTM) is used to model the temporal features of the multi-source data. The Bi-LSTM network consists of a forward LSTM and a backward LSTM, which can capture the forward and backward temporal dependencies of the data simultaneously. Let the input feature sequence be X = {x1, x2, …, x t}, where x t is the feature vector at time t, and the output of Bi-LSTM is:

[0066] h t = Bi_LSTM(x t , h t-1 )

[0067] where h t is the hidden state at time t, which contains the outputs of the forward and backward LSTMs.

[0068] Then, an attention mechanism is introduced based on Bi-LSTM to dynamically adjust the weights of the multi-source data, enabling the model to focus on the features most relevant to spasm assessment. The weighted feature vector is input into the fully connected layer and the Softmax layer to complete the mapping from features to spasm categories, and the probability distribution of the spasm assessment result is output. Using the multi-source data and the scoring data of the expert system, the Bi-LSTM network and the attention mechanism are jointly trained, and the cross-entropy loss function and the gradient descent method are used to optimize the model parameters until the model converges.

[0069] Based on the spasm assessment model and the real-time collected rotational speed data of the wrist joint, the angular velocity data of rotation, the angular change data of the wrist joint, the pressure change data of the whole palm, the acceleration change data of the palm, the torque change data of the wrist joint, the electromyogram change data of related muscles, and the scoring data of the expert system, the spasm assessment result is given on the data display module.

[0070] Combined with Figure 2 , Figure 3 , a wrist joint spasm assessment device based on multi-source data fusion mainly consists of a front active support plate 1, a rear fixed support plate 2, a connecting shaft 3, an angle acquisition module 4, a torque acquisition module 5, a pressure acquisition module 6,

[0071] an electromyogram acquisition module 7, an active connecting rod 8, a lead screw 9, a fixed block 10, and a motor 11.

[0072] The opposite ends of the front movable support plate 1 and the rear fixed support plate 2 are staggeredly overlapped and hinged by a connecting shaft 3. One end of the connecting shaft 3 is fixedly connected to an angle acquisition module 4, and the other end is fixedly connected to a torque acquisition module 5. The lower end of the front movable support plate 1 is fixedly connected to a pressure acquisition module 6. The front movable support plate 1 and the pressure acquisition module 6 are hinged to the upper part of a movable link 8. The lower part of the movable link 8 is connected to a lead screw 9 through a fixing block 10. The fixing block 10 drives the lead screw 9 to rotate through the rotation of a motor 11 to achieve movement. The front movable support plate 1 is pushed by the movable link 8 to rotate around the connecting shaft 3.

[0073] The angle acquisition module 4, the pressure acquisition module 6, the torque acquisition module 5, and the electromyogram acquisition module 7 communicate with a microprocessor module, and the microprocessor module communicates with a host computer. The microprocessor module mainly includes an A / D conversion module, a signal acquisition module, and a signal sending module.

[0074] The front movable support plate 1 is hinged to the upper part of the movable link 8. The lower part of the movable link 8 is connected to the lead screw 9 through the fixing block 10. The fixing block 10 drives the lead screw 9 to rotate through the rotation of the motor 11 to achieve movement. The front movable support plate 1 is pushed by the movable link 8 to rotate around the connecting shaft 3.

[0075] The microprocessor module mainly includes an A / D conversion module, a signal acquisition module, and a signal sending module; the host computer includes a signal receiving module, a signal processing module, and a data display module. The signal receiving module receives multi-source data transmitted from the signal sending module, including the rotation speed data of the wrist joint, the rotation angular velocity data, the wrist joint angle change data, the overall palm pressure change data, the palm acceleration change data, the wrist joint torque change data, the electromyogram change data of related muscles, and the expert system scoring data. The signal processing module is used to preprocess and analyze the received multi-source data, including data filtering, denoising, feature extraction, and data fusion, and input the processed data into a spasm evaluation model to generate a spasm evaluation result. The processed data is displayed and stored in real time through the data display module.

