Prediction method, device, system and storage medium for limb spasticity after stroke

Abstract

The present invention provides a prediction method, device, system and storage medium for limb spasticity after stroke, which relates to the technical field of stroke prognosis assessment. The method includes: collecting static EEG signal data and static magnetic resonance imaging data of the person to be tested in a resting state and obtaining a resting state feature vector; collecting dynamic EEG signal data and dynamic electromyogram in a motion task state and obtaining a task state feature vector; integrating and splicing and obtaining a state score through a multi-layer perceptron to obtain a prediction result. The present invention combines static and dynamic data to capture the patient's comprehensive neurophysiological state, uses a multi-layer perceptron to extract key features from complex data, and constructs an accurate prediction model, which not only improves the accuracy and reliability of the prediction, but also can provide a fast and accurate spasm risk assessment for the clinic.

Description

Technical Field

[0001] The present invention relates to the technical field of stroke prognosis assessment, and more specifically, to a method, device, system and storage medium for predicting limb spasticity after stroke for non-diagnostic purposes. Background Art

[0002] Stroke is one of the leading causes of adult disability worldwide, and its sequelae seriously affect the patient's quality of life and survival rate. Post-stroke spasticity (PSS) is one of the common complications, which is characterized by persistent and unconscious muscle contraction, severely limiting the patient's ability to move and daily life. According to statistics, about 25% of stroke patients develop PSS within two weeks of onset, and this proportion will increase to 38% to 50% after six months to one year. This condition not only reduces the patient's quality of life, but also increases the burden on caregivers and socioeconomic costs.

[0003] At present, the assessment and prediction of PSS mainly rely on clinical subjective assessment scales, such as the Modified Ashworth Scale (MAS) and the Tardieu scale. These scales determine the severity of spasticity by evaluating the patient's muscle tension, providing a basis for clinical diagnosis and assessment. However, these methods rely on the subjective feelings of doctors or therapists, have certain assessment errors, and are difficult to accurately quantify the degree of spasticity.

[0004] Existing technologies have obvious defects in the assessment and prediction of PSS. First, due to the lack of objective assessment tools, the judgment of the degree of spasticity is highly subjective, which may lead to significant differences in the assessment results between different assessors. Secondly, most of the current assessment methods are limited to static or semi-static tests, which are difficult to fully reflect the dynamic challenges encountered by patients in daily life. In addition, existing assessment methods usually require professional operation and interpretation, which increases the difficulty of monitoring and managing PSS in a home environment for ordinary caregivers and family members of patients.

[0005] In summary, the existing PSS assessment and prediction methods cannot effectively reflect the actual spasticity status and treatment needs of patients due to their lack of objectivity and real-time performance. The limitations of these technologies not only affect the precise guidance of patient recovery, but may also lead to irrational allocation of treatment resources. Therefore, the development of a technology that can objectively, accurately and in real time assess limb spasticity after stroke is an urgent problem that needs to be solved, which will greatly promote the patient's rehabilitation process and improve their quality of life. Summary of the invention

[0006] In view of this, in view of the above technical problems, the present invention provides a method for predicting limb spasticity after stroke, comprising:

[0007] Collecting static EEG signal data and static MRI data of the person to be tested in a resting state, and obtaining a resting state feature vector based on the static EEG signal data and the static MRI data;

[0008] Collecting dynamic EEG signal data and dynamic electromyogram of the person to be tested in a motion task state, and obtaining a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram;

[0009] The resting state feature vector and the task state feature vector are integrated and spliced, and a state score is obtained through a multi-layer perceptron, and a prediction result of the person to be tested is obtained through the state score.

[0010] Preferably, obtaining a resting state feature vector based on the static EEG signal data and the static MRI data includes:

[0011] Calculating a static EEG brain connection map based on the static EEG signal data; and calculating a static fMRI brain connection map based on the static magnetic resonance imaging data;

[0012] The static EEG brain connection map and the static fMRI brain connection map are used to form a node-level fusion feature vector from a node granularity; and the static EEG brain connection map and the static fMRI brain connection map are spliced ​​through an attention mechanism to obtain a global-level feature vector;

[0013] The node-level fusion feature vector and the global-level feature vector are concatenated to form the resting-state feature vector.

[0014] Preferably, the step of calculating a static EEG brain connectivity map based on the static EEG signal data comprises:

[0015] The distances between the channels of the static EEG signal are calculated according to the Euclidean distance, thereby constructing a distance matrix; the calculation expression of the distance matrix is: ij =||x i -x j || 2 ; Among them, D ij represents the Euclidean distance between channels i and j, x i represents the three-dimensional topological coordinates of channel i, x j represents the three-dimensional topological coordinates of channel j;

[0016] According to the distance matrix, each channel is clustered using the k-nearest neighbor algorithm to construct the static EEG brain connection map in which each channel is associated with other k channels.

[0017] Preferably, the calculating of a static fMRI brain connectivity map based on the static magnetic resonance imaging data comprises:

[0018] Dividing the brain image in the static magnetic resonance imaging data into a plurality of unit brain regions;

[0019] Extracting time series data of the unit brain area;

[0020] Calculate the Pearson correlation coefficient of the time series data of any two unit brain regions; the calculation expression of the Pearson correlation coefficient is: Among them, ρ X,Y represents the correlation coefficient of the time series X and Y of the two unit brain regions, cov(X,Y) is the covariance of the two time series, σ X and σ Y Represents the standard deviation of X and Y series respectively; μ X and μ Y Represent the means of X and Y series respectively;

[0021] The connection relationship between different unit brain regions is screened by a preset threshold to obtain the static fMRI brain connection map.

[0022] Preferably, the step of forming a node-level fusion feature vector from the static EEG brain connection map and the static fMRI brain connection map at a node granularity comprises:

[0023] Arrange the nodes of the static EEG brain connection map and the static fMRI brain connection map in rows, respectively, and concatenate them to obtain a representation matrix corresponding to the static EEG brain connection map and a representation matrix corresponding to the static fMRI brain connection map;

[0024] Transposing the representation matrix corresponding to the static EEG brain connection map and the representation matrix corresponding to the static fMRI brain connection map respectively, and performing matrix multiplication processing to obtain a corresponding node-level fusion matrix;

[0025] The numerical distribution in each node-level fusion matrix is ​​counted in the form of a histogram, and the node-level fusion feature vector is formed by the frequency of each group of the histogram.

