Peripheral nerve injury electromyographic signal analysis and evaluation method

Abstract

This invention relates to the field of electromyography (EMG) signal technology, specifically to a method for analyzing and evaluating EMG signals in peripheral nerve injury. The method includes the following steps: acquiring raw EMG signals and environmental noise reference signals. This invention discretizes the intrinsic mode function (IMF) components and the environmental noise reference signal, calculates the length of the discrete signal; calculates the cross-correlation coefficient based on the length, sorts the components to obtain the correlation coefficient of the largest component and determines a threshold, filters effective IMF components according to the threshold, and then integrates them into a set of effective components; processes the IMF components within the set to obtain denoised components, and superimposes the denoised components to obtain a pure EMG signal, effectively removing power frequency interference and patient jitter artifacts, avoiding the accidental deletion of weak EMG signals from early, minor nerve injuries by traditional fixed thresholds, improving the purity of EMG signals with low signal-to-noise ratios, and preserving effective component characteristics even at low signal-to-noise ratios, thus improving the integrity of preserved effective features.

Description

Technical Field

[0001] This invention relates to the field of electromyography (EMG) signal technology, and more specifically, to a method for analyzing and evaluating EMG signals in peripheral nerve injury. Background Technology

[0002] A method for analyzing and evaluating electromyographic signals of peripheral nerve injury is proposed. Its core purpose is to quantitatively assess and grade the state of peripheral nerve injury. By systematically processing the electromyographic signals of muscles innervated by the ulnar nerve (such as the abductor digiti minimi), it ultimately outputs three core indicators: nerve conduction integrity (NCI), muscle contraction function (MCI), and damage severity comprehensive (DSI) and their corresponding grades (no obvious damage / mild / moderate / severe). This provides objective and accurate injury status data support for medical and nursing management. Its core value lies in solving the problems of difficulty in identifying mild injuries in the early stage and the single assessment dimension, filling the gap in sensitivity and comprehensiveness of traditional electromyographic signal analysis methods. Traditional electromyography (EMG) signal analysis and assessment methods typically involve four main stages: signal acquisition, denoising, feature extraction, and evaluation. These methods primarily acquire muscle EMG signals from patients using surface EMG sensors. In the denoising stage, methods such as wavelet denoising based on fixed thresholds or empirical mode decomposition (EMD) are mainly employed. These methods filter out noise by setting a fixed threshold or directly remove high-frequency components after mode decomposition. However, these methods lack the ability to adaptively identify and separate noise components, making it difficult to effectively handle situations with weak signals and low signal-to-noise ratios in early, minor injuries, and easily leading to the accidental deletion of useful information.

[0003] In the feature extraction stage, traditional electromyography (EMG) signal analysis and evaluation methods mainly rely on single time-domain or frequency-domain features such as root mean square (RMS), median frequency (MF), and average power frequency (MPF), without making full use of nonlinear features and the correlation information between multiple features. The evaluation dimensions are relatively isolated, making the EMG signal distortion caused by early nerve injury very weak. The expressive power of a single feature is limited. At the same time, it is easily affected by power frequency interference and patient motion artifacts in the acquisition environment, resulting in insufficient differentiation between normal signals and slightly damaged signals, and reduced overall evaluation sensitivity. In summary, traditional electromyography (EMG) signal analysis and assessment methods still face two major technical bottlenecks in assessing EMG signals in patients with early-stage, mild nerve damage, resulting in insufficient assessment efficacy: On the one hand, in the early stages of nerve injury, the electromyographic (EMG) signals of the neuromuscular system are weak and the effective EMG signal-to-noise ratio is extremely low. They mainly rely on single-dimensional time-domain features (such as root mean square value, integral EMG value) or frequency-domain features (such as median frequency, average power frequency). However, the power frequency (50Hz / 60Hz) interference commonly present in the acquisition environment and the motion artifacts that cannot be completely avoided by patients during the examination (such as artifacts caused by muscle tremors or electrode displacement) will highly overlap with the target weak EMG signal, forming complex background noise. Therefore, traditional EMG signal analysis and evaluation methods using traditional fixed threshold denoising methods are too rigid and easily over-remove the weak, low-frequency, or low-amplitude EMG signal components that remain in the early stages of nerve injury, resulting in the loss of key electrophysiological information. Actual studies have shown that in early mild nerve injury scenarios, the recognition sensitivity of these traditional feature extraction methods is generally reduced, and the purity and feature integrity of the EMG signal cannot meet the requirements for early screening. Therefore, we provide a method for analyzing and evaluating EMG signals in peripheral nerve injury. Summary of the Invention

[0004] The purpose of this invention is to provide a method for analyzing and evaluating electromyographic signals of peripheral nerve injury, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for analyzing and evaluating electromyographic signals of peripheral nerve injury, comprising the following steps: 1. A method for analyzing and evaluating electromyographic signals of peripheral nerve injury, characterized by comprising the following steps: S1. Acquire the patient's raw electromyographic signals using an electromyography sensor, and simultaneously acquire environmental noise reference signals using a noise sensor; S2. The original electromyographic signal is decomposed using a variational mode decomposition algorithm to obtain the set of intrinsic mode function components and the set of center frequencies; S3. Calculate the cross-correlation coefficient between each intrinsic mode function component and the environmental noise reference signal, and filter the effective intrinsic mode function components based on the cross-correlation coefficient; S4. Perform wavelet denoising on the selected effective intrinsic mode function components, obtain the denoised components, and superimpose them to obtain a pure electromyographic signal. S5. Extract time-domain features, frequency-domain features, and nonlinear feature vectors from the pure electromyography signal to obtain the extracted feature vectors; S6. Calculate the Pearson correlation coefficient and the comprehensive nerve injury index based on the extracted feature vectors, and assess the patient's nerve injury status based on the comprehensive nerve injury index.