[0076] Refer to Figure 4 , the host computer interface consists of a data display module 14, a data visualization module 15, and a control input module 17. The data input module 14 includes a spasm occurrence indicator 12 and a multi-source data and running time display 13. The data visualization module 15 includes a multi-source data visualization interface 16. The input control module 17 includes an emergency stop button 18, a control input device 19, and a buffer data display 20.

[0077] Furthermore, by controlling the input device 19, the data storage path, the selected baud rate, sampling frequency, and sampling time can be input. The spasm occurrence indicator 12 will give a prompt when detecting the occurrence of the user's spasm. The multi-source data collected from the electromyogram data, pressure data, angle data, and torque data will be displayed in real time on the multi-source data and running time display 13. The multi-source data obtained by the four sensors in real time is visualized through the multi-source data visualization interface 16 to show the data change trend. The emergency stop button 18 can achieve an emergency stop of the device operation.

[0078] The wrist spasm evaluation device of the present invention adopts a three-layer composite structure. The three-layer composite structure compresses the internal space of the device by using multi-layer stacking and combination, improves the overall integration, and reduces the volume of the device.

[0079] The first layer of the three-layer composite structure is the interaction layer, which is composed of a front movable support plate, a rear fixed support plate, a connecting shaft, an angle acquisition module, a torque acquisition module, and a user wrist fixing strap. The front movable support plate has multiple hollow grooves and through holes, which are symmetrically distributed left and right and are used to fix the hand and screws. The rear fixed support plate contacts the support column, and there are multiple groups of hollow grooves and through holes on the surface of the rear support plate, which are used to fix the forearm and the support column. The support column is fixedly connected to the rear support plate and the aluminum profile. The connecting shaft is fixed by a rotating shaft fixing plate, and the rotating shaft fixing plate has through holes for fixing the rotating shaft and connecting to the front rotating support plate. The angle acquisition module includes an angle sensor and an angle sensor fixing base, and the torque acquisition module includes a torque sensor and a torque sensor fixing base. The angle acquisition module and the torque acquisition module are respectively fixed at both ends of the rotating shaft. The torque sensor fixing base is fixed on the rotating shaft fixing plate by screws, and the angle sensor fixing base has through holes for fixing the angle sensor and the rotating shaft.

[0080] The second layer of the three-layer composite structure is the execution layer, which is composed of a link execution module, a transmission system module, a pressure acquisition module, a motion control module, and a low-voltage power supply module. The link execution module includes a pressure sensor hinge fixer, a slider hinge fixer, and a support link. The pressure sensor hinge fixer includes upper and lower parts, which are connected by screws and connected to the support link. The slider hinge fixer also consists of upper and lower parts, which are connected by screws and fixed to the slider. The transmission system module includes a ball screw, a slider, and a coupling, which are used to transmit the rotation of the motor and realize the linear motion of the slider. The motion control module includes a motor, a motor drive, and a power supply, all of which are fixed on the second layer. The low-voltage power supply module mainly provides power for the electromyogram sensor and is also fixed on the second layer.

[0081] The third layer of the three-layer composite structure is the control layer, which mainly fixes the switching power supply, the driver, the microprocessor, the emergency stop and control buttons.

[0082] The host computer mainly includes a data display module, a data visualization module, and a control input module.

[0083] The data display module includes a spasm occurrence indicator, a multi-source data and running time display; the data visualization module includes a multi-source data visualization interface; the control input module includes an emergency stop button, a control input device, and a buffer data display. Data storage paths are input through the control input device, and the baud rate, sampling frequency, and sampling time are selected; the spasm occurrence indicator gives a prompt when the occurrence of the user's spasm is detected, and the multi-source data and running time display is used to display electromyogram data, pressure data, angle data, and torque data in real time; the emergency stop button is used to achieve an emergency stop of the device operation.

[0084] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A wrist spasticity assessment method based on multi-source data fusion, characterized in that: include: By promoting the rotation of the wrist, multi-source dynamic motion data and muscle electromyography data of the wrist and palm are collected in real time. The real-time fusion of multi-source data is achieved based on the collected multi-source data combined with the expert system scoring data; The multivariate linear regression algorithm is combined with the differential evolution algorithm to establish and optimize the linear regression model by combining the collected multi-source data with the expert system scoring data; Based on multi-source data fusion, a bidirectional long short-term memory network is used in combination with the attention mechanism to dynamically adjust the weights of multi-source data, build a spasm assessment model, and realize the assessment of wrist spasm.