[0026] Preferably, the step of obtaining a global feature vector by concatenating the static EEG brain connection map and the static fMRI brain connection map through an attention mechanism comprises:

[0027] Input the representation matrices into the corresponding attention mechanisms to select the top-k nodes;

[0028] Taking the average of the feature vectors of all the top-k nodes, and respectively obtaining the global vector representation information of the static EEG brain connection map and the static fMRI brain connection map;

[0029] The vector evidence information corresponding to the static EEG brain connection map and the static fMRI brain connection map is spliced ​​to form the global level feature vector.

[0030] Preferably, obtaining the task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram comprises:

[0031] The dynamic EEG signal data is subjected to wavelet decomposition to construct a dynamic EEG feature vector; and an energy value feature is obtained according to the dynamic EEG signal data;

[0032] Combining the dynamic EEG feature vector and the energy value feature to obtain a dynamic feature vector;

[0033] Extracting dynamic sEMG features according to the dynamic electromyogram;

[0034] The energy value feature and the dynamic sEMG feature are reduced in dimension by a support vector machine to construct a weight matrix;

[0035] Based on the weight matrix, the dynamic EEG signal data and the dynamic electromyogram are classified by the classification model of the support vector machine to obtain a classification result matrix;

[0036] Multiply the classification result matrix by the weight matrix to obtain a weighted classification result vector;

[0037] Acquire MEP detection information, and extract MEP feature vectors corresponding to the MEP detection information;

[0038] The weighted classification result vector, the MEP feature vector, the dynamic feature vector and the dynamic sEMG feature are concatenated to obtain the task state feature vector.

[0039] In addition, to solve the above problems, the present invention also provides a prediction device for limb spasticity after stroke, comprising:

[0040] A static characterization module, used to collect static EEG signal data and static MRI data of the person to be tested in a resting state, and obtain a resting state feature vector based on the static EEG signal data and the static MRI data;

[0041] A dynamic characterization module, used for collecting dynamic EEG signal data and dynamic electromyogram of the person to be tested in a motion task state, and obtaining a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram;

[0042] The evaluation prediction module is used to integrate and splice the resting state feature vector and the task state feature vector, and obtain a state score through a multi-layer perceptron, and obtain a prediction result of the person to be tested through the state score.

[0043] In addition, to solve the above problems, the present invention also provides a prediction system for limb spasm after stroke, including a memory and a processor, wherein the memory stores a prediction program for limb spasm after stroke, and the processor runs the prediction program for limb spasm after stroke so that the prediction system for limb spasm after stroke executes the prediction method for limb spasm after stroke as described above.

[0044] In addition, to solve the above problems, the present invention also provides a computer-readable storage medium, on which a prediction program for limb spasm after stroke is stored. When the prediction program for limb spasm after stroke is executed by a processor, the prediction method for limb spasm after stroke as described above is implemented.

[0045] The present invention provides a prediction method, device, system and storage medium for post-stroke limb spasticity. The evaluation method is applied to the evaluation and prediction of the clinical state of post-stroke hemiplegia for non-diagnostic purposes, including: collecting static EEG signal data and static magnetic resonance imaging data of the person to be tested in a resting state, and obtaining a resting state feature vector based on the static EEG signal data and the static magnetic resonance imaging data; collecting dynamic EEG signal data and dynamic electromyogram of the person to be tested in a motion task state, and obtaining a task state feature vector based on the dynamic EEG signal data and the dynamic electromyogram; integrating and splicing the resting state feature vector and the task state feature vector, and obtaining a state score through a multi-layer perceptron, and obtaining a prediction result of the person to be tested through the state score.

[0046] The method for predicting post-stroke limb spasticity provided by the present invention has the following beneficial effects:

[0047] First, by combining static and dynamic data, this method can more comprehensively capture the patient's neurophysiological state and provide more accurate prediction results. Static data reflects the patient's basic neurophysiological conditions, while dynamic data can reveal the patient's neural response and muscle coordination during actual activities. This comprehensive analysis helps to comprehensively assess the risk and degree of spasticity.

[0048] Secondly, through the application of multi-layer perceptron, this method can effectively extract key features from large amounts of complex data and obtain a more accurate prediction model through learning, thereby improving the accuracy and reliability of the prediction.

[0049] In addition, this method has strong real-time and operability, and can provide clinicians with rapid and accurate spasm risk assessment, which helps to adjust treatment plans in a timely manner and optimize resource allocation, thereby improving treatment efficiency and patients' quality of life.

[0050] In conclusion, this prediction method can provide scientific and effective technical support for the prediction and management of limb spasticity after stroke through highly integrated and intelligent data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a structural schematic diagram of the hardware operating environment involved in the embodiment of the method for predicting limb spasticity after stroke of the present invention;

[0052] Figure 2 Schematic diagram of the process of Example 1 of the method for predicting limb spasticity after stroke of the present invention;

[0053] Figure 3 This is a schematic flow chart of the refinement of step S100 in Example 2 of the method for predicting limb spasticity after stroke of the present invention;

[0054] Figure 4 Schematic diagram of the process of refining step S110 (S111-S112) in Example 2 of the method for predicting limb spasticity after stroke of the present invention;

[0055] Figure 5 Schematic diagram of the process of refining step S110 (S113-S116) in Example 2 of the method for predicting limb spasticity after stroke of the present invention;

[0056] Figure 6 Schematic diagram of the process of refining step S120 (S121-S123) in Example 2 of the method for predicting limb spasticity after stroke of the present invention;

[0057] Figure 7 Schematic diagram of the process of refining step S120 (S124-S126) in Example 2 of the method for predicting limb spasticity after stroke of the present invention;

[0058] Figure 8 This is a schematic flow chart of the refinement of step S200 in Example 3 of the method for predicting limb spasticity after stroke of the present invention;

[0059] Fig. 9 The figure is a schematic diagram of module connections of the device for predicting limb spasticity after stroke of the present invention.