[0006] As a further improvement to this technical solution, the specific steps of the variational mode decomposition algorithm in S2 include: Initialize the known set of intrinsic mode function components, the set of center frequencies, and the Lagrange multipliers; The eigenmode function components and center frequency are updated iteratively using the alternating direction multiplier method; The iteration termination condition is determined based on the dynamic convergence threshold. The iteration is terminated when the relative error of the maximum intrinsic mode function component and the absolute error of the maximum center frequency are both less than the dynamic convergence threshold in two consecutive iterations. The set of intrinsic mode function components and the set of center frequencies of the decomposed original electromyographic signal are then output.

[0007] As a further improvement to this technical solution, the specific steps for setting the dynamic convergence threshold include: The initial maximum threshold is recorded by initializing the convergence threshold. Based on the initial maximum threshold and the current iteration number Achieving a dynamic convergence threshold using the maximum number of iterations .

[0008] As a further improvement to this technical solution, the specific steps for effective component screening in S3 include: Discretize the intrinsic mode function components and the environmental noise reference signal into discrete signals, and calculate the length of the discrete signals; The cross-correlation function value is calculated based on the length of the discrete signal, the cross-correlation coefficient is calculated based on the cross-correlation function value, and the correlation coefficient of the largest component is obtained by sorting the cross-correlation coefficients. Calculate the adaptive cross-correlation threshold based on the correlation coefficient of the maximum component; Eigenmode function components with cross-correlation coefficients less than the adaptive cross-correlation threshold are identified as valid eigenmode function components.

[0009] As a further improvement to this technical solution, the specific steps of the adaptive cross-correlation threshold include: The cross-correlation coefficients are sorted from highest to lowest to obtain the correlation coefficient of the largest component. The adaptive cross-correlation threshold is calculated using the correlation coefficient of the maximum component. .

[0010] As a further improvement to this technical solution, the wavelet denoising process in S4 uses db4 wavelet for 3-level decomposition, and the specific steps include: The processed wavelet coefficients are obtained by performing wavelet decomposition on the effective intrinsic mode function components. The iteration step size is estimated based on the highest level detail coefficients of the wavelet, and then the processed wavelet coefficients are inversely transformed to obtain the denoised components.

[0011] As a further improvement to this technical solution, the time-domain feature vector in S5 includes the root mean square value, the integral electromyographic value, and the peak factor; Frequency domain feature vectors include average power frequency and median frequency; nonlinear features include approximate entropy and sample entropy. It also includes calculating the muscle synergy coefficient as a coefficient feature, which is calculated based on the mean, standard deviation, and covariance of the pure electromyographic signal.

[0012] As a further improvement to this technical solution, the Pearson correlation coefficient is calculated from the features extracted in S6, and the specific steps include the following: The time domain, frequency domain, nonlinear eigenvectors, and coefficient features are combined in pairs to calculate the Pearson correlation coefficient between the two features.

[0013] As a further improvement to this technical solution, the specific steps of the comprehensive neurological injury index in S6 include: By standardizing the extracted features using Z-score, a standardized feature vector is obtained. The sensitivity weight for nerve injury was calculated based on the analysis of variance values; the association weight was calculated based on the Pearson correlation coefficient. The final weight is obtained by combining the nerve injury sensitivity weight and the correlation weight; A fusion feature value is generated by fusing standardized feature vectors with final weights, and a comprehensive index of nerve damage is calculated based on the fusion feature value.

[0014] As a further improvement to this technical solution, the specific steps of S6 in assessing the patient's nerve damage are as follows: Obtain known normal comprehensive index values ​​and the range of comprehensive index values ​​for minor nerve damage from the medical database; The presence of nerve damage in a patient is determined by comparing the comprehensive index value of nerve injury with the ranges of normal and mild nerve injury comprehensive index values. When the comprehensive index value of nerve injury is greater than or equal to the normal comprehensive index value, the patient is determined to have no nerve injury. When the comprehensive index value of nerve injury is within the range of the comprehensive index value of mild nerve injury, the patient is determined to have mild nerve injury.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This method for analyzing and evaluating electromyographic (EMG) signals of peripheral nerve injury discretizes the intrinsic mode function (IMF) components and the environmental noise reference signal, and calculates the length of the discrete signal. Based on the length of the discrete signal, the cross-correlation coefficient is calculated, and the maximum component correlation coefficient is obtained by sorting. An adaptive cross-correlation threshold is then determined based on this. Effective IMF components are selected according to the adaptive cross-correlation threshold and then integrated into a set of effective components. The IMF components within the set are processed by processing wavelet coefficients and then undergoing inverse transform to obtain denoised components. The denoised components are then superimposed to obtain a pure EMG signal. This method effectively removes power frequency interference from the screening room and artifacts caused by patient jitter during examination, avoiding the accidental deletion of weak EMG signals from early, minor nerve injuries by traditional fixed thresholds. It improves the purity of EMG signals with low signal-to-noise ratios and retains the characteristics of effective components even at low signal-to-noise ratios, thus improving the integrity of the retained effective features.

