Surface electromyography method based on FastICA and contractile force

Abstract

This invention relates to a surface electromyography (SEMG) decomposition method based on FastICA and contractile force, belonging to the field of EEMG signal processing technology. First, a high-density array-type SEMG signal is pre-processed by filtering. Second, FastICA is used to decompose low-level contractile force EEMG signals, extracting the motor unit action potential sequence matrix from this level of muscle force signal. Then, FastICA is used to decompose high-level contractile force signals, and the extracted motor unit action potential sequence is subtracted from the action potential sequence matrix of the previous level. Finally, the above decomposition algorithm is iterated and optimized, with the constraint of minimizing the distance error between adjacent level potential sequence matrices. The improved decomposition method has high accuracy, is simple to implement, and can decompose a large number of motor units at medium to high contractile force levels.

Description

Technical Field

[0001] This invention relates to a surface electromyography decomposition method based on FastICA and contractile force, belonging to the field of electromyography signal processing technology. Background Technology

[0002] Electromyography (EMG) signals are signals generated by the action potential sequences of all motor units involved in muscle control during muscle contraction. These sequences are filtered by muscle, subcutaneous fat, and skin tissue, and then superimposed at the electrodes. EMG signal decomposition involves breaking down the acquired signal into individual motor unit action potential sequences. By analyzing these sequences, the activity state of motor neurons can be determined. EMG signal decomposition technology has significant applications in motor control and the diagnosis of clinical neurological diseases. Based on the acquisition method, EMG signals can be divided into two types: surface EMG signals and needle electrode EMG signals. Compared to needle electrode acquisition, surface electrodes adhere well to the skin, capturing a wealth of spatiotemporal information about motor units. Furthermore, surface electrodes are non-invasive and simple to operate, making them commonly used for EMG signal detection.

[0003] Current research generally employs two methods for decomposing surface electromyography (SEM) signals: one is a cluster-based template matching method, and the other is blind source separation. Many studies have proposed that blind source separation can be used to decompose multi-channel SEM signals. Among these blind source separation techniques, rapid independent component analysis (XComponent Analysis) is the most widely used and has achieved good results. However, this is limited to the decomposition of SEM signals at low contractile force levels. At medium to high contractile force levels, due to severe waveform superposition of motor units in the SEM signals, it is difficult to distinguish between them. Furthermore, the filtering effect of the surface electrode array causes these methods to extract only a very small number of motor units from high-density SEM signals, resulting in unstable decomposition results and low accuracy. Therefore, the decomposition of high-density SEM signals remains a challenging problem when waveform superposition is significant. Summary of the Invention

[0004] In view of the above problems, the object of the present invention is to provide a high-density surface electromyography signal decomposition method based on the correlation between FastICA and contractile force level.

[0005] This invention provides a surface electromyography method based on FastICA and contractile force, characterized in that it includes step 1:

[0006] High-density surface electromyography signals were collected under different levels of contractile force.

[0007] Step 2: Preprocess the raw electromyographic signals;

[0008] Step 3: Decompose the surface electromyographic signals of each level of contractile force using rapid independent component analysis;

[0009] Step 4: Determine the contraction force level. When the contraction force level is low, directly output the surface electromyography signal decomposition results. When the contraction force level is medium to high, it is necessary to extract the motor unit action potential sequence under 40% MVC and medium to high contraction force at the same time as decomposition. The low level is 10% to 40% MVC, and the medium to high level is 50% to 100% MVC.

[0010] Step 5: After decomposing the electromyographic signals using FastICA, extract the motor unit action potential sequence matrix;

[0011] Step 6: Perform the Fast ICA decomposition method from Step 3 to decompose the surface electromyography signals under medium to high levels of contractile force;

[0012] Step 7: Extract the action potential sequence matrix of motor units under high-level contractile force signals based on Step 5.

[0013] Step 8: Based on the characteristics of high-density sEMG signals during constant force contraction, the signals generated by the same MU at different firing times have certain similarities between two adjacent contraction force levels. Since FastICA has a high accuracy rate in decomposing surface electromyography signals of low-level contraction forces, in order to improve the accuracy of surface electromyography signal decomposition of medium and high-level contraction forces, the action potential sequence extracted from each medium and high-level contraction force and its previous level contraction force signal is made as close as possible, and the difference between the action potential sequence matrices of motor units under two adjacent levels is calculated.

[0014] Step 9: Iterate through steps 6, 7, and 8. Each time Fast ICA is used to decompose the surface electromyography signals of medium- to high-level contractile forces, a motor unit sequence matrix can be extracted. Using different sequence matrices Subtract the contractile force S0 from the previous level and select the electromyographic signal decomposition result corresponding to the smallest error e;

[0015] Step 10: Repeat steps 5, 6, 7, 8, and 9 to sequentially decompose the surface electromyographic signals of medium to high contractile force levels.

[0016] Step 3 includes the first step: performing mean removal and whitening processing on the denoised electromyographic signals;

[0017] Step 2: Initialize the decomposition vector W j (i), take ||W j(0)||=1, i is the number of independent components to be extracted, i=1,...,k, j is the source signal index to be decomposed, let j=1;

[0018] Step 3: Iteratively calculate the p+1 independent components of W.

[0019] Where G is a non-polynomial function, here we take...