[0060] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0061] The embodiments of the present invention are described in detail below, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions.

[0062] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0063] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0064] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0065] like Figure 1 , which is a schematic diagram of the structure of the hardware operating environment of the terminal involved in the embodiment of the present invention.

[0066] The prediction system for post-stroke limb spasticity provided by the embodiment of the present invention can be a PC, or a mobile terminal device such as a smart phone, a tablet computer or a portable computer. The prediction system for post-stroke limb spasticity may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005 and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen, an input unit such as a keyboard, and a remote controller, and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory, such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001. Optionally, the prediction system for post-stroke limb spasticity may also include an RF (Radio Frequency, radio frequency) circuit, an audio circuit, a WiFi module, and the like. In addition, the prediction system for limb spasticity after stroke may also be configured with other sensors such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc., which will not be described in detail here.

[0067] Those skilled in the art will understand that Figure 1 The prediction system for post-stroke limb spasticity shown in the figure does not constitute a limitation thereof, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently. Figure 1 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a data interface control program, a network connection program, and a prediction program for post-stroke limb spasticity.

[0068] In summary, the post-stroke limb spasticity prediction method of the present invention comprehensively captures the patient's neurophysiological state by combining static and dynamic data, thereby improving the accuracy of the prediction. Static data reveals the basic neural conditions, while dynamic data displays the neural responses and muscle coordination in actual activities, which helps to comprehensively assess the risk of spasticity. This method uses a multi-layer perceptron to extract key features from complex data, forming an accurate prediction model and enhancing the reliability of the prediction. With strong real-time and operability, this method can quickly provide spasticity risk assessment, optimize treatment plans and resource allocation, significantly improve treatment efficiency and patient quality of life, and provide effective technical support for post-stroke spasticity management.

[0069] Embodiment 1:

[0070] Reference Figure 2Embodiment 1 of the present invention provides a method for predicting limb spasticity after stroke, which is applied to the evaluation and prediction of limb spasticity after stroke for non-diagnostic purposes. The evaluation method comprises:

[0071] Step S100, collecting static EEG signal data and static MRI data of the person to be tested in a resting state, and obtaining a resting state feature vector based on the static EEG signal data and the static MRI data;

[0072] The above-mentioned persons to be tested may be patients who need to undergo a predictive evaluation of limb spasticity after stroke.

[0073] The static electroencephalogram signal data mentioned above may be static EEG signal data.

[0074] Among them, EEG, Electroencephalogram, is a method of measuring the electrical activity of the brain.

[0075] The static magnetic resonance imaging data mentioned above may be fMRI signal data.

[0076] Among them, fMRI, Functional Magnetic Resonance Imaging, is an imaging technology that uses magnetic fields and radio waves to measure changes in brain blood flow.

[0077] Collect EEG data and MRI data from the person under test in a resting state. Static EEG data can be collected by wearing a specific headgear device equipped with multiple electrodes that can record the electrical activity of the brain in a completely resting state.

[0078] At the same time, through the nuclear magnetic resonance imaging technology (fMRI), the structural and functional images of the brain can be obtained. Based on these data, data processing technology (such as signal processing algorithms and image processing software) is used to generate a resting state feature vector, which can reflect the basic neurophysiological state of the patient's brain.

[0079] Static data provide a baseline status of the patient's nervous system and help identify neurophysiological abnormalities that may lead to spasticity.

[0080] Specifically, the data can be processed and analyzed by commonly used EEG analysis software (such as EEGLAB or FieldTrip) and MRI image processing software (such as SPM or FSL) to generate feature vectors.

[0081] Step S200, collecting dynamic EEG signal data and dynamic electromyogram of the person to be tested in a motion task state, and obtaining a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram;

[0082] The above steps involve collecting dynamic EEG signals and dynamic electromyogram (dynamic EMG data) from patients while they perform specific motor tasks.

[0083] Dynamic electroencephalogram (EEG) signals can also be captured by an electrode head cap, while electromyography (EMG) records the electrical activity of specific muscle groups when they are active. After processing, these data can generate a task-state feature vector that reflects the patient's brain and muscle coordination function when performing a motor task.

[0084] Among them, EMG, Electromyography, is a method of recording the bioelectric signals of muscles when they are at rest or active.

[0085] Dynamic data can more realistically reflect the patient's nerve and muscle responses during daily life activities and provide a basis for assessing spasticity risk in the real world.

[0086] Analysis of dynamic data may use time series analysis techniques combined with machine learning algorithms such as random forests or support vector machines to process and classify large amounts of motion and non-motion state data.

[0087] Step S300, integrating and splicing the resting state feature vector and the task state feature vector, and obtaining a state score through a multi-layer perceptron, and obtaining a prediction result of the person to be tested through the state score.

[0088] As mentioned above, the feature vectors of the resting state and the task state are integrated and spliced, and the synthesized data vector is analyzed using a multi-layer perceptron to generate a state score. This score is based on the output of the deep learning algorithm to predict the risk of spasticity of the person being tested.

[0089] The above status score is calculated based on the model, which may involve a comparison threshold. When the status score exceeds this threshold, the prediction result is positive, which means that the patient has a higher risk of spasm; if the score is lower than the threshold, the prediction result is negative, indicating a lower risk. For example, the value range can be set to 0 to 1 points. The higher the score, the better the patient's status and the lower the probability of PSS. Conversely, the lower the probability of PSS.

[0090] The key to the above steps is how to design and train a multi-layer perceptron model so that it can accurately extract features related to post-stroke limb spasticity from a large amount of complex neurophysiological data and make accurate predictions based on them. This usually requires a large amount of labeled data to train the model, and to optimize the model parameters through methods such as cross-validation to improve the accuracy and generalization of the prediction.

[0091] As mentioned above, Multilayer Perceptron (MLP) is an artificial neural network structure used to handle complex pattern recognition and prediction problems. It is a type of feedforward neural network, which means that the data flow in the network has only one direction from the input to the output layer, with no loop in between.

[0092] Using deep learning models after integrating static and dynamic feature vectors can enhance the accuracy and robustness of predictions and enable learning and prediction from more complex data relationships.

