A method for monitoring muscle spasms and related health indicators

Abstract

The present invention discloses a method for monitoring muscle spasms and related health indicators, which belongs to the field of health monitoring methods, and includes the following steps: collecting EMG signals and classifying and processing them to obtain EMG classification data; collecting one-dimensional ultrasonic signals and performing data segmentation on them to obtain ultrasonic fine segmentation data; inputting the EMG classification data and the ultrasonic fine segmentation data into a fusion model; realizing the alignment and fusion of multimodal features in a unified feature space through a multimodal feature space alignment encoder; feature alignment and calibration, so that the features of different modes are comparable and consistent in the same space, so as to perform effective fusion and analysis; fusing the calibrated and aligned EMG signals and ultrasonic data features through a multimodal fusion module, using displacement mapping to rotate, scale and fuse features, and capturing shared and unique features between different modes. The present invention can accurately monitor muscle spasms and related health indicators, and thus timely warn of chronic strain.

Description

Technical Field

[0001] The present invention relates to the field of health monitoring methods, in particular to a method for monitoring muscle spasms and related health indicators. Background Art

[0002] The way people live and work has changed dramatically, and the life expectancy has been greatly extended, which has led to a huge change in the spectrum of human diseases, and a significant increase in the incidence of strain diseases and degenerative diseases. However, these strains and degenerations are all happening without our knowledge, especially the strain of our spine. By the time obvious symptoms appear, the strain is already very serious and has led to the onset of the disease. By then, many lesions are irreversible. Therefore, it is particularly important to judge and warn of our chronic strains, as well as to judge the effectiveness of many treatment methods.

[0003] However, there is no instrument on the market that can detect muscle strain, let alone early warning devices or muscle efficacy evaluation devices. Our strain first manifests itself in muscle ischemia and spasm, and decreased muscle elasticity. Therefore, how to accurately monitor muscle spasm and related health indicators has become a problem that needs to be solved urgently.

[0004] Therefore, those skilled in the art provide a method for monitoring muscle spasms and related health indicators to solve the problems raised in the above background technology. Summary of the invention

[0005] The purpose of the present invention is to provide a method for monitoring muscle spasms and related health indicators, which can accurately monitor muscle spasms and related health indicators, and then timely warn of chronic strain to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for monitoring muscle spasms and related health indicators, comprising the following steps:

[0008] Collecting EMG signals and classifying and processing them to obtain EMG classification data;

[0009] Collecting one-dimensional ultrasonic signals and performing data segmentation on them to obtain ultrasonic fine segmentation data;

[0010] Input EMG classification data and ultrasound fine segmentation data into the fusion model;

[0011] The alignment and fusion of multimodal features in a unified feature space are achieved through a multimodal feature space alignment encoder;

[0012] Feature alignment and calibration make the features of different modalities comparable and consistent in the same space for effective fusion and analysis;

[0013] The calibrated and aligned EMG signal and ultrasound data features are fused through the multimodal fusion module, and displacement mapping is used to rotate, scale and fuse features to capture shared and unique features between different modalities;

[0014] Perform multiple loop iterations;

[0015] Output the fusion and matching results of EMG data and ultrasound data for more accurate monitoring of muscle spasms and collaborative analysis of multi-source data;

[0016] A mechanism to ensure that data from different modalities are aligned in feature space by spatially aligning feature vectors.

[0017] As a further solution of the present invention: the classification and processing process of the EMG signal is specifically as follows:

[0018] Model pre-training, i.e. training the EMG model using datasets from multiple subjects;

[0019] EMG signal generation, i.e. generating synthetic data from EMG signals of unknown states and symptoms through an autoregressive conditional generation process;

[0020] EMG classifier training, that is, using real and synthetic EMG datasets to train the classifier to improve its classification accuracy;

[0021] The EMG signal is classified and processed by the trained classifier to obtain EMG classification data.

