A human lower limb chronic pain discrimination method based on surface electromyogram multi-dimensional feature fusion, a storage medium and a device

Abstract

The present application belongs to the technical field of surface electromyography evaluation, and provides a human lower limb chronic pain discrimination method based on surface electromyography signal multi-dimensional feature fusion, a storage medium and a device. The method uses a self-made electromyography collection device to collect surface electromyography (sEMG) signals of selected lower limb muscle groups to form a data training set and a test set. Meanwhile, with the aid of sEMG multi-dimensional features in the time domain and the frequency domain, a supervised learning of the training set data is performed by combining a Gaussian kernel function and a support vector machine. Finally, the type of electromyography data in the test set is discriminated, and the pain symptom recognition accuracy is calculated. Through Gaussian noise introduction, sEMG multi-dimensional feature fusion and support vector machine classification learning, the discrimination accuracy of lower limb chronic pain symptoms can be effectively enhanced. The present application can provide an objective, quantitative and scientific analysis means for the auxiliary diagnosis of lower limb chronic pain symptoms, improve the work efficiency of doctors in diagnosis, reduce unnecessary examinations, and avoid the waste of medical resources.

Description

Technical Field

[0001] This invention belongs to the field of surface electromyography (SEMG) assessment technology, specifically a method, storage medium, and device for identifying chronic pain in the human lower limbs based on the fusion of multi-dimensional features of surface electromyography signals. Background Technology

[0002] Pain is one of the most common clinical symptoms and a complex physiological and psychological activity. The International Association for the Study of Pain defines pain as: an unpleasant sensory and emotional experience accompanied by present or potential tissue damage. Currently, pain has become the fifth vital sign after body temperature, pulse, respiration, and blood pressure, and is receiving increasing attention from the international community and the medical community. In fact, pain is a subjective experience, and traditional medical diagnosis relies heavily on patients' subjective descriptions. Analyzing patients' pain status through objective and scientific techniques is of great significance for standardizing diagnostic procedures and improving the accuracy of pain diagnosis.

[0003] Current methods for subjective pain assessment commonly include numerical assessment and visual analogue (VAS) to describe pain levels. However, due to individual differences in subjective perception, there are always variations in the identification of pain. This is especially true for certain groups, such as young children, the deaf and mute, and the elderly, who face varying degrees of difficulty in accurately describing their pain. In addition to these methods, electroencephalography (EEG) can be used to observe pain-related neural activity in the brain to assess pain intensity. However, this technology is costly, and patients often have their EEG signals collected only in specialized hospitals, significantly limiting its widespread clinical application. Furthermore, pain assessment requires expert knowledge and clinical experience, which can be challenging in regions or areas with limited medical resources, making it difficult to find experienced experts to diagnose pain symptoms.

[0004] Surface electromyography (sEMG) is the bioelectric current generated by the contraction of muscles on the human body's surface. The nervous system controls muscle activity (contraction or relaxation), and different muscle fiber motor units in the skin produce different signals simultaneously. Acquiring the electromyographic signals from the skin surface is called surface electromyography, or sEMG. Surface electromyography signals are one-dimensional action potential sequences, belonging to non-stationary micro-electrical signals, with amplitudes ranging from 0-1.5 mV, useful signal frequencies between 0-500 Hz, and the main energy concentrated between 20-150 Hz. sEMG signals possess many dimensional characteristics, which have important applications in research, clinical practice, and exercise physiology. Furthermore, electromyography devices are relatively inexpensive and have mature applications, offering several advantages in the field of pain management, providing crucial information about pain mechanisms and treatment.

[0005] Pain is one of the most important indicators of clinical symptoms and is crucial for pain treatment. Accurate pain assessment can provide doctors with necessary assistance. To help doctors determine whether a patient has chronic lower limb pain symptoms, this invention collects lower limb electromyography (EMG) signals from the patient, processes and analyzes the action-evoked EMG data, and, through data augmentation and machine learning model training, ultimately provides a conclusion on whether the patient has chronic pain, offering a novel objective evaluation method for chronic pain symptoms. Summary of the Invention

[0006] This invention provides a method, storage medium, and device for identifying chronic pain in the human lower limbs based on the fusion of multi-dimensional features of surface electromyography signals, comprising the following steps:

[0007] S1. Select several subjects, including N healthy volunteers and M patients with chronic lower limb pain. Apply electrode patches to selected areas on the subjects' bodies, connect a power source, and turn on the self-made surface electromyography (sEMG) signal acquisition device. The acquisition module records the subjects' surface electromyography (sEMG) signals at a set frequency. The acquired data is divided into a training set and a test set. The training set data is labeled 0 or 1 for the healthy group and the pain group, respectively.