[0093] Specifically, the implementation of multi-layer perceptrons can be done with the help of existing deep learning frameworks such as TensorFlow or PyTorch, which provide the tools and libraries needed to build and train deep neural networks.

[0094] Through this series of steps, the present invention can provide a comprehensive and accurate method for predicting limb spasticity after stroke, which can help medical professionals better understand the patient's spasticity risk and formulate corresponding treatment and management plans accordingly.

[0095] The method for predicting post-stroke limb spasticity provided in this embodiment has the following beneficial effects:

[0096] First, by combining static and dynamic data, this method can more comprehensively capture the patient's neurophysiological state and provide more accurate prediction results. Static data reflects the patient's basic neurophysiological conditions, while dynamic data can reveal the patient's neural response and muscle coordination during actual activities. This comprehensive analysis helps to comprehensively assess the risk and degree of spasticity.

[0097] Secondly, through the application of multi-layer perceptron, this method can effectively extract key features from large amounts of complex data and obtain a more accurate prediction model through learning, thereby improving the accuracy and reliability of the prediction.

[0098] In addition, this method has strong real-time and operability, and can provide clinicians with rapid and accurate spasm risk assessment, which helps to adjust treatment plans in a timely manner and optimize resource allocation, thereby improving treatment efficiency and patients' quality of life.

[0099] In conclusion, this prediction method can provide scientific and effective technical support for the prediction and management of limb spasticity after stroke through highly integrated and intelligent data analysis.

[0100] Embodiment 2:

[0101] Reference Figure 3Embodiment 2 of the present invention provides a method for predicting limb spasticity after stroke. Based on the above embodiment 1, step S100, obtaining a resting state feature vector based on the static EEG signal data and the static MRI data, comprises:

[0102] Step S110, calculating a static EEG brain connection map based on the static EEG signal data; and calculating a static fMRI brain connection map based on the static magnetic resonance imaging data;

[0103] As described above, static EEG data are used to construct a map representing the connections between brain channels, and static functional MRI data are then used to construct a functional connectivity map between brain regions.

[0104] Among them, through step S110, a connection diagram reflecting the connection between EEG channels in a static state and an fMRI brain connection diagram reflecting the functional connection between various brain regions can be obtained.

[0105] Static EEG brain connectivity maps reveal the network connectivity characteristics of the brain in a resting state, which helps to understand the function and structure of brain networks. fMRI brain connectivity maps provide a high-resolution view of the brain's static functional connectivity, which helps to identify brain areas related to diseases.

[0106] Specifically, neuroscience analysis software (such as EEGLAB) can be used to first perform data preprocessing (filtering, artifact removal, etc.), and then perform distance measurement and connection map construction.

[0107] In addition, fMRI brain connectivity maps can be analyzed using fMRI analysis software (such as SPM or FSL) for data preprocessing and connectivity map analysis.

[0108] Step S120, constructing a node-level fusion feature vector from the node granularity of the static EEG brain connection map and the static fMRI brain connection map; and concatenating the static EEG brain connection map and the static fMRI brain connection map through an attention mechanism to obtain a global-level feature vector;

[0109] As mentioned above, the brain connectivity maps of EEG and fMRI are fused at the node level, and the EEG and fMRI features of each brain region are integrated to obtain a node-level feature vector that integrates EEG and fMRI information. In this step, the data of the two modalities are fused to improve the comprehensiveness of the features and the accuracy of the prediction.

[0110] Specifically, it can be done through simple data splicing or using more complex feature fusion algorithms, such as weighted averaging, principal component analysis (PCA), etc.

[0111] As mentioned above, the attention mechanism is used to identify key information in EEG and fMRI data and integrate global features. The attention mechanism can highlight important connection patterns and ignore unimportant information, thereby obtaining a global feature vector containing key information.

[0112] The attention mechanism can improve the model's sensitivity to key features, thereby improving the accuracy of predictions.

[0113] Specifically, an attention network can be designed, for example, using the attention mechanism in a convolutional neural network (CNN) or a recurrent neural network (RNN).

[0114] The complementarity of EEG and fMRI data is utilized to enhance the richness of features and the accuracy of the prediction model.

[0115] As mentioned above, the attention mechanism is a mechanism in deep learning that aims to simulate the human attention mechanism, enabling the model to focus on the most important parts of the input data. This mechanism is very useful in a variety of tasks, such as natural language processing, image recognition, and bioinformatics.

[0116] Step S130, concatenating the node-level fusion feature vector and the global-level feature vector to form the resting-state feature vector.

[0117] In the above, the node-level and global-level feature vectors are combined to form a complete resting-state feature vector. Specifically, the feature vectors obtained in step S120 are concatenated to form a comprehensive feature vector containing local and global information.

[0118] This step provides a feature representation that integrates multi-level information, which helps improve the performance of the prediction model. Specifically, the two vectors can be directly concatenated in sequence, or feature selection and dimensionality reduction techniques can be used to avoid the dimensionality disaster.

[0119] For further reference, Figure 4 The step S110, calculating a static EEG brain connection map based on the static EEG signal data, includes:

[0120] Step S111, calculating the distance between each channel of the static EEG signal according to the Euclidean distance, thereby constructing a distance matrix; the calculation expression of the distance matrix is: ij =||x i -x j || 2 ; Among them, D ij represents the Euclidean distance between channels i and j, x i represents the three-dimensional topological coordinates of channel i, x j Represents the three-dimensional topological coordinates of channel j.

[0121] In the above, Euclidean distance is used to quantify the similarity between EEG signals in different channels. For each channel pair (i, j), the Euclidean distance between them is calculated. The Euclidean distance is the straight-line distance between two points in three-dimensional space, resulting in a distance matrix where each element D ij is the distance between channels i and j.

[0122] As mentioned above, Euclidean distance is an intuitive and easy-to-calculate distance metric that can reflect the spatial proximity between channels.

[0123] Specifically, the Euclidean distance can be calculated using vector operations, such as using the NumPy library in Python.

[0124] Step S112, clustering each channel using a k-nearest neighbor algorithm according to the distance matrix, and constructing the static EEG brain connection map in which each channel is associated with other k channels.