[0022] As a further solution of the present invention: the specific process of data segmentation of the one-dimensional ultrasonic signal is:

[0023] Input signal, i.e., input one-dimensional ultrasonic signal;

[0024] Primary U-Net, that is, passing the signal through a primary U-Net structure;

[0025] Focus on region sampling, where the output of the primary U-Net is used to generate candidate regions, i.e., sampling regions;

[0026] Region of interest segmentation, i.e., sampling based on the region of interest is used to crop the original signal;

[0027] The second level U-Net passes the cropped signal through another more refined U-Net;

[0028] Output segmentation, i.e., by performing a rough segmentation based on sampling of the region of interest, the output of the secondary U-Net is used for fine segmentation to predict the peak position of muscle spasm;

[0029] Classification, that is, the lowest layer (64) of the first-level U-Net is flattened and classified through the fully connected layer (1024→128→x).

[0030] As a further solution of the present invention: the U-Net includes a downsampling path (4→8→16→32→64) and an upsampling path (32→16→8→4), and each level uses double 1D convolution.

[0031] As a further solution of the present invention: the dual 1D convolution includes convolution, batch normalization and ELU activation.

[0032] As a further solution of the present invention: the layers of the U-Net are down-sampled by maximum pooling and up-sampled by convolution, and the corresponding down-sampling and up-sampling layers are connected using attention blocks.

[0033] As a further solution of the present invention: the EMG signal is collected by a monitor that is continuously attached to the surface of the skin close to the muscle.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] This application can continuously and in real time monitor muscle strain degenerative diseases, and can provide real-time analysis decisions and alarms to reduce or even eliminate strain degenerative diseases. It allows users to gradually develop work, life and exercise habits that do not cause further strain and degeneration, thereby completely getting rid of similar equipment and the troubles of strain and degeneration. At the same time, it can also provide the same functions of other related health indicators and can optimize the combination. In addition, data from different modalities (EMG and ultrasound) are effectively fused and matched for medical diagnosis, human-computer interface or other application scenarios that require collaborative analysis of multi-source data. The complexity of the architecture shows that it can process high-dimensional, heterogeneous data, and extract useful comprehensive features through multi-step processing, thereby effectively improving the accuracy of monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flowchart of a method for monitoring muscle cramps and related health indicators;

[0037] Figure 2 A schematic diagram of the classification and processing process of EMG signals in a method for monitoring muscle spasms and related health indicators;

[0038] Figure 3A schematic diagram of the autoregressive generation process in a method for monitoring muscle spasms and related health indicators;

[0039] Figure 4 A schematic diagram of an EMG generation model in a method for monitoring muscle spasms and related health indicators;

[0040] Figure 5 Schematic diagram of data segmentation of one-dimensional ultrasound signals in a method for monitoring muscle spasms and related health indicators. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] As mentioned in the background technology of this application, research has found that there is currently no instrument on the market that can specifically detect muscle strain, nor any early warning device, nor any therapeutic efficacy evaluation device for muscles. Our strain first manifests itself in ischemic spasms of our muscles and decreased muscle elasticity. How to accurately monitor muscle spasms and related health indicators has become a defect that urgently needs to be solved.

[0043] In order to solve the above-mentioned defects, the present application discloses a method for monitoring muscle spasms and related health indicators, which can accurately monitor muscle spasms and related health indicators, and then timely warn of chronic strain.

[0044] The following will describe in detail how the solution of the present application solves the above technical problems in conjunction with the accompanying drawings.

[0045] See also Figure 1 In an embodiment of the present invention, a method for monitoring muscle spasms and related health indicators includes the following steps: collecting EMG (electromyography) signals and classifying and processing them to obtain EMG classification data; collecting one-dimensional ultrasound signals and performing data segmentation on them to obtain ultrasound fine segmentation data; inputting the EMG classification data and the ultrasound fine segmentation data into a fusion model; aligning and fusion of multimodal features in a unified feature space through a multimodal feature space alignment encoder, Figure 1Two parallel processing flows are shown in , representing different processing layers or modules. These modules include layer normalization + feature mapping + feature integration, attention module, convolution + activation + feature mapping (multiple loop iterations), feature integration, etc., which are used to extract and process features respectively; feature alignment and calibration, so that the features of different modalities (such as images, text, audio, etc.) are comparable and consistent in the same space for effective fusion and analysis; the calibrated and aligned EMG signal and ultrasound data features are fused through the multimodal fusion module, and the displacement mapping is used to rotate, scale and fuse the features to capture the shared and unique features between different modalities; multiple loop iterations are performed; the fusion and matching results of EMG data and ultrasound data are output for more accurate monitoring of muscle spasms and collaborative analysis of multi-source data; a mechanism to ensure that different modal data are aligned in the feature space by spatially aligning feature vectors. This application effectively fuses and matches data from different modalities (EMG and ultrasound) for medical diagnosis, human-computer interface or other application scenarios that require collaborative analysis of multi-source data. The complexity of the architecture shows that it can handle high-dimensional, heterogeneous data and extract useful comprehensive features through multi-step processing.