[0008] S2. Guide the subject to repeat multiple sets of lower limb movements and collect surface electromyography data induced by continuous movements;

[0009] S3. Based on S2, a high-pass filter is used to remove baseline noise and high-frequency interference to obtain a 10Hz-160Hz sEMG filtered signal.

[0010] S4. Based on S3, the sEMG filtered signal is converted to the frequency domain and the energy spectrum is estimated by short-time Fourier transform. The sEMG signal segments below the given energy threshold are eliminated, and the electromyographic potentials induced by lower limb movements are retained.

[0011] S5. Based on S4, Gaussian noise is added to the electromyography data in the training set to artificially synthesize electromyography signals, thereby expanding the training set data.

[0012] S6. Based on S5, extract multi-dimensional features of the training set sEMG from the time domain and time-frequency domain respectively;

[0013] S7. Based on S6, the Gaussian kernel function kernel scale is optimized by combining sEMG multi-dimensional features with Bayesian optimization algorithm for support vector machine classification algorithm.

[0014] S8. Based on S7, the optimal Gaussian kernel function and support vector machine (SVM) are used to perform binary classification of the test set sEMG joint feature data into pain and health, and the accuracy of pain symptom estimation is calculated.

[0015] S9. Randomly allocate the collected electromyographic data to the training set and the test set according to the set ratio, repeat S1 to S8 multiple times (e.g., 1000 times), calculate the average accuracy of pain symptom estimation, and obtain stable evaluation results.

[0016] In S5, the artificially synthesized electromyography (EMG) signal is added to the EMG data of patients with chronic lower limb pain only to achieve the purpose of small sample data augmentation. In S6, the multi-dimensional feature extraction includes root mean square (RMS), mean absolute value (ABSMean), median frequency (MDF), and mean frequency (MNF).

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] 1. This invention is based on a self-developed surface electromyography signal acquisition and storage medium and device. The device is easy to operate, the electrode patch acquisition position is convenient, and the subject's designed action is simple, which helps to promote its application in communities and homes.

[0019] 2. This invention, by introducing Gaussian noise data enhancement and sEMG multi-dimensional feature fusion classification, can effectively improve the accuracy of chronic pain identification and enhance the robustness of identification, providing an effective sEMG quantitative analysis method for chronic pain diagnosis. Attached Figure Description

[0020] Figure 1 This is a structural diagram of the electromyography signal acquisition and analysis device of the present invention;

[0021] Figure 2 This is a flowchart of the electromyography signal acquisition device of the present invention;

[0022] Figure 3 This is a flowchart of the calculation unit of the electromyography signal acquisition device of the present invention;

[0023] Figure 4 This is a schematic diagram of surface electromyography in healthy subjects after pretreatment according to the present invention;

[0024] Figure 5 This is a schematic diagram of surface electromyography of a pain subject after pretreatment according to the present invention;

[0025] Figure 6 A comparison of surface electromyography signals before and after introducing Gaussian noise (SNR=15) for this invention;

[0026] Figure 7 A schematic diagram illustrating the optimization process of introducing a Gaussian kernel function for Gaussian noise (SNR=15) in this invention;

[0027] Figure 8This is a schematic diagram illustrating the pain recognition accuracy on a test set under different Gaussian signal-to-noise ratios according to the present invention.