[0016] 2. This method for analyzing and assessing electromyographic signals of peripheral nerve injury involves obtaining ANOVA values ​​from labeled peripheral nerve injury samples and calculating nerve injury sensitivity weights based on these ANOVA values. Correlation weights are calculated based on Pearson correlation coefficients, and the final weights are obtained by linking these nerve injury sensitivity weights. A fusion feature value is generated by fusing standardized feature vectors with the final weights. Using normal human indicators from a medical database as a benchmark, the method assesses nerve conduction integrity and muscle contraction function indicators. A comprehensive nerve injury index is calculated based on these two indexes, and the patient's nerve injury status is determined according to this comprehensive index. The method comprehensively considers the sensitivity and independence of features through a fusion strategy, suppressing redundant feature interference and strengthening core features that are indicative of nerve injury. It also establishes a correlation assessment dimension between nerve conduction function and muscle contraction function, breaking the limitations of traditional isolated indicators. This allows for a comprehensive and accurate reflection of the patient's nerve injury index values, improving the overall accuracy of the assessment. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall steps of the present invention. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides a method for analyzing and evaluating electromyographic signals in peripheral nerve injury. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps: S1 includes the following method steps: S1.1 First, guide the patient into the waiting room. Collect the patient's heart rate using a heart rate monitor worn on the patient's wrist. Obtain the collected heart rate and then determine the standard heart rate range. Use the collected heart rate and the standard heart rate to determine if the patient needs to calm down. If the collected heart rate exceeds the standard heart rate range, have the patient sit quietly for 5 minutes to calm down. At the same time, use a metal detector installed in the waiting room to detect metal jewelry on the patient's upper limbs. Remove the metal jewelry to avoid electromagnetic interference. When the collected heart rate is within the standard heart rate range and the metal jewelry on the upper limbs has been removed, lead the patient into the screening room. The patient should be seated (meaning that the patient should be seated for subsequent related operations or in a specific state), with the upper limbs naturally placed on the operating table, palms facing up, and the little finger and thumb kept relaxed to prevent signal distortion caused by active muscle contraction. Secondly, clean the skin of the abductor digitorum brevis muscle (innervated by the ulnar nerve) on the patient's right hand, then apply conductive ointment, and attach the electromyography (EMG) sensor to the skin surface of the abductor digitorum brevis muscle (innervated by the ulnar nerve) on the patient's right hand, ensuring good contact between the EMG sensor and the patient's skin. At the same time, attach the noise sensor to the skin area of ​​the patient's right forearm where there is no muscle attachment, and keep it away from the EMG sensor to avoid EMG signal interference. By properly deploying the sensors, EMG signals and environmental noise reference signals can be accurately collected. Simultaneously activate the electromyography (EMG) sensor and noise sensor to acquire signals, and use the EMG sensor and noise sensor respectively at acquisition frequencies. and collection duration Collect primitive electromyographic signals of the abductor digiti minimi muscle from the patient. and ambient noise reference signal Randomly acquire raw electromyographic signals and the number of signal points corresponding to the original electromyographic signal Calculate the amplitude of the original electromyography (EMG) signal based on the original EMG signal. The amplitude of the original electromyographic signal can be obtained by taking its absolute value (or its magnitude if it is a complex signal). The mean amplitude can be calculated based on the original electromyographic signal amplitude and the number of signal points. ,in, This refers to the label, indicating the first... The original electromyography (EMG) signal amplitudes are calculated, and then the amplitude standard deviation is calculated based on the original EMG signal amplitudes, the number of signal points, and the amplitude mean. The amplitude range is determined based on the mean amplitude and standard deviation of the amplitude. The amplitude range is typically set as follows: ,in It is an empirical coefficient, typically set to a value of [value missing]. or The value is or The principle: In statistics, if signal amplitude data approximately follows a normal distribution (Gaussian distribution), then according to the properties of the normal distribution, about 95.45% of the data will fall within the mean amplitude. about Double standard deviation Within this range, approximately 99.73% of the data will fall within the mean amplitude. about Double standard deviation Within the range; The amplitude and range of the original electromyographic signal are used to determine whether there is significant shaking in the patient's limb. When the amplitude of the original electromyographic signal is within the range, it is determined that there is no significant shaking in the patient's limb. When the amplitude of the original electromyographic signal exceeds the range, it is determined that there is significant shaking in the patient's limb. S1.2. The primitive electromyographic signal of the abductor digiti minimi muscle is decomposed using the Variational Mode Decomposition (VMD) algorithm, and then the objective function of the Variational Mode Decomposition is solved using the Alternating Direction Multiplier Method (ADMM). The specific algorithm formula is as follows: ; Retrieve the set of known intrinsic mode functions from the database. Known center frequency set and Lagrange multipliers Then, for the known set of intrinsic mode functions Known center frequency set and known Lagrange multipliers Initialization is performed by assigning random initial values ​​(or zero values) to the known set of intrinsic mode functions, denoted as the initial set of intrinsic mode functions. ,in, It refers to the first in the set Given a set of known center frequencies, assign initial frequency values ​​to the set of known center frequencies (based on a reasonable allocation of frequencies acquired from electromyography signal acquisition), denoted as the initial set of known center frequencies. Then, assign initial values ​​(usually zero vectors) to the known Lagrange multipliers, denoted as the initial Lagrange multipliers. Initialize the iteration count timer to 0 and set the maximum number of iterations. And the convergence threshold, and obtain the preset number of decomposition layers from the database. Based on the known set of intrinsic mode functions and the preset number of decomposition layers, the constraints are formulated as follows: ,in, This represents a constraint condition that requires the decomposed product to meet certain conditions. The sum of the individual IMF components equals the patient's original electromyographic signal of the abductor digiti minimi muscle, ensuring the accuracy and completeness of the decomposition. To address the optimization challenges posed by constraints, a known set of intrinsic mode functions is used. Center frequency set and Lagrange multipliers By introducing the augmented Lagrangian function, the constrained problem is transformed into an unconstrained problem, thus deriving the augmented Lagrangian function. The final output is obtained by augmenting the Lagrange function. Initial iteration eigenmode function set Center frequency set Lagrange multipliers Specific algorithm formula: ;in, The penalty factor is used to enforce the satisfaction of constraints (combining the characteristics of the original electromyographic signal of the abductor digiti minimi muscle (1000Hz sampling, 20-500Hz bandwidth), with the core being to satisfy the constraints without distorting the electromyographic signal characteristics of weak injuries; the larger the penalty factor, the higher the constraint satisfaction, but too large a factor can easily lead to distortion of the original electromyographic signal of the abductor digiti minimi muscle; if too small a factor, the constraints will fail. Referring to the VMD decomposition experience of similar original electromyographic signals of the abductor digiti minimi muscle, the penalty factor is set to 2000, which ensures that the sum of the intrinsic mode function components after decomposition is equal to the original signal, and avoids excessive suppression of the weak electromyographic signal characteristics of early minor injuries, thus adapting to the needs of low signal-to-noise ratio scenarios). These are Lagrange multipliers used to balance the deviation between the objective function optimization and the constraints. Indicates time Find the partial derivative. This refers to the Dirac function, a generalized function that is often used in signal processing to represent a unit impulse signal. It is the imaginary unit ( This term, along with the Dirac function, participates in the convolution operation and is used to construct the mathematical model of the VMD algorithm. This