[0046] In this embodiment, the classification process of the EMG signal is specifically as follows: model pre-training, that is, using data sets from multiple subjects to train the EMG model, wherein each subject has multiple abnormal states, each abnormal state includes multiple abnormal symptoms, and each abnormal symptom has a different number of EMG signal collection devices, Figure 2 The waveforms of EMG signals collected at different positions are shown in FIG. Figure 3 and Figure 4 As shown, synthetic data is generated from EMG signals of unknown states and symptoms through an autoregressive conditional generation process. The new EMG data is synthesized based on existing samples, wherein a large amount of synthetic data is generated based on a small amount of new data using a trained generation model to expand the training data set; EMG classifier training, that is, using real and synthetic EMG data sets to train the classifier to improve its classification accuracy; the EMG signal is classified and processed by the trained classifier to obtain EMG classification data. It should be noted that the EMG signal is a muscle signal, and an autoregressive generation model is used when generating synthetic data. The autoregressive generation model is shown in Figure 3As shown in the figure, the autoregressive generative model is a generative model based on probability theory, which generates new data sequences by learning the conditional probability distribution in the data sequence. This model is widely used in many fields, including time series analysis, natural language processing, image generation, and music generation. The basic idea of ​​the autoregressive generative model is to use historical information in the data sequence to predict future data points. In statistics, the autoregressive model (AR model) is a method for processing time series data. It uses the previous periods of the same variable (such as x1 to xt-1) to predict the performance of the current period (xt), and assumes a linear relationship between them. This model has been extended to a wider range of generative models to generate new data sequences with similar statistical properties to the original data.

[0047] In this embodiment, if Figure 5As shown in the figure, the specific process of data segmentation of one-dimensional ultrasound signal is as follows: input signal, that is, input one-dimensional ultrasound signal; primary U-Net, that is, the signal passes through a primary U-Net structure; region of interest sampling, wherein the output of the primary U-Net is used to generate a candidate region (512 to 612), and the signal in this region is sampled to form a sampling region; region of interest segmentation, that is, based on the region of interest sampling, the original signal is cropped; secondary U-Net, the cropped signal passes through another finer U-Net (similar in structure to the primary U-Net); output segmentation, that is, through rough segmentation based on region of interest sampling, the output of the secondary U-Net is used for fine segmentation to predict the peak position of muscle spasm; classification, that is, the lowest layer (64) of the primary U-Net is flattened and classified through a fully connected layer (1024→128→x) to facilitate understanding of the overall context of the signal and avoid affecting subsequent processing steps, such as the strategy adjustment of muscle spasm peak detection, the number and index of output channels, so as to guide the system to focus on the most relevant or important signal channels, which helps to reduce the impact of noise and improve the accuracy of subsequent processing. The whole pipeline is designed to handle signal classification and muscle spasm peak position prediction tasks simultaneously. Two U-Nets provide perception domains of different scales and are connected by a sampling method based on the attention region, realizing a coarse-to-fine processing strategy. Among them, for the focus area sampling stage, the purpose of this stage is to narrow the focus area, improve processing efficiency and accuracy, provide more concentrated input for the primary U-Net, and focus on areas that may contain features of interest (such as muscle spasm peaks); the working principle of this stage is to use the output of the primary U-Net to generate candidate regions (shown in the figure as a range of 512 to 612), sample within this candidate area, create a sampling area, calculate the peak probability, and determine the location that is most likely to contain the target feature; the functions of this stage are preliminary positioning (preliminary determination of the most relevant part of the signal), dimensionality reduction (through selective sampling, reducing the amount of data that needs to be finely processed), and guiding attention (providing a key focus area for subsequent fine processing). In addition, based on the results of the focus area sampling, the original signal is cropped, and the cropped signal is sent to the secondary U-Net for more detailed analysis, so as to improve efficiency (avoid fine processing of the entire signal and save computing resources), enhance accuracy (by focusing on the most relevant area, improving the accuracy of subsequent processing), and improve adaptability (being able to dynamically adjust the focus area according to the characteristics of different input signals). Finally, multi-scale analysis is achieved by combining the coarse and fine stages to realize multi-scale feature extraction from global to local, which effectively bridges the primary and secondary processing stages and optimizes the performance and efficiency of the entire network. It is particularly suitable for processing tasks that require precise positioning such as muscle spasm detection.It should be noted that U-Net is an end-to-end image segmentation model based on convolutional neural network (CNN). The model is famous for its unique U-shaped structure and excellent performance on image segmentation tasks. It is particularly suitable for the field of biomedical image segmentation, but it has also been widely used in other image segmentation tasks. The overall structure of U-Net is similar to the capital letter U, hence the name U-Net. This structure enables the network to learn both low-level and high-level features of the image at the same time. Skip connections play an important role in U-Net, as they help capture multi-scale information and improve the accuracy of image segmentation. Skip connections combine low-level features in the encoder with high-level features in the decoder, allowing the network to better restore detail information in the image. U-Net has no fully connected layers, which allows the network to accept input images of any size and output segmentation results of the same size as the input image. Due to its simple and effective structure, U-Net is able to achieve faster inference speed while maintaining high segmentation accuracy.