2. The wrist spasticity assessment method based on multi-source data fusion according to claim 1, characterized in that: During the assessment process, multi-source data are monitored in real time, the motor speed is adjusted through a feedback mechanism, and the spasticity assessment results are presented in real time in the data display module.

3. The wrist spasticity assessment method based on multi-source data fusion according to claim 1, characterized in that: The multi-source data include wrist joint rotation speed data, rotation angular velocity data, wrist joint angle change data, palm overall pressure change data, palm acceleration change data, wrist joint torque change data, and electromyographic data of five muscles, namely, radial flexor carpi, ulnar flexor carpi, radial extensor carpi longus, radial extensor carpi brevis, and ulnar extensor carpi, as well as expert system scoring data.

4. The wrist spasticity assessment method based on multi-source data fusion according to claim 3, characterized in that: The wrist joint rotation speed data, rotation angular velocity data, wrist joint angle change data, and palm acceleration change data are completed by the angle acquisition module, which uses an incremental encoder to detect and count position changes in real time, collects the angle and position data of the support plate rotation, and calculates the speed and acceleration information during the movement; the overall palm pressure change data is completed by the pressure acquisition module, and the foil strain gauge and alloy steel elastomer are combined to record the pressure change information during the movement of the support plate; the wrist joint torque change data is completed by the torque acquisition module, and one side of the uniaxial torque sensor is fixed to the bearing seat through a connecting piece, and the other side is connected and fixed to the connecting piece that fixes the support plate rotating shaft to collect the torque information during the movement; the electromyographic data is completed by the electromyographic acquisition module, and the electromyographic sensor is used to collect the electromyographic data of five muscles: radial flexor carpi, ulnar flexor carpi, radial extensor carpi longus, radial extensor carpi brevis and ulnar extensor carpi.

5. The wrist spasticity assessment method based on multi-source data fusion according to claim 1, characterized in that: The multivariate linear regression algorithm is combined with the differential evolution algorithm to establish and optimize the linear regression model by combining the collected multi-source data with the expert system scoring data. Specifically: Assume that the electromyographic data is x1, the pressure data is x2, the torque data is x3, and the expert system scoring data is y, then the multivariate linear regression equation can be expressed as: y=m0+m1x1+m2x2+m3x3 Among them, m0 is a constant term, and m1, m2, and m3 are partial regression coefficients of electromyographic data, pressure data, and torque data, respectively; The differential evolution algorithm performs parameter fitting and optimization on the linear regression model. First, the population is initialized. Each individual in the population consists of four parameters a0, a1, a2, and a3, representing a potential solution. The population initialization formula is: in, is the initial value of the i-th individual of the j-th parameter, and is the maximum and minimum value of parameter j, rand ji (0,1) is a random number generated by a uniform random number generator; then the individuals in the population are mutated to generate mutation carriers; the mutation formula is: Where g is the number of iterations, is the mutation vector of the i-th individual in the g-th generation, is the basis vector, are two different individuals in the population, and F = [0, 1] is a scaling factor used to scale the difference between parameters; then a crossover operation is performed to generate a test vector; the crossover formula is: in is the test vector of the i-th individual of the g-th generation, and Cr is the crossover rate. By comparing the fitness values ​​of the test vector and the original vector, the better individual is selected to enter the next generation. The selection formula is: Where f() is the objective function, which represents the error of the regression model. is the new individual after selection, is the test vector of the i-th individual in the g-th generation, is the original vector of the i-th individual in the g-th generation; the mutation, crossover and selection operations are repeated until the preset number of iterations or error threshold is reached; finally, the differential evolution algorithm outputs the optimized regression coefficient to complete the establishment of the multivariate linear regression model.