represents the convolution operation. This refers to the complex exponential function, used for frequency domain transformation of electromyographic signals. The square of the 2-norm of a vector or function is used to measure the energy of an electromyographic signal. Transforming a time-domain problem into a frequency-domain problem (Fourier transform denoted as...) The inverse transform is denoted as In the frequency domain correspond , correspond , correspond ,in, In the frequency domain, angular frequency is represented. The amplitude / phase information at that point is used to derive the following: Frequency domain update formula: The updated frequency domain results are obtained, and then an inverse Fourier transform is performed on the updated frequency domain results to obtain the updated values ​​of the eigenmode function components in the time domain. The specific algorithm formula is as follows: ; According to the center frequency Describe IMF components The main oscillation frequency is fixed and updated. Wheel intrinsic mode function set , minimize The update formula (also based on frequency domain solution) is obtained, thus yielding the updated value of the main oscillation center angular frequency. The specific algorithm formula is as follows: ; Due to the The initial iteration of the Lagrange multipliers is used to correct the deviation of the constraints, and the update formula is as follows (direct update in the time domain): ;in, The iteration step size (usually taken as...) ); The initial maximum threshold is recorded by initializing the convergence threshold. The initial maximum threshold was determined by combining the characteristics of the abductor digiti minimi electromyography (EMG) signal (1000Hz sampling, 20-500Hz bandwidth) and the VMD decomposition constraints. The core objective was to achieve rapid convergence without losing the EMG signal characteristics of weak lesions. An excessively large initial maximum threshold would lead to insufficient convergence, while an excessively small threshold would increase invalid iterations. Therefore, referencing similar experiences with low signal-to-noise ratio EMG signal decomposition, a record of the initial maximum threshold was established. This approach adapts to the weak distortion characteristics of electromyographic signals from early, minor injuries, ensuring both efficient termination of the iteration and meeting the constraint that the sum of the intrinsic mode function components after decomposition equals the original signal, thus guaranteeing decomposition accuracy. It is based on the initial maximum threshold and the current iteration number. and maximum number of iterations Achieve dynamic convergence threshold ; Due to the convergence characteristics of the VMD frequency domain quadratic optimization (positively correlated with the frequency domain concentration of the original electromyographic signal of the abductor digiti minimi muscle), the spectral concentration of the original electromyographic signal of the abductor digiti minimi muscle (1000Hz sampling, 20-500Hz bandwidth) is anchored, and the effective frequency domain range of the signal is narrow and the spectrum is concentrated. Therefore, the quadratic optimization of the VMD frequency domain subproblem is achieved in the first half of the iteration ( This can achieve over 80% convergence, matching the rapid early convergence characteristic of exponential decay; Where, the denominator is taken Used to adapt to the actual convergence trend of the primitive electromyographic signal of the abductor digiti minimi muscle, which is fast at the beginning and slow at the end - the first half of the process ( Iterations need to quickly approximate the optimal solution, using the denominator. Allow the dynamic convergence threshold to decay rapidly to reduce invalid iterations; the second half ( The attenuation is slowed down to match the slow convergence characteristics of the fine adjustment in the later stage of VMD, so as to avoid excessive attenuation and loss of the electromyographic signal characteristics of weak damage. Update values ​​based on eigenmode function components in the time domain Updated value of the center angular frequency of the main oscillation With the Initial iteration eigenmode function set Center frequency set The dynamic convergence threshold is used to determine whether to terminate the iteration. The formula for this determination is as follows: ,and ; Determine to terminate the iteration; where, and This can be understood as two consecutive iterations or two consecutive rounds of iteration. This refers to the relative error of the maximum eigenmode function components; This refers to the absolute error of the maximum center frequency; At the same time, if the number of iterations Regardless of whether the above judgment conditions are met, the iteration is terminated (to prevent infinite iteration). When the iteration terminates, the set of eigenmode function components of the final decomposed primitive electromyographic signal of the abductor digiti minimi muscle is output. and the set of center frequencies of the primitive electromyographic signals of the abductor digiti minimi muscle. , That is ,in, , That is The corresponding center frequency of the original electromyographic signal of the abductor digiti minimi muscle is denoted as . The intrinsic mode function components of the original electromyography (EMG) signal of the abductor digiti minimi muscle are ultimately decomposed into high-frequency intrinsic mode function components corresponding to power frequency interference and patient jitter artifacts, as well as low-frequency intrinsic mode function components corresponding to the effective EMG signal. The original EMG signal is decomposed into high-frequency noise components (power frequency interference, jitter artifacts) and low-frequency effective intrinsic mode function components. The decomposition accuracy is ensured by using a dynamic convergence threshold, while preserving the low-frequency features corresponding to weak injuries. This avoids the loss of low-frequency features caused by fixed threshold decomposition, and can accurately separate noise and effective EMG signals, improving the preservation rate of weak features. S1.3 Discretize the intrinsic mode function components of the original electromyographic signal of the abductor digiti minimi muscle and the environmental noise reference signal into discrete signals of the abductor digiti minimi muscle. Discrete signals with environmental noise ,in, This refers to the time index of a discrete signal, which is used to calculate the length of the discrete signal based on the acquisition frequency and acquisition duration. Then, based on the discrete signal of the abductor digiti minimi muscle, the discrete signal of the environmental noise, and the length of the discrete signal, the cross-correlation function value between the abductor digiti minimi muscle and the environmental noise is calculated. The specific algorithm formula is as follows: ,in, This refers to delay, with a value range of [value missing]. In actual calculations, for simplification, only the following calculations need to be performed: The cross-correlation value at time t, because it reflects the correlation between two signals in the case of no delay; Meanwhile, the cross-correlation function value between the abductor digiti minimi muscle and environmental noise. That is The discrete signal of the abductor digiti minimi muscle is calculated based on the discrete signal of the abductor digiti minimi muscle. Autocorrelation function value at The specific algorithm formula is as follows: Then, based on the discrete environmental noise signal, calculate the discrete environmental noise signal in... Autocorrelation function value at The specific algorithm formula is as follows: Combining the cross-correlation function between the abductor digiti minimi muscle and environmental noise in The value at which the cross-correlation coefficient of the components is calculated. The cross-correlation coefficients of the components are sorted from highest to lowest to obtain the correlation coefficient of the component with the largest cross-correlation coefficient. The adaptive cross-correlation threshold is calculated using the correlation coefficient of the maximum component. Among them, take and The basis for this approach is the noise characteristics of the abductor digiti minimi muscle electromyography (EMG) signal. The statistical value of the cross-correlation coefficient between the effective EMG signal and power frequency interference and jitter artifacts is typically below 0.3. Therefore, 0.3 is used as the basic threshold to initially filter out high-noise components. Combined with actual samples of effective EMG signals (sampled at 1000Hz), the correlation coefficient of the largest component is mostly within... The coefficient of 0.05 allows the adaptive cross-correlation threshold to dynamically adapt to the noise intensity, avoiding the rigidity of a fixed threshold. The adaptive cross-correlation threshold setting retains weak damage feature components with low cross-correlation coefficients with noise, and can adjust the adaptive cross-correlation threshold according to the noise intensity. When the noise is strong, the adaptive cross-correlation threshold is moderately increased to avoid erroneously deleting effective intrinsic mode function components; when the noise is weak, the adaptive cross-correlation threshold is decreased to enhance denoising. Compared with a fixed threshold, the effective intrinsic mode function