[0048] In this embodiment, U-Net includes a downsampling path (4→8→16→32→64) and an upsampling path (32→16→8→4), and each level uses double 1D convolution. Among them, the downsampling path is mainly responsible for extracting features from the input image and gradually reducing the spatial size (i.e., width and height) of the feature map while increasing the depth (i.e., number of channels) of the feature map. The upsampling path is responsible for restoring the feature map extracted in the downsampling path to the size of the original image and gradually refining the segmentation result in the process. The numbers (4→8→16→32→64) and (32→16→8→4) refer to the depth (i.e., number of channels) of the feature map, not the width or height of the feature map.

[0049] In this embodiment, the double 1D convolution includes convolution, batch normalization and ELU (exponential linear unit) activation. This setting means that two one-dimensional convolution operations are applied consecutively in a neural network layer, and batch normalization and ELU (Exponential Linear Units) activation functions are applied after each convolution operation. This configuration can enhance the feature extraction capability of the network while improving the stability and efficiency of training. The double 1D convolution (including convolution, batch normalization and ELU activation) can build a powerful neural network layer for processing one-dimensional data and extracting useful features. Among them, convolution is an important operation in analytical mathematics. It is a mathematical operation that generates a third function through two functions f and g. Its essence is a special integral transformation, which represents the integral of the overlapping part function value product of the function f and g after flipping and translation over the overlapping length. In deep learning, especially in convolutional neural networks (CNNs), convolution operations are often used to extract local features of input data. Batch normalization is a commonly used technique in the training process of neural networks, which aims to solve the problem of internal covariate shift, that is, the input distribution of each layer changes continuously during the training process, causing the network to adapt continuously. Batch normalization accelerates the training of neural networks by normalizing the input of each layer to keep it in a relatively stable distribution. ELU is a nonlinear activation function used to introduce nonlinear characteristics into neural networks. The ELU function performs better when processing negative values ​​and can effectively solve the "dead neuron" problem of the ReLU (rectified linear unit) function when processing negative values.

[0050] In this embodiment, the layers of U-Net are downsampled by maximum pooling and upsampled by upconvolution, and the corresponding downsampling and upsampling layers are connected by attention blocks. Maximum pooling is a type of pooling operation, which achieves downsampling by taking the maximum value of the local area of ​​the input feature map. Maximum pooling can reduce the size of the feature map and reduce the spatial dimension of the data, thereby reducing the amount of calculation and the amount of parameters. At the same time, it can also extract the most significant features of the local area, enhance the feature expression ability of the network, and have certain translation invariance and rotation invariance. In the downsampling path of U-Net, each layer usually contains one or more convolution layers to extract features, followed by a maximum pooling layer to reduce the size of the feature map. As the network goes deeper, the size of the feature map gradually decreases, while the depth of the feature map (i.e., the number of channels) gradually increases. Upconvolution (also called transposed convolution or deconvolution) is a special convolution operation that can enlarge the low-resolution feature map to high resolution through the learned convolution kernel. Upconvolution can restore the spatial size of the feature map, so that the network can gradually restore the detailed information of the image in the upsampling path. At the same time, it can also combine deep high-level semantic information with shallow low-level detail information to improve the accuracy of image segmentation. In the upsampling path of U-Net, each layer usually contains one or more up-convolution layers to enlarge the size of the feature map, followed by one or more convolution layers to further extract and refine the features. In addition, U-Net also splices the feature map in the downsampling path with the feature map of the corresponding level in the upsampling path through skip connections to fuse feature information at different levels. U-Net performs downsampling through maximum pooling to extract high-level semantic information, and upsampling through up-convolution to restore image detail information. At the same time, feature information at different levels is fused through skip connections, thereby achieving high-precision image segmentation. This unique structure has enabled U-Net to be widely used and have excellent performance in fields such as medical image segmentation and satellite image segmentation.