[0125] As mentioned above, K-Nearest Neighbors (KNN) is a basic and intuitive machine learning algorithm for classification and regression problems. The KNN algorithm performs classification or regression by measuring the distance between different feature values. For a new input sample, the algorithm finds the K sample points in the training data set that are closest to it (i.e., K nearest neighbors), and then predicts the category or value of the new sample based on the categories or values ​​of these neighbors.

[0126] The k-NN algorithm is used to cluster according to the Euclidean distance between channels to construct a brain connection map. For each channel i, the k channels closest to it are found and these channels are considered to be connected to channel i, and a connection map is constructed in which each channel is connected to k other channels.

[0127] k-NN clustering is a simple and effective method to reveal local connectivity patterns among channels.

[0128] This can be achieved using the k-NN algorithm in machine learning libraries such as scikit-learn.

[0129] For example, there are 5 EEG channels, each with its coordinates in 3D space (x i ,y i ,z i ). First, calculate the Euclidean distance between each channel pair and build a 5x5 distance matrix. Then, choose k=3, for each channel, find the 3 channels closest to it and mark them as connected to the channel. Finally, a graph describing the connection between these 5 channels is obtained.

[0130] For further reference, Figure 5 The step S110, calculating a static fMRI brain connectivity map based on the static MRI data, comprises:

[0131] Step S113, dividing the brain image in the static MRI data into a plurality of unit brain regions;

[0132] As mentioned above, the fMRI images of the brain are divided into several smaller, relatively independent functional regions, which are called unit brain areas.

[0133] Use image segmentation techniques such as threshold-based segmentation, cluster segmentation, or segmentation methods that use anatomical prior knowledge.

[0134] By dividing the brain into smaller units, we can get a series of well-defined unit brain areas, each of which corresponds to a specific part of the image. Decomposing the brain into smaller units helps to analyze the activity patterns of specific brain areas and the interactions between them.

[0135] Specifically, for example, image segmentation can be performed using image processing software or programming libraries (such as ITK, SimpleITK, or Python's scikit-image).

[0136] Step S114, extracting the time series data of the unit brain area;

[0137] As described above, the time series data of BOLD (blood oxygen level dependent) signals are extracted from each unit brain region, and the signals of the fMRI image sequence of each unit brain region are extracted at time points, thereby obtaining the BOLD signals of each unit brain region that change with time.

[0138] Time series data can reflect the dynamic changes in brain activity and provide a basis for further analysis of interactions between brain regions.

[0139] To achieve signal extraction through programming, it may be necessary to preprocess the raw fMRI data, such as denoising and correction.

[0140] Step S115, calculating the Pearson correlation coefficient of the time series data of any two of the unit brain regions; the calculation expression of the Pearson correlation coefficient is: Among them, ρ X,Y represents the correlation coefficient of the time series X and Y of the two unit brain regions, cov(X,Y) is the covariance of the two time series, σ X and σ Y Represents the standard deviation of X and Y series respectively; μ X and μ Y Represent the means of the X and Y series respectively.

[0141] As mentioned above, the Pearson correlation coefficient is used to quantify the linear relationship between the time series data of any two unit brain regions, thereby obtaining a correlation coefficient matrix representing the correlation between different unit brain regions.

[0142] The Pearson correlation coefficient is a standard metric that quantifies the synchrony between two signals. It can be calculated using statistical packages or library functions in programming languages, such as Python's NumPy library.

[0143] Step S116, filtering the connection relationship between different unit brain regions through a preset threshold to obtain the static fMRI brain connection map.

[0144] The functional connectivity matrix is ​​thresholded to screen the connectivity of brain regions, thereby obtaining an fMRI brain connectivity map. A preset correlation coefficient threshold is used to determine whether the connection between unit brain regions is significant. Each value in the correlation coefficient matrix is ​​checked, and connections above the threshold are considered significant, thereby obtaining a simplified brain connectivity map containing only significant connections.

[0145] Threshold screening can remove noise and unimportant connections and highlight important brain network structures.

[0146] Specifically, the threshold screening logic is implemented in programming, and the threshold can be determined in combination with statistical tests.

[0147] For further reference, Figure 6 The step S120, forming a node-level fusion feature vector from the static EEG brain connection map and the static fMRI brain connection map at the node granularity, comprises:

[0148] Step S121, arranging the nodes of the static EEG brain connection map and the static fMRI brain connection map in rows, respectively, and splicing them to obtain a representation matrix corresponding to the static EEG brain connection map and a representation matrix corresponding to the static fMRI brain connection map;

[0149] As mentioned above, in terms of node granularity, the node data of the EEG brain connection map and the fMRI brain connection map are arranged in row order, and the feature vectors of each are concatenated to form two matrices, which are the representation matrix corresponding to the static EEG brain connection map and the representation matrix corresponding to the static fMRI brain connection map.

[0150] This step involves converting the brain connectivity map data from EEG and fMRI into a format that can be further analyzed.

[0151] In the above steps, the node data of the EEG brain connection map and the fMRI brain connection map are arranged in row order to form row vectors, and then these row vectors are concatenated to form two new matrices corresponding to the EEG and fMRI data respectively.

[0152] In this step, two representation matrices are obtained respectively, the number of rows of each matrix is ​​equal to the number of nodes of the original connection graph, and the number of columns is equal to the number of nodes of the original connection graph multiplied by the number of nodes (because it is the expansion of the NxN matrix).

[0153] This processing method converts graph data into matrix data, which is convenient for statistical analysis and subsequent machine learning processing.

[0154] Specifically, it can be implemented using array or matrix operations in a programming language, such as using Python's NumPy library.

[0155] Step S122, transposing the representation matrix corresponding to the static EEG brain connection map and the representation matrix corresponding to the static fMRI brain connection map respectively, and performing matrix multiplication processing to obtain a corresponding node-level fusion matrix;

[0156] Step S123, counting the distribution of values ​​in each of the node-level fusion matrices in the form of a histogram, and constructing the node-level fusion feature vector through the frequency of each group of the histogram.

[0157] As mentioned above, the transpose of the EEG image representation matrix and the fMRI image representation matrix is ​​multiplied to obtain a node-level fusion matrix, and the distribution of the values ​​in the matrix is ​​statistically analyzed in the form of a histogram, and the frequency of each group of the histogram constitutes a node-level fusion feature vector.