6. The wrist spasticity assessment method based on multi-source data fusion according to claim 1, characterized in that: Based on multi-source data fusion, a bidirectional long short-term memory network is used in combination with an attention mechanism to dynamically adjust the weights of multi-source data and build a spasm assessment model to evaluate wrist spasm. Specifically: First, data preprocessing and feature extraction are performed to preprocess the collected multi-source data, including denoising, normalization and feature extraction; through filtering and feature selection, wrist joint motion features, mechanical features, muscle electromyography features, time series features, and expert system scoring data related to spasticity are extracted to form feature vectors; Then, the bidirectional long short-term memory network is used to model the temporal characteristics of multi-source data; The Bi-LSTM network consists of a forward LSTM and a backward LSTM, which can capture the forward and backward temporal dependencies of the data at the same time. Suppose the input feature sequence is X = {x1, x2, ..., x t }, where x t is the feature vector at time t, and the output of Bi-LSTM is: h t =Bi_LSTM(x t ,h t-1 ) Among them, h t is the hidden state at time t, including the output of forward and backward LSTM; An attention mechanism is introduced based on Bi-LSTM to dynamically adjust the weights of multi-source data, so that the model can focus on the features most relevant to spasticity assessment. The weighted feature vector is input into the fully connected layer and the Softmax layer to complete the mapping of features to spasm categories and output the probability distribution of the spasm assessment results. The Bi-LSTM network and the attention mechanism are jointly trained using multi-source data and expert system scoring data, and the cross entropy loss function and gradient descent method are used to optimize the model parameters until the model converges.

7. A wrist joint spasm assessment device based on multi-source data fusion, characterized in that: The device is used to implement any of the methods described in claims 1 to 6, comprising a front movable support plate (1) and a rear fixed support plate (2), wherein the facing ends of the front movable support plate (1) and the rear fixed support plate (2) are staggered and overlapped and hinged through a connecting shaft (3), one end of the connecting shaft (3) is fixedly connected to an angle acquisition module (4), and the other end is fixedly connected to a torque acquisition module (5), and the lower end of the front movable support plate (1) is fixedly connected to a pressure acquisition module (6); the angle acquisition module (4), the pressure acquisition module (6), the torque acquisition module (5), and the electromyography acquisition module (7) communicate with a microprocessor module, and the microprocessor module communicates with a host computer; The front movable support plate (1) is hinged to the upper part of the movable connecting rod (8), and the lower part of the movable connecting rod (8) is connected to the lead screw (9) via a fixed block (10). The fixed block (10) drives the lead screw (9) to rotate through the motor (11) to achieve movement, and the front movable support plate (1) is pushed by the movable connecting rod (8) to achieve rotation around the connecting shaft (3).

8. The wrist spasticity assessment device based on multi-source data fusion according to claim 7, characterized in that: The microprocessor module mainly includes an A / D conversion module, a signal acquisition module and a signal sending module. The host computer includes a signal receiving module, a signal processing module and a data display module. The signal receiving module receives multi-source data transmitted from the signal sending module, and uses the signal processing module to pre-process and analyze the received multi-source data, including data filtering, denoising, feature extraction and data fusion, and inputs the processed data into the spasm assessment model to generate a spasm assessment result; the processed data is displayed and stored in real time through the data display module.

9. The wrist spasticity assessment device based on multi-source data fusion according to claim 7, characterized in that: The upper computer interface is composed of a data display module (14), a data visualization module (15) and a control input module (17); the data display module (14) includes a spasm occurrence indicator (12) and a multi-source data and run time display (13); the data visualization module (15) includes a multi-source data visualization interface (16); the control input module (17) includes an emergency stop button (18), a control input device (19) and a buffer data display (20); the data storage path is input through the control input device (19), and the baud rate, sampling frequency and sampling time are selected; the spasm occurrence indicator (12) gives a prompt when the occurrence of user spasm is detected, and the multi-source data and run time display (13) is used to display electromyographic data, pressure data, angle data and torque data in real time; the emergency stop button (18) is used to realize the tight stop of the device operation.

10. 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 according to any one of claims 1 to 6 are implemented.

Images