[0051] In this embodiment, the EMG signal is collected by a monitor (i.e., an EMG signal collection device) that is continuously attached to the skin surface close to the muscle. More specifically, the EMG signal collection device is attached to the skin surface near the muscle to collect information related to blood circulation, tension and elasticity of the muscle, analyze and judge, set a threshold, and issue an early warning when the blood circulation is reduced and the spasm reaches a certain degree due to strain, prompting us to take appropriate rest, exercise or treatment measures to improve blood circulation and spasm before continuing to work. It is also possible to monitor the effects of our various treatment plans in real time, and judge indicators such as the speed of effectiveness, the degree of improvement of the effect, and the durability of the effect. It is also possible to increase the measurement of blood oxygen saturation, heart rate, respiration, body temperature, blood pressure, and even blood sugar to give relevant health tips. A wireless link is set to the indicator display system, and it can also be uploaded to the cloud and then transmitted to the mobile phone APP (application) to prompt relevant health indicators, and health guidance is given after analysis, and alarms and prompts can be given. It is also possible to evaluate the efficacy and effectiveness of various treatment methods and functional exercise methods, as well as the duration of the efficacy, which can be used as a general evaluation standard. The diameter of the EMG signal collection device is about 2 cm, and it is waterproof. It does not need to be removed for bathing or swimming. It can be removed for cleaning or replacement in about 2 weeks. After cleaning, it can be applied to the skin with a strong patch for continued use, or it can be directly replaced. It can be opened or tightly closed, and a small button battery can be replaced. It contains chips with different functions. It can collect data such as blood circulation, spasm degree, and viscoelasticity of local muscles. If necessary, chips for monitoring blood oxygen saturation, heart rate, respiration, body temperature, blood pressure, and even blood sugar data can be added. These chips can be selected according to the specific circumstances of different users, or an optimized combination can be selected. For example, some people only need equipment with blood pressure indicators, so only equipment with blood pressure monitoring can be selected. It can also be combined. For example, if body temperature and blood sugar indicators are needed, equipment with body temperature and blood sugar indicators can be provided. It also has a wireless transmission chip. It transmits data to a portable device that integrates data storage equipment, data analysis equipment, display equipment, and alarm equipment. The above-mentioned device is composed of several integrated chips, including a data storage device, which will receive and store all collected data and upload it to the data analysis device in real time. In the device, the data will be sorted, summarized, and analyzed to draw preliminary conclusions. The conclusions will be displayed on the display device through the display and transmitted to the alarm device. The alarm device is set according to the preset threshold and different situations. The setting can be adjusted manually under the guidance of professional doctors. At the same time, the device can also upload data to the cloud and then transmit it to designated smart terminal devices such as mobile phones, computers, tablets, sports watches and other devices.A dedicated APP can be used to provide feedback to professional physicians, who can also make clinical decisions in real time based on different situations, and provide feedback to users to guide them to adopt certain treatment plans, such as rest, proper exercise, necessary treatment measures, etc. These suggestions will also be transmitted to the display device in real time for the user to see directly, and can also be fed back to the alarm device to issue appropriate alarms to prompt the user. There is also a real-time and continuous evaluation system for the above functions to evaluate the effects of rest, treatment, exercise methods and plans, and to design software based on specific situations to draw and chart based on real-time data, and make comparisons to reflect continuous real-time effects and establish a new evaluation standard system.