retention rate is improved, which is suitable for the processing needs of electromyographic signals of early mild damage. The validity of the intrinsic mode function (IMF) components of the primitive electromyographic (EMG) signal of the abductor digiti minimi muscle is determined by using the cross-correlation coefficients of the components and an adaptive cross-correlation threshold. If the cross-correlation coefficient is greater than or equal to the adaptive cross-correlation threshold, the IMF components of the primitive EMG signal of the abductor digiti minimi muscle are deemed invalid and removed. If the cross-correlation coefficient is less than the adaptive cross-correlation threshold, the IMF components of the primitive EMG signal of the abductor digiti minimi muscle are deemed valid and retained. The valid IMF components of the primitive EMG signal of the abductor digiti minimi muscle are then integrated into a set of abductor digiti minimi components. By filtering out effective intrinsic mode function components that are less affected by noise through the cross-correlation coefficient with the noise reference signal, the problem of erroneous deletion of effective signals by traditional fixed threshold is avoided. The processed wavelet coefficients were obtained by performing a db4 wavelet 3-level decomposition on each component in the abductor digiti minimi muscle component set. ;in, These are wavelet coefficients. The iteration step size is estimated using the highest level detail coefficients of the wavelet. The processed wavelet coefficients are then subjected to inverse transform using the iteration step size to obtain the denoised abductor digiti minimi muscle component. Finally, the denoised abductor digiti minimi muscle components are superimposed to obtain the pure electromyographic signal of the abductor digiti minimi muscle. The effective intrinsic mode function components after screening are further denoised to obtain a pure electromyographic signal, thus solving the problem of electromyographic signal distortion caused by power frequency interference and patient tremor artifacts in the examination scenario. S2 includes the following method steps: S2.1 Calculate the root mean square value of the pure electromyographic signal of the abductor digiti minimi muscle based on the pure electromyographic signal and discrete signal length using the root mean square formula. The root mean square value of the pure electromyographic signal of the abductor digiti minimi muscle reflects the contractile strength of the nerve-innervated muscle. The integral electromyography (EMG) value of the pure EMG signal of the abductor digiti minimi muscle is calculated using the integral EMG formula based on the pure EMG signal and the length of the discrete signal of the abductor digiti minimi muscle. Then, the pure electromyographic signals of the little finger abductor muscle are sorted from high to low to obtain the pure electromyographic signal of the maximum little finger abductor muscle. It can reflect the muscle's contraction endurance. The peak factor of the pure electromyographic signal of the abductor digiti minimi is calculated based on the root mean square value of the pure electromyographic signal of the abductor digiti minimi and the pure electromyographic signal of the abductor digiti minimi. It can reflect the instantaneous fluctuations of muscle contraction, and combine the root mean square value, integral electromyography value and peak factor of the pure electromyography signal of the abductor digiti minimi muscle into a time-domain feature vector of the pure electromyography signal of the abductor digiti minimi muscle. Fast Fourier Transform was performed on the pure electromyographic signal of the abductor digiti minimi muscle to obtain the frequency domain signal of the abductor digiti minimi muscle. ,in, It refers to the sampling frequency mentioned in S1.1. The length of the discrete signal is used to calculate the power spectral density of the pure electromyographic signal of the abductor digiti minimi muscle based on the frequency domain signal of the abductor digiti minimi muscle and the length of the discrete signal. The average power frequency of the pure electromyographic signal of the abductor digiti minimi muscle was calculated using the average power frequency formula based on the acquisition frequency and the power spectral density of the pure electromyographic signal of the abductor digiti minimi muscle. Then, the median frequency of the pure electromyographic signal of the abductor digiti minimi muscle was calculated using the power spectral density of the abductor digiti minimi muscle. Among them, 20 is limited. Up to 500 This frequency range is primarily based on physiological considerations: In electromyography (EMG) signal analysis, signals at different frequency ranges contain different physiological information. EMG signals generated by muscle activity have a specific frequency distribution, and most physiologically significant signals are usually concentrated within a specific frequency range. Up to 500 This range likely encompasses the main frequency components related to muscle contraction and nerve conduction, filtering out frequencies below 20. Potential low-frequency noise interference and frequencies above 500 High-frequency noise interference can be eliminated to more accurately assess the electromyographic signal status of the patient's muscles. The average power frequency and median frequency of the abductor digiti minimi muscle are integrated into the frequency domain feature vector of the pure electromyographic signal of the abductor digiti minimi muscle. S2.2 Calculate the similarity tolerance of the pure electromyographic signal of the abductor digiti minimi muscle based on the pure electromyographic signal of the abductor digiti minimi muscle. ,in, This refers to the standard deviation. The approximate entropy of the pure electromyographic signal of the abductor digiti minimi muscle is calculated using the approximate entropy formula based on the similarity tolerance of the pure electromyographic signal of the abductor digiti minimi muscle. ,in, This refers to embedding dimensions of At that time, the proportion of pattern pairs that satisfy the similarity tolerance of the pure electromyographic signal of the abductor digiti minimi muscle is determined, and then the sample entropy of the pure electromyographic signal of the abductor digiti minimi muscle is calculated according to the similarity tolerance using the sample entropy formula. and, among which, and It is a key statistic in the process of calculating sample entropy, which integrates the approximate entropy of the pure electromyographic signal of the abductor digiti minimi muscle and the sample entropy into a nonlinear feature vector of the pure electromyographic signal of the abductor digiti minimi muscle. The mean of the pure electromyographic (EMG) signal of the abductor digiti minimi muscle was calculated using the pure EMG signal and the length of the discrete signal. Then, the standard deviation and covariance of the pure EMG signal were calculated using the pure EMG signal, the length of the discrete signal, and the mean of the pure EMG signal. The muscle coordination coefficient of the pure EMG signal was calculated using the standard deviation and covariance of the pure EMG signal. This muscle coordination coefficient was used as the coefficient feature of the pure EMG signal. The four basic features of the pure EMG signal—time-domain feature vector, frequency-domain feature vector, nonlinear feature vector, and coefficient feature—were arbitrarily paired, and the Pearson correlation coefficient between any two features was calculated. A 4×4 feature correlation matrix was constructed based on the Pearson correlation coefficient. The time-domain feature vector (reflecting the amplitude change of the pure electromyographic signal of the abductor digiti minimi) and the frequency-domain feature vector (reflecting the spectral distribution) are adapted to the characteristics of the pure electromyographic signal of the abductor digiti minimi with a sampling frequency of 1000Hz and a bandwidth of 20-500Hz. The nonlinear feature matches the complex nature of neuro-electromagnetic signals. The coefficient feature (muscle synergy coefficient of the pure electromyographic signal of the abductor digiti minimi) corresponds to the physiological mechanism of neuro-muscle synergy. The four types of features completely characterize the nerve injury state from different dimensions, breaking through the limitation that a single feature cannot fully capture the injury information, avoiding the distortion of weight allocation due to feature redundancy, and solving the problem that the electromyographic signal of early minor injury is easily masked. By calculating the Pearson correlation coefficient to construct a 4×4 feature correlation matrix, redundant features are accurately identified, providing a basis for subsequent weight allocation. Ultimately, the specificity and sensitivity of nerve injury assessment are improved, ensuring that no abnormal changes in nerve conduction and muscle contraction function are missed. S3 includes the following steps: S3.1. The four basic features of the pure electromyographic signal of the abductor digiti minimi muscle are normalized by Z-score to obtain the normalized feature vector of the pure electromyographic signal of the abductor digiti minimi muscle. This eliminates the dimensional differences between different features, avoiding the impact of different dimensions on subsequent weight calculations