[0158] First, the node data of the EEG brain connection map and the fMRI brain connection map are arranged in rows. For each connection map, its N×N matrix (N is the number of nodes) is converted into N row vectors, each row vector contains all the connection values ​​of a node, thus obtaining two matrices, respectively called the EEG map representation matrix and the fMRI map representation matrix, each matrix has the same number of rows and columns as the number of nodes.

[0159] The two representation matrices are transposed. The transposition operation is to swap the rows and columns of the matrix, that is, the i-th row and j-th column of the original matrix becomes the j-th row and i-th column in the transposed matrix.

[0160] The rows of the transposed matrix represent the columns of the original matrix, that is, all the connection values ​​of each node now become column vectors.

[0161] The multiplication operation is performed by transposing the step matrix. Since both matrices are N×N, their multiplication is feasible and the result is an N×N node-level fusion matrix.

[0162] Each element in the node-level fusion matrix represents a certain comprehensive measure of EEG and fMRI data at the corresponding node.

[0163] The obtained node-level fusion matrix is ​​statistically analyzed for numerical distribution. All the values ​​in the matrix are divided into several intervals (buckets), and the number of values ​​in each interval is counted to form a histogram.

[0164] The frequencies of each group of the obtained histogram are used as elements of the feature vector. This means that the frequency corresponding to each bucket of the histogram is a dimension of the feature vector. In this step, a feature vector is obtained, whose length is equal to the number of buckets of the histogram, and the value of each dimension is the frequency of the corresponding numerical interval.

[0165] The above method allows the fusion of two different neuroimaging data, which may reveal patterns or features that cannot be discovered by a single data source. Through the histogram, the original node-level connection data is compressed into a more compact feature vector for subsequent processing. The histogram provides statistical characteristics of the data distribution, which helps to understand the central tendency and dispersion of the data.

[0166] For further reference, Figure 7 The step S120, wherein the static EEG brain connection map and the static fMRI brain connection map are concatenated through an attention mechanism to obtain a global feature vector, comprises:

[0167] Step S124, inputting the representation matrix into the corresponding attention mechanism to select the top-k nodes;

[0168] As mentioned above, the representation matrices of EEG and fMRI are used as input, and the attention mechanism is used to identify the most important nodes in each matrix. The attention mechanism is used to assign a weight to each node to indicate its importance in the overall network.

[0169] Select the k nodes with the highest weights, which are considered to be the most critical in the current task.

[0170] Specifically, the attention mechanism can highlight the most important features in the data, which helps improve the accuracy and efficiency of subsequent analysis.

[0171] This can be implemented using the attention module in deep learning frameworks such as TensorFlow or PyTorch.

[0172] Step S125, taking the average of the feature vectors of all the top-k nodes, and respectively obtaining the global vector representation information of the static EEG brain connection map and the static fMRI brain connection map;

[0173] As mentioned above, the average feature vectors of the top-k nodes in the EEG and fMRI representation matrices are calculated.

[0174] Step S126, concatenating the vector evidence information corresponding to the static EEG brain connection map and the static fMRI brain connection map to form the global level feature vector.

[0175] In the above steps, global feature fusion processing is performed.

[0176] Specifically, the representation matrix of the graph is sent to each attention module to select the top-k nodes, and the feature vectors of these nodes are averaged to obtain the global vector representation information of the brain connection of the two neuroimaging images, and the two vector representation information are concatenated to form a global feature vector.

[0177] The advantage of the above method is that it can comprehensively consider the data of two different modalities and highlight the most important features through the attention mechanism, which helps to improve the performance of subsequent analysis or prediction tasks. However, choosing a suitable k value and implementing the attention mechanism may need to be adjusted according to the specific task and dataset.

[0178] Embodiment 3:

[0179] Reference Figure 8 Embodiment 3 of the present invention provides a method for predicting limb spasticity after stroke. Based on the above embodiment 1, the step S200, obtaining a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram, includes:

[0180] Step S210, performing wavelet decomposition on the dynamic EEG signal data to obtain a dynamic EEG feature vector; and obtaining an energy value feature according to the dynamic EEG signal data;

[0181] It should be noted that when a person performs an actual action, the ERS / ERD phenomenon will occur. In this embodiment, after obtaining the dynamic EEG signal (EEG signal) of the person to be tested in the state of the motion task, the wavelet decomposition method is used to construct the feature vector, and the intention feature information is obtained in the frequency band related to the motion. Specifically, the wavelet decomposition technology is used to process the dynamic EEG signal data, extract the feature vector, and obtain the energy value feature.

[0182] The dynamic EEG signal is decomposed into components of different frequencies through wavelet decomposition, from which the eigenvector is extracted; at the same time, the energy value of the signal in each frequency band is calculated to obtain the eigenvector and related energy value describing the dynamic EEG activity.

[0183] As mentioned above, wavelet decomposition can effectively analyze signals in both time and frequency dimensions, helps to identify frequency changes within a local time period, and is particularly suitable for non-stationary signal analysis.

[0184] For example, you can use the PyWavelets library in Python to perform wavelet transforms and extract frequency domain and energy features.

[0185] For example, perform k-level wavelet decomposition on the signal, and suppose k1, k2, …, kn are n EEG signal sub-bands related to lower limb movement. Then take the average of these n finite-length sequences to obtain the mean of the n coefficients X1, X2, …Xn.

[0186] In addition to the coefficient characteristics, the energy value can also reflect the characteristic information of the EEG signal to a certain extent. For example, the energy value of the sub-band is defined as the square sum of the amplitudes of the sampling points. The sampling time of the EEG signal band k1, k2, ..., kn related to lower limb movement is ts, and the corresponding unit time energy value is P1, P2, ..., Pn.

[0187] Step S220, combining the dynamic EEG feature vector and the energy value feature to obtain a dynamic feature vector;

[0188] Assume that the EEG data has M channels, and each channel selects n relevant sub-frequency bands after decomposition. Then, combining the wavelet decomposition coefficients and energy value features, a feature vector containing 2×M×n eigenvalues ​​is constructed.