[0052] The present invention can continuously and in real time monitor muscle strain degenerative diseases, and can provide real-time analysis decisions and alarms, so that strain degenerative diseases can be reduced or even eliminated. It allows users to gradually develop work, life and exercise habits that do not cause further strain and degeneration, thereby completely getting rid of similar equipment and the troubles of strain and degeneration. At the same time, it can also provide the same functions of other related health indicators and can optimize the combination. In addition, data from different modalities (EMG and ultrasound) are effectively fused and matched for medical diagnosis, human-computer interface or other application scenarios that require collaborative analysis of multi-source data. The complexity of the architecture shows that it can process high-dimensional, heterogeneous data, and extract useful comprehensive features through multi-step processing, thereby effectively improving the accuracy of monitoring results.

[0053] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0054] What is described above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for monitoring muscle spasm health indicators, characterized in that: The following steps are involved: Collecting EMG signals and classifying and processing them to obtain EMG classification data; Collecting one-dimensional ultrasonic signals and performing data segmentation on them to obtain ultrasonic fine segmentation data; Input EMG classification data and ultrasound fine segmentation data into the fusion model; The alignment and fusion of multimodal features in a unified feature space are achieved through a multimodal feature space alignment encoder; Feature alignment and calibration make the features of different modalities comparable and consistent in the same space for effective fusion and analysis; The calibrated and aligned EMG signal and ultrasound data features are fused through the multimodal fusion module, and displacement mapping is used to rotate, scale and fuse features to capture shared and unique features between different modalities; Perform multiple loop iterations; Output the fusion and matching results of EMG data and ultrasound data for monitoring muscle spasms and collaborative analysis of multi-source data; A mechanism to ensure alignment of data from different modalities in feature space by spatially aligning feature vectors; The EMG data and ultrasound data are processed in two parallel processes, the processing process includes layer normalization, feature mapping and feature integration of the data, and then inputting the data into the attention module for feed-forward processing, combining multiple convolutions, activations, feature mappings, and multiple cycles of iterations, and then integrating the features for the feature alignment and calibration; The classification and processing process of the EMG signal is specifically as follows: Model pre-training, i.e., using data sets from multiple subjects to train the EMG generation model, wherein each subject has a healthy state and multiple abnormal states, each abnormal state includes multiple abnormal symptoms, and each abnormal symptom has a different number of EMG signal collection devices, and the EMG generation model includes self-branching, context branching, and multi-attention module integration to predict the input data and output; EMG signal generation, extracting effective EMG signal fragments from EMG signals of unknown states and symptoms, combining the model pre-training data, and generating synthetic data through the EMG generation model using an autoregressive conditional generation process; EMG classifier training, i.e., using the EMG signals of unknown states and symptoms and the synthetic data to train the classifier to improve its classification accuracy; The EMG signal is classified and processed by the trained classifier to obtain EMG classification data.

2. A method for monitoring muscle spasm health indicators according to claim 1, characterized in that: The specific process of data segmentation of the one-dimensional ultrasonic signal is as follows: Input signal, i.e., input one-dimensional ultrasonic signal; Primary U-Net, that is, passing the signal through a primary U-Net structure; Focus on region sampling, where the output of the primary U-Net is used to generate candidate regions, i.e., sampling regions; Region of interest segmentation, i.e., sampling based on the region of interest is used to crop the original signal; Secondary U-Net, passing the cropped signal through another U-Net structure; Output segmentation, i.e., by performing a rough segmentation based on sampling of the region of interest, the output of the secondary U-Net is used for fine segmentation to predict the peak position of muscle spasm; Classification, that is, the lowest layer of the primary U-Net is flattened and classified through the fully connected layer.

3. A method for monitoring muscle spasm health indicators according to claim 2, characterized in that: The U-Net includes a downsampling path and an upsampling path, and each level uses double 1D convolution.

4. A method for monitoring muscle spasm health indicators according to claim 3, characterized in that: The dual 1D convolution includes convolution, batch normalization and ELU activation.

5. A method for monitoring muscle spasm health indicators according to claim 4, characterized in that: The layers of the U-Net are downsampled by maximum pooling and upsampled by convolution, and the corresponding downsampling and upsampling layers are connected using attention blocks.

6. A method for monitoring muscle spasm health indicators according to claim 5, characterized in that: The EMG signal is collected by a monitor that is continuously attached to the surface of the skin close to the muscle.

Images