and feature fusion; Obtain labeled normal samples, early-stage peripheral nerve injury samples, intermediate-stage peripheral nerve injury samples, and severe peripheral nerve injury samples, along with the corresponding observations for each sample. Then, divide the samples into... Each group is divided into groups, and the sample size and total sample size are recorded for each group, based on the corresponding observations, the number of groups, the sample size in each group, and the total sample size. Calculate the population mean and group mean separately. Then, based on the population mean, group mean, number of groups, and sample size, calculate the sum of squared deviations between groups. Within-group sum of squares At the same time, the degrees of freedom between sample groups are calculated using the number of groups. and within-sample group degrees of freedom Then, the mean square between sample groups is calculated using the sum of squares of deviations between sample groups, the sum of squares of deviations within sample groups, and the degrees of freedom between and within sample groups, respectively. and within-sample group mean square Then, the variance analysis values ​​are calculated based on the mean squares between and within groups of the sample. Analysis of variance can test whether there are significant differences in the means of a feature between different groups. The larger the value of the analysis of variance, the more significant the difference of the feature between different groups, and the better it can distinguish different nerve damage around the patient's muscles. The nerve injury sensitivity weights were calculated using analysis of variance values ​​and standardized eigenvectors of pure electromyography signals. The nerve injury sensitivity weight reflects the sensitivity of the standardized feature vector of each pure electromyographic signal to early nerve injury in the patient. Calculate the average correlation coefficient of each feature with other features based on the Pearson correlation coefficient between the two features. The average correlation coefficient reflects the degree of association between two features, and the association weight is then calculated based on the average correlation coefficient. Association weight is used to measure the independence of features. The higher the association weight, the weaker the association between the feature and other features, and the stronger the independence. The final weight is calculated using the nerve injury sensitivity weight and the correlation weight. This formula comprehensively considers the sensitivity and independence of features, giving higher weights to features that are sensitive to and highly independent of early neurological injury in patients. By calculating sensitivity weights and correlation weights, and combining them to obtain the final weight, it highlights features that are sensitive to and highly independent of early neurological injury in patients, suppressing the influence of redundant features. Then, it fuses features using the standardized feature vector of the pure electromyographic signal of the abductor digiti minimi muscle and the final weight, obtaining the fused feature value of the pure electromyographic signal of the abductor digiti minimi muscle. The specific algorithm formula is as follows: This integrates multiple basic features into a single comprehensive feature, highlighting the role of important features and reducing interference from redundant features. The features (time domain / frequency domain / nonlinear / coefficient features) corresponding to the nerve injury sensitivity weights are matched with the physiological dimensions of peripheral nerve injury, namely muscle contraction intensity (time domain), nerve conduction frequency characteristics (frequency domain), neuromuscular nonlinear synergy (nonlinear), and muscle fiber activation synergy (coefficient features). Early minor injuries will first cause subtle differences in these physiological links, so the feature sensitivity conforms to the pathophysiological law of nerve injury. The correlation weight is based on the principle that nerve conduction and muscle contraction are independent physiological processes. The Pearson correlation coefficient is used to quantify feature redundancy (such as frequency domain and nonlinear features belonging to different physiological mechanisms, with low redundancy), avoiding overweighting of features of the same physiological dimension. The final weighting formula simultaneously focuses on both physiological difference sensitivity and physiological process independence, covering the core physiological dimensions of neuromuscular injury while avoiding interference from redundant features. Its principle aligns with the physiological changes observed in clinical nerve injuries, effectively improving the accuracy of early injury feature identification. S3.2 Obtain the average power frequency, median frequency, and fusion characteristic value of the pure electromyographic signal of the abductor digiti minimi muscle from known normal individuals in the medical database, and use them as the reference value for the average power frequency of the pure electromyographic signal of the abductor digiti minimi muscle. Median frequency reference value Fusion feature benchmark value The neural conduction integrity index was assessed using the formula based on the average power frequency, average power frequency baseline, median frequency, and median frequency baseline of the pure electromyographic signal of the abductor digiti minimi muscle. The study comprehensively considers the ratio of average power frequency and median frequency to the baseline value of normal individuals, and obtains a comprehensive index through weighted averaging to reflect the integrity of the conduction function of the ulnar nerve and median nerve. Moreover, the nerve conduction integrity index value directly assesses the integrity of nerve conduction function, making up for the limitations of traditional single conduction velocity assessment. Fusion feature value of pure electromyographic signal of abductor digiti minimi muscle Fusion characteristic benchmark value of pure electromyographic signal of abductor digiti minimi muscle Calculate the muscle contraction function index value of the pure electromyographic signal of the abductor digiti minimi muscle. By comparing the fusion characteristics with the baseline values ​​of fusion characteristics in the normal population, an index reflecting the patient's muscle contraction function is obtained, which reflects the synergy between nerve conduction and muscle contraction, and connects the synergy between nerve conduction and muscle contraction, thus making up for the shortcomings of the traditional single assessment dimension. A large amount of clinical data was acquired from a medical database, including the values ​​of nerve conduction integrity and muscle contraction function indices of the pure electromyography (EMG) signal of the abductor digiti minimi muscle in different patients, as well as the actual nerve damage status. Then, multiple regression analysis was used, with the actual severity of nerve damage as the dependent variable and the values ​​of nerve conduction integrity and muscle contraction function indices of the pure EMG signal of the abductor digiti minimi muscle as independent variables, to establish a regression model. The weighting coefficients of the nerve conduction integrity and muscle contraction function indices on the severity of nerve damage were calculated using the regression model. and Then, by combining the nerve conduction integrity index value and muscle contraction function index value of the pure electromyographic signal of the abductor digiti minimi muscle, the comprehensive index value of nerve damage in the abductor digiti minimi muscle can be evaluated. Then, retrieve known normal comprehensive index values ​​of the abductor digiti minimi muscle from the medical database. Range of comprehensive index values ​​for minor nerve damage in the abductor digiti minimi muscle Range of comprehensive index values ​​for moderate nerve injury in the abductor digiti minimi muscle Comprehensive index value of severe nerve damage to the abductor digiti minimi muscle The presence of nerve damage in patients is determined by comparing the comprehensive nerve damage index of the abductor digiti minimi muscle with the normal comprehensive index, the range of the comprehensive index for mild nerve damage, the range of the comprehensive index for moderate nerve damage, and the comprehensive index for severe nerve damage. When the comprehensive nerve damage index of the abductor digiti minimi muscle is greater than or equal to the normal comprehensive index, the patient is considered to have no nerve damage. When the comprehensive nerve damage index of the abductor digiti minimi muscle is within the range of the mild nerve damage comprehensive index, the patient is considered to have mild nerve damage. When the comprehensive nerve damage index of the abductor digiti minimi muscle is within the range of the moderate nerve damage comprehensive index, the patient is considered to have moderate nerve damage. When the comprehensive nerve damage index of the abductor digiti minimi muscle is less than the comprehensive index for severe nerve damage, the patient is considered to have severe nerve damage. By comprehensively considering nerve conduction and muscle contraction function indicators, the severity of nerve damage in patients can be assessed more comprehensively and accurately.