[0028] Figure 9 This is a schematic diagram of the subject's foot-supported position according to the present invention;

[0029] Figure 10 This is a schematic diagram of the electrode placement in the subject electromyography signal acquisition system of the present invention. Detailed Implementation

[0030] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0031] like Figures 1 to 9 As shown:

[0032] Example: A method, storage medium, and device for identifying chronic pain in the human lower limbs based on the fusion of multi-dimensional features of surface electromyography signals, comprising the following steps:

[0033] S1. Connect the power supply and turn on the acquisition device. Attach electrode patches to the selected area. The electromyography (EMG) signal acquisition module will acquire signals from the selected body part at the set acquisition frequency. The system block diagram is shown below. Figure 1 As shown in the figure. The electromyography (EMG) signal acquisition uses an EMG signal acquisition device and Ag / Agcl surface electrodes. The EMG signal gain is 2000, the input impedance is >100MΩ, the sampling bandwidth is 50-1000Hz, the sensitivity is 1uV, and the signal sampling rate is 1700 times / second. The acquisition process and calculation processing flow are as follows: Figure 2 , 3 As shown.

[0034] S2. Thirty healthy volunteers aged 22-30 and five male patients aged 20-30 with chronic lower limb pain were selected. The subjects were asked to repeatedly perform 34 sets of toe-raising exercises, and their lower limb electromyography (EMG) data were collected. The toe-raising position was as follows: Figure 9 As shown; the electrode pads for the gastrocnemius muscle and hand are placed as follows. Figure 10 As shown.

[0035] S3. Based on S2, a Butterworth high-pass filter is used to remove baseline noise and high-frequency components, extracting the sEMG signal from 10Hz to 160Hz. The filtered EMG signals from healthy and pain subjects are shown below. Figure 4 , 5 As shown;

[0036] S4. Based on S3, the sEMG signal is converted to the frequency domain by short-time Fourier transform and its energy spectrum is estimated. sEMG signal segments below a given energy threshold are eliminated, while action-induced electromyographic potentials are retained.

[0037] S5. Based on S4, new electromyographic signals are artificially synthesized by adding Gaussian noise to the electromyographic data of patients in the training set, thereby expanding the amount of training set data. In order to verify the impact of different signal-to-noise ratio Gaussian noise conditions on the classification accuracy of small sample datasets, this invention generates artificially synthesized electromyographic signals with Gaussian signal-to-noise ratios of SNR = 5, 10, 15, 20, 25, 30, 35, and 40, respectively. Figure 6 A comparison diagram of the potentials of artificially generated electromyographic signals and original electromyographic signals is given when SNR=10;

[0038] S6. Based on S5, multi-dimensional feature extraction is performed on real raw electromyography signals and artificially generated electromyography signals in the time domain and time-frequency domain. Multi-dimensional feature extraction includes root mean square value (RMS), mean absolute value (ABSMean), median frequency (MDF), mean frequency (MNF), etc.

[0039] S7. Building upon S6, the Bayesian optimization algorithm is used to optimize the kernel scale of the Gaussian kernel function, with the number of iterations set to 30. The iteration process is as follows: Figure 7 As shown;

[0040] S8. Based on S7, the Gaussian kernel function with the optimal estimated parameters and the support vector machine (SVM) are used to classify the joint feature data of the electromyographic signals in the test set, and finally output the classification results of pain and health groups. Figure 8 The accuracy of pain state recognition under different Gaussian kernel functions and kernel scales is presented.

[0041] S9. Repeat S1 to S8 multiple times (1000 times), record the accuracy of pain identification each time and calculate the average value as a stable pain symptom assessment result.

[0042] As can be seen from the above, this invention relates to a classification technology for chronic lower limb pain based on the fusion of multi-dimensional features of surface electromyography (EMG) signals, which can perform binary classification (health or pain) on EMG signal data collected from subjects. Firstly, by employing a Gaussian noise data enhancement strategy, it was found that adding Gaussian noise with an SNR of 40 to the EMG signals of the training set patients can effectively improve the classification accuracy of the subjects (see...). Figure 8 Secondly, the kernel scaling parameter (KerS) of the Gaussian kernel function was obtained using Bayesian optimization. The kernel scaling parameter was then assigned integer values ​​from 1 to 5 using an empirical method. Experimental results showed that when KerS was set to 1, the pain discrimination accuracy reached its highest level of 92.77%. Based on these experimental results, it is recommended to superimpose Gaussian noise with an SNR of 40 onto the patient's electromyography data, and set the kernel scaling parameter KerS of the Gaussian kernel function to 1.