[0189] Step S230, extracting dynamic sEMG features according to the dynamic electromyogram;

[0190] It should be noted that the frequency of electromyographic signals is usually between 10 and 500 Hz.

[0191] In this step, the collected original signal, i.e., the dynamic electromyogram, can be processed as follows:

[0192] (1) Perform 50Hz notch processing on dynamic electromyography;

[0193] (2) Use wavelet threshold denoising method to remove noise.

[0194] (3) Extract EMG features from different angles such as time domain, frequency domain, time-frequency domain and chaotic characteristics to obtain dynamic sEMG features, so as to further combine multiple features to analyze the signal.

[0195] Step S240, reducing the dimension of the energy value feature and the dynamic sEMG feature through a support vector machine to construct a weight matrix;

[0196] As mentioned above, the feature vectors of dynamic EEG and sEMG are processed using the support vector machine (SVM) algorithm to reduce the dimension and construct the weight matrix. The SVM algorithm is applied to reduce the dimension of the features, optimize the feature space, and extract the most effective features.

[0197] The above,weight matrix, describes the importance of each feature in classification.

[0198] As mentioned above, SVM (Support Vector Machine) is a powerful classifier that can effectively handle high-dimensional data and avoid overfitting by maximizing the margin of the decision boundary.

[0199] Specifically, the scikit-learn library in Python can be used to implement SVM model training and weight extraction.

[0200] Step S250, based on the weight matrix, classify the dynamic EEG signal data and the dynamic electromyogram through the classification model of the support vector machine to obtain a classification result matrix;

[0201] As mentioned above, after obtaining the weight matrix, the weight matrix here is obtained during the support vector machine (SVM) training process, which is learned based on the training data (i.e., the previously collected EEG and sEMG data). The purpose of the weight matrix is ​​to optimize and determine the importance and contribution of each feature when the model processes the feature.

[0202] The classification model (here SVM model) uses a weight matrix as the basis for determining the influence of each feature. The model is trained to distinguish between different stimulus effects or motion states, such as identifying specific types of limb movements or reactions.

[0203] Based on the weight matrix, the dynamic EEG signal data and the dynamic electromyogram are classified by the support vector machine (SVM) classification model. Here, the weight matrix is ​​learned by the SVM model during the training process and is used to adjust the model's emphasis on each feature during classification, thereby optimizing the classification decision. The dynamic EEG signal data and the dynamic electromyogram data are used as input, processed by the SVM classification model, and output as specific classification results. These results help identify and analyze the brain and muscle activity patterns of patients when performing specific motor tasks. This method helps to better understand and predict limb spasticity after stroke in medical diagnosis and treatment planning.

[0204] The obtained test data (the dynamic EEG signal data and the dynamic EMG) can first be preprocessed, time aligned and feature extracted, and then respectively input into the model for classification, and the classification results are represented in the form of column vectors. If the prediction is the i-th stimulation effect, the i-th row is set to 1, and the remaining rows are set to 0.

[0205] Step S260, performing a dot multiplication of the classification result matrix and the weight matrix to obtain a weighted classification result vector;

[0206] In the above, the classification result is multiplied by the weight matrix to calculate the weighted classification result. Through matrix multiplication, the classification result matrix is ​​multiplied by the weight matrix to obtain the weighted result.

[0207] The weighted classification result vector provides the classification output adjusted according to the feature importance.

[0208] The weighting process takes into account the influence of each feature and improves the reliability of the results.

[0209] For example, all model column vectors are combined into the classification result matrix R ij , and then multiply the weight matrix by the classification result matrix.

[0210] Step S270, obtaining MEP detection information, and extracting a MEP feature vector corresponding to the MEP detection information;

[0211] As mentioned above, feature vectors are extracted from muscle evoked potential (MEP) detection data. MEP signals are analyzed to extract relevant bioelectric features.

[0212] Step S280, concatenating the weighted classification result vector, the MEP feature vector, the dynamic feature vector and the dynamic sEMG feature to obtain the task state feature vector.

[0213] All extracted feature vectors (weighted classification results, MEP features, dynamic EEG and sEMG features) are merged into a task-state feature vector.

[0214] As mentioned above, the three new fused feature vectors obtained are concatenated through the MEP, EEG response delay and electromyography input model to construct the task state vector representation.

[0215] In addition, reference Fig. 9 This embodiment also provides a prediction device for post-stroke limb spasticity, comprising:

[0216] The static characterization module 10 is used to collect static EEG signal data and static MRI data of the person to be tested in a resting state, and obtain a resting state feature vector based on the static EEG signal data and the static MRI data;

[0217] A dynamic characterization module 20 is used to collect dynamic EEG signal data and dynamic electromyogram of the person to be tested in a motion task state, and obtain a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram;

[0218] The evaluation prediction module 30 is used to integrate and splice the resting state feature vector and the task state feature vector, and obtain a state score through a multi-layer perceptron, and obtain a prediction result of the person to be tested through the state score.

[0219] In addition, an embodiment of the present application also provides a prediction system for limb spasm after stroke, including a memory and a processor, wherein the memory stores a prediction program for limb spasm after stroke, and the processor runs the prediction program for limb spasm after stroke so that the prediction system for limb spasm after stroke executes the prediction method for limb spasm after stroke as described above.

[0220] In addition, a computer-readable storage medium is provided in an embodiment of the present application, on which a prediction program for limb spasm after stroke is stored. When the prediction program for limb spasm after stroke is executed by a processor, the prediction method for limb spasm after stroke as described above is implemented.

[0221] In short, the post-stroke limb spasticity prediction method provided in the embodiment of the present application captures the patient's comprehensive neurophysiological state by combining static and dynamic data, extracts key features from complex data using a multilayer perceptron, and constructs an accurate prediction model. This method not only improves the accuracy and reliability of the prediction, but also has real-time and operability, can provide fast and accurate spasm risk assessment for clinicians, help timely adjust treatment plans and optimize resource allocation, significantly improve treatment efficiency and patient quality of life, and provide scientific and effective technical support for the prediction and management of post-stroke limb spasticity.

[0222] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, including a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention. The above is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly used in other related technical fields, are similarly included in the patent protection scope of the present invention.