[0020] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing and evaluating electromyographic signals in peripheral nerve injury, characterized in that: The methods and steps include the following: S1. Acquire the patient's raw electromyographic signals using an electromyography sensor, and simultaneously acquire environmental noise reference signals using a noise sensor; S2. The original electromyographic signal is decomposed using a variational mode decomposition algorithm to obtain the set of intrinsic mode function components and the set of center frequencies; S3. Calculate the cross-correlation coefficient between each intrinsic mode function component and the environmental noise reference signal, and filter the effective intrinsic mode function components based on the cross-correlation coefficient; S4. Perform wavelet denoising on the selected effective intrinsic mode function components, obtain the denoised components, and superimpose them to obtain a pure electromyographic signal. S5. Extract time-domain features, frequency-domain features, and nonlinear feature vectors from the pure electromyography signal to obtain the extracted feature vectors; S6. Calculate the Pearson correlation coefficient and the comprehensive nerve injury index based on the extracted feature vectors, and assess the patient's nerve injury status based on the comprehensive nerve injury index.

2. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 1, characterized in that: The specific steps of the variational mode decomposition algorithm in S2 include: Initialize the known set of intrinsic mode function components, the set of center frequencies, and the Lagrange multipliers; The eigenmode function components and center frequency are updated iteratively using the alternating direction multiplier method; The iteration termination condition is determined based on the dynamic convergence threshold. The iteration is terminated when the relative error of the maximum intrinsic mode function component and the absolute error of the maximum center frequency are both less than the dynamic convergence threshold in two consecutive iterations. The set of intrinsic mode function components and the set of center frequencies of the decomposed original electromyographic signal are then output.

3. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 2, characterized in that: The specific steps for setting the dynamic convergence threshold include: The initial maximum threshold is recorded by initializing the convergence threshold. Based on the initial maximum threshold and the current iteration number Achieving a dynamic convergence threshold using the maximum number of iterations .

4. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 1, characterized in that: The specific steps for effective component screening in S3 include: Discretize the intrinsic mode function components and the environmental noise reference signal into discrete signals, and calculate the length of the discrete signals; The cross-correlation function value is calculated based on the length of the discrete signal, the cross-correlation coefficient is calculated based on the cross-correlation function value, and the correlation coefficient of the largest component is obtained by sorting the cross-correlation coefficients. Calculate the adaptive cross-correlation threshold based on the correlation coefficient of the maximum component; Eigenmode function components with cross-correlation coefficients less than the adaptive cross-correlation threshold are identified as valid eigenmode function components.

5. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 4, characterized in that: The specific steps for setting the adaptive cross-correlation threshold include: The cross-correlation coefficients are sorted from highest to lowest to obtain the correlation coefficient of the largest component. The adaptive cross-correlation threshold is calculated using the correlation coefficient of the maximum component. .

6. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 5, characterized in that: The wavelet denoising process in S4 uses db4 wavelet for 3-level decomposition, and the specific steps include: The processed wavelet coefficients are obtained by performing wavelet decomposition on the effective intrinsic mode function components. The iteration step size is estimated based on the highest level detail coefficients of the wavelet, and then the processed wavelet coefficients are inversely transformed to obtain the denoised components.

7. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 5, characterized in that: The time-domain feature vector in S5 includes root mean square value, integral electromyography value, and peak factor. The frequency domain eigenvectors include the average power frequency and the median frequency; Nonlinear characteristics include approximate entropy and sample entropy; It also includes calculating the muscle synergy coefficient as a coefficient feature, which is calculated based on the mean, standard deviation, and covariance of the pure electromyographic signal.

8. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 1, characterized in that: The Pearson correlation coefficient is calculated based on the features extracted in S6. The specific steps include the following: The time domain, frequency domain, nonlinear eigenvectors, and coefficient features are combined in pairs to calculate the Pearson correlation coefficient between the two features.

9. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 1, characterized in that: The specific steps for the comprehensive neurological injury index in S6 include: By standardizing the extracted features using Z-score, a standardized feature vector is obtained. The sensitivity weight for nerve injury was calculated based on the analysis of variance values; the association weight was calculated based on the Pearson correlation coefficient. The final weight is obtained by combining the nerve injury sensitivity weight and the correlation weight; A fusion feature value is generated by fusing standardized feature vectors with final weights, and a comprehensive index of nerve damage is calculated based on the fusion feature value.

10. The method for analyzing and evaluating electromyographic signals of peripheral nerve injury according to claim 1, characterized in that: The specific steps for assessing the patient's nerve damage using S6 are as follows: Obtain known normal comprehensive index values ​​and the range of comprehensive index values ​​for minor nerve damage from the medical database; The presence of nerve damage in a patient is determined by comparing the comprehensive index value of nerve injury with the ranges of normal and mild nerve injury comprehensive index values. When the comprehensive index value of nerve injury is greater than or equal to the normal comprehensive index value, the patient is determined to have no nerve injury. When the comprehensive index value of nerve injury is within the range of the comprehensive index value of mild nerve injury, the patient is determined to have mild nerve injury.

Images