[0043] In summary, this method for identifying chronic lower limb pain based on the fusion of multi-dimensional features of surface electromyography signals has the following effects:

[0044] 1. This invention is based on a self-developed surface electromyography signal acquisition and storage medium and device. The device is easy to operate, the electrode patch acquisition position is convenient, and the subject's designed action is simple. This device is conducive to the promotion and application in communities and homes.

[0045] 2. This invention, by introducing Gaussian noise data enhancement and sEMG multi-dimensional feature fusion classification, can effectively improve the accuracy of chronic pain symptom identification and enhance the robustness of the algorithm, providing an effective sEMG quantitative analysis method for chronic pain diagnosis; this invention is conducive to intelligent pain assessment, reduces unnecessary examinations, and avoids the waste of medical resources.

[0046] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A device for identifying chronic pain in the human lower limbs based on the fusion of multi-dimensional features of surface electromyography signals, characterized in that, include: At least one processor; At least one memory for storing computer programs; When the processor executes the computer program stored in the memory, it performs the following steps: S1. Select several subjects, including N healthy volunteers and M patients with chronic lower limb pain. Attach electrode patches to the selected sites on the subjects, connect the power supply and turn on the self-made surface electromyography signal acquisition device. The acquisition module records the surface electromyography signals of the subjects according to the set frequency. The acquired data is divided into training set and test set. The training set data is labeled as 0 or 1 according to the healthy group and the pain group. S2. Guide the subject to repeat multiple sets of lower limb movements and collect surface electromyography data induced by continuous movements; S3. Based on S2, a high-pass filter is used to remove baseline noise and high-frequency interference to obtain a surface electromyography (EMG) filter signal of 10Hz-160Hz. S4. Based on S3, the surface electromyography (EMG) filtering signal is converted to the frequency domain and the energy spectrum is estimated by short-time Fourier transform. The surface EMG signal segments below the given energy threshold are eliminated, and the EMG potentials induced by lower limb movements are retained. S5. Based on S4, Gaussian noise is added to the electromyography data in the training set to artificially synthesize electromyography signals, thereby expanding the training set data. S6. Based on S5, extract multi-dimensional features of surface electromyography from the training set in the time domain and time-frequency domain respectively. S7. Based on S6, the multidimensional features of surface electromyography are combined with the Bayesian optimization algorithm to optimize the kernel scale of the Gaussian kernel function for use in the support vector machine classification algorithm. S8. Based on S7, the optimal Gaussian kernel function and support vector machine are used to classify the surface electromyography joint feature data of the test set into pain group and healthy group, and the accuracy of pain symptom estimation is calculated. S9. Randomly allocate the collected electromyography data to the training set and the test set according to the set ratio. Repeat S1 to S8 multiple times to calculate the average accuracy of pain symptom estimation and obtain stable evaluation results.

2. The apparatus as described in claim 1, characterized in that: In S5, the artificially synthesized electromyography (EMG) signal is only added to the EMG data of patients with chronic lower limb pain to achieve the purpose of enhancing small sample data.

3. The apparatus as described in claim 1, characterized in that: In S6, the multi-dimensional feature extraction includes root mean square value, average absolute value, median frequency, and average frequency.

4. The apparatus as claimed in claim 1, characterized in that: In S7, the Gaussian kernel function kernel scale is selected iteratively using a Bayesian optimization algorithm, which includes the following steps:

1. Define the hyperparameters, including the search range for kernel scale and bounding box constraint level, both of which have a search range of 0.001-1000; 2. Select the model classification error as the objective function and minimize the error. The input is the hyperparameter value, and the output is the classification error.

3. Randomly initialize hyperparameters and use a Gaussian process to establish a prior probability distribution, which is used to predict the objective function value; Fourth, select the Expected Improvement acquisition function, assign a score to each possible hyperparameter combination, and then select the hyperparameter combination with the highest score as the next hyperparameter evaluation value. At the same time, pass the evaluation value to the objective function for error calculation, and continuously iterate the above steps until the maximum number of iterations is reached or complete convergence is achieved.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the apparatus as described in any one of claims 1 to 4.

Images