Claims

1. A method for predicting limb spasticity after stroke, characterized in that: include: Collecting static EEG signal data and static MRI data of the person to be tested in a resting state, and obtaining a resting state feature vector based on the static EEG signal data and the static MRI data; Collecting dynamic EEG signal data and dynamic electromyogram of the person to be tested in a motion task state, and obtaining a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram; The resting state feature vector and the task state feature vector are integrated and spliced, and a state score is obtained through a multi-layer perceptron, and a prediction result of the person to be tested is obtained through the state score.

2. The method for predicting post-stroke limb spasticity according to claim 1, characterized in that: The step of obtaining a resting state feature vector based on the static EEG signal data and the static MRI data includes: Calculating a static EEG brain connection map based on the static EEG signal data; and calculating a static fMRI brain connection map based on the static magnetic resonance imaging data; The static EEG brain connection map and the static fMRI brain connection map are used to form a node-level fusion feature vector from a node granularity; and the static EEG brain connection map and the static fMRI brain connection map are spliced ​​through an attention mechanism to obtain a global-level feature vector; The node-level fusion feature vector and the global-level feature vector are concatenated to form the resting-state feature vector.

3. The method for predicting post-stroke limb spasticity as claimed in claim 2, characterized in that: The step of calculating a static EEG brain connection diagram based on the static EEG signal data includes: The distances between the channels of the static EEG signal are calculated according to the Euclidean distance, thereby constructing a distance matrix; the calculation expression of the distance matrix is: ij =||x i -x j || 2 ; Among them, D ij represents the Euclidean distance between channels i and j, x i represents the three-dimensional topological coordinates of channel i, x j represents the three-dimensional topological coordinates of channel j; According to the distance matrix, each channel is clustered using the k-nearest neighbor algorithm to construct the static EEG brain connection map in which each channel is associated with other k channels.

4. The method for predicting post-stroke limb spasticity according to claim 2, characterized in that: The step of calculating a static fMRI brain connectivity map based on the static nuclear magnetic resonance imaging data comprises: Dividing the brain image in the static magnetic resonance imaging data into a plurality of unit brain regions; extracting time series data of the unit brain area; Calculate the Pearson correlation coefficient of the time series data of any two unit brain regions; the calculation expression of the Pearson correlation coefficient is: Among them, ρ X,Y represents the correlation coefficient of the time series X and Y of the two unit brain regions, cov(X,Y) is the covariance of the two time series, σ X and σ Y Represents the standard deviation of X and Y series respectively; μ X and μ Y Represent the means of X and Y series respectively; The connection relationship between different unit brain regions is screened by a preset threshold to obtain the static fMRI brain connection map.

5. The method for predicting post-stroke limb spasticity as claimed in claim 2, characterized in that: The step of forming a node-level fusion feature vector from the static EEG brain connection map and the static fMRI brain connection map at a node granularity includes: Arrange the nodes of the static EEG brain connection map and the static fMRI brain connection map in rows, respectively, and concatenate them to obtain a representation matrix corresponding to the static EEG brain connection map and a representation matrix corresponding to the static fMRI brain connection map; Transposing the representation matrix corresponding to the static EEG brain connection map and the representation matrix corresponding to the static fMRI brain connection map respectively, and performing matrix multiplication processing to obtain a corresponding node-level fusion matrix; The numerical distribution in each node-level fusion matrix is ​​counted in the form of a histogram, and the node-level fusion feature vector is formed by the frequency of each group of the histogram.

6. The method for predicting post-stroke limb spasticity according to claim 5, characterized in that: The method of obtaining a global feature vector by splicing the static EEG brain connection map and the static fMRI brain connection map through an attention mechanism includes: Input the representation matrices into the corresponding attention mechanisms to select the top-k nodes; Taking the average of the feature vectors of all the top-k nodes, and respectively obtaining the global vector representation information of the static EEG brain connection map and the static fMRI brain connection map; The vector evidence information corresponding to the static EEG brain connection map and the static fMRI brain connection map is spliced ​​to form the global level feature vector.

7. The method for predicting post-stroke limb spasticity according to claim 1, characterized in that: The step of obtaining a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram comprises: The dynamic EEG signal data is subjected to wavelet decomposition to construct a dynamic EEG feature vector; and an energy value feature is obtained according to the dynamic EEG signal data; Combining the dynamic EEG feature vector and the energy value feature to obtain a dynamic feature vector; Extracting dynamic sEMG features according to the dynamic electromyogram; The energy value feature and the dynamic sEMG feature are reduced in dimension by a support vector machine to construct a weight matrix; Based on the weight matrix, the dynamic EEG signal data and the dynamic electromyogram are classified by the classification model of the support vector machine to obtain a classification result matrix; Multiply the classification result matrix by the weight matrix to obtain a weighted classification result vector; Acquire MEP detection information, and extract MEP feature vectors corresponding to the MEP detection information; The weighted classification result vector, the MEP feature vector, the dynamic feature vector and the dynamic sEMG feature are concatenated to obtain the task state feature vector.

8. A device for predicting limb spasticity after stroke, characterized in that: include: A static characterization module, used to collect static EEG signal data and static MRI data of the person to be tested in a resting state, and obtain a resting state feature vector based on the static EEG signal data and the static MRI data; A dynamic characterization module, used for collecting dynamic EEG signal data and dynamic electromyogram of the person to be tested in a motion task state, and obtaining a task state feature vector according to the dynamic EEG signal data and the dynamic electromyogram; The evaluation prediction module is used to integrate and splice the resting state feature vector and the task state feature vector, and obtain a state score through a multi-layer perceptron, and obtain a prediction result of the person to be tested through the state score.

9. A prediction system for limb spasticity after stroke, characterized in that: It comprises a memory and a processor, wherein the memory stores a prediction program for limb spasticity after stroke, and the processor runs the prediction program for limb spasticity after stroke so that the prediction system for limb spasticity after stroke executes the prediction method for limb spasticity after stroke as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a prediction program for limb spasticity after stroke, and when the prediction program for limb spasticity after stroke is executed by a processor, the prediction method for limb spasticity after stroke as described in any one of claims 1-7 is implemented.

Images