[0020] G(x) = log(cosh(x));

[0021] Step 4: Orthogonalize independent components, calculation formula

[0022]

[0023] Step 5: Normalization, the calculation formula is as follows:

[0024] w j (p+1)=w j (p+1) / ||w j (p+1)|| 2 ;

[0025] Step 6: If w j If the convergence fails, return to step 3, where W = [w1, w2, ..., w k ] T .

[0026] The decomposed surface electromyographic signal can be represented as:

[0027]

[0028] Where a mn (τ) represents the waveform of the nth motion unit in channel m, τ represents the delay of the firing sequence, s n (t-τ) represents the firing time sequence of the nth motion unit, a mn (τ)s n (t-τ) represents the motor unit action potential sequence of the nth motor unit in the m channel. Before decomposing the surface electromyography signal of the high contractile force level, according to the formula (4) a mn (τ)s n (t-τ) Extract all motion unit action potential sequences decomposed under the contractile force signal of the previous adjacent level, and form the motion unit action potential sequence matrix S0.

[0029] The formula for the difference is:

[0030] The beneficial effects of this invention are:

[0031] This invention provides a surface electromyography (sEMG) decomposition method based on FastICA and contractile force. Addressing the difficulty in decomposing high-density sEMG signals under medium to high levels of contractile force, this invention offers a method based on FastICA and the correlation between contractile force levels. According to the characteristics of multi-channel sEMG signals during constant force contraction—signals generated by the same motor unit (MU) at different force levels exhibit a certain correlation, and this correlation is higher in adjacent levels—FastICA is improved by extracting the motor unit action potential sequence matrix between two adjacent contractile force levels. The sequence matrix of the lower level acts as a constraint matrix. The principle is to minimize the error of the motor unit action potential sequence matrix between the two levels, and the FastICA decomposition algorithm is iterated. This allows FastICA to obtain the optimal decomposition result under the constraint of the action potential sequence matrix, resulting in a more accurate decomposition of sEMG signals under medium to high levels of contractile force. Attached Figure Description

[0032] Figure 1 This is a flowchart of the surface electromyography method based on FastICA and contractile force according to the present invention.

[0033] Figure 2 This is a schematic diagram of the electromyographic signal decomposition results of the surface electromyographic decomposition method based on FastICA and contractile force according to the present invention.

[0034] Figure 3 This is a schematic diagram of another decomposition result of the electromyographic signal from the surface electromyographic decomposition method based on FastICA and contractile force according to the present invention.

[0035] Figure 4 This is a schematic diagram comparing the results of the surface electromyography method based on FastICA and contractile force according to the present invention. Detailed Implementation

[0036] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0037] like Figures 1 to 4 As shown: This invention provides a surface electromyography (sEMG) decomposition method based on FastICA and contractile force. On the basis of the original FastICA decomposition method, based on the correlation of the action potential sequences of motor units of adjacent contractile forces, a process of subtracting action potential sequences between adjacent contractile forces is introduced, which improves the accuracy of sEMG signal decomposition of medium and high-level contractile forces to a certain extent and can decompose more motor units.

[0038] The electromyographic signals used in this study were collected at the University of Houston and the Texas Medical Center. Eight healthy subjects participated in the data collection. High-density surface electromyographic signals at different levels of contractile force were collected on the skin surface of the biceps brachii of the healthy subjects using a high-density array electrode with 128 channels. The subjects' biceps brachii performed isometric voluntary contractions, and the contractile force was controlled at nine progressive levels of maximum voluntary contractile force (MVC), namely 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, and 100% of the maximum voluntary contractile force.

[0039] Step 2: Preprocess the raw electromyography signal, remove baseline drift by moving average method, and denoise by Butterworth filtering and notch filtering.

[0040] Step 3: Decompose the surface electromyography signals of each level of contractile force using the Fast Independent Component Analysis (FICA) method. The FastICA decomposition is performed in the following six steps:

[0041] The first step is to perform mean removal and whitening processing on the denoised electromyographic signals.

[0042] The second step is to initialize the decomposition vector W. j (i), take ||W j (0)||=1, i is the number of independent components to be extracted, i=1,...,k, j is the source signal number to be decomposed, let j=1.

[0043] The third step is to iteratively calculate the p+1 independent component forces of W, using the following formula:

[0044]

[0045] Where G is a non-polynomial function, here we take G(x) = log(cosh(x)).

[0046] The fourth step is to orthogonalize the independent components. The calculation formula is as follows:

[0047]

[0048] Step 5: Normalization, calculated using the following formula:

[0049] w j (p+1)=w j (p+1) / ||w j (p+1)|| 2 (3)

[0050] Step 6, if w j If the convergence fails, return to step 3, where W = [w1, w2, ..., w k ] T .

[0051] Step 4: Determine the contraction force level. When the contraction force level is low, directly output the surface electromyography signal decomposition results. When the contraction force level is medium to high, the motor unit action potential sequence under 40% MVC and medium to high contraction force needs to be extracted at the same time. The low level is 10% to 40% MVC, and the medium to high level is 50% to 100% MVC.

[0052] Step 5: After decomposing the electromyographic signal using FastICA, extract the motor unit action potential sequence matrix. The decomposed surface electromyographic signal can be represented as:

[0053]

[0054] Where a mn (τ) represents the waveform of the nth motion unit in channel m, τ represents the delay of the firing sequence, s n (t-τ) represents the firing time sequence of the nth motion unit, a mn (τ)s n (t-τ) represents the motor unit action potential sequence of the nth motor unit in the m channel. Before decomposing the surface electromyography signal of the medium-high contractile force level (50%~100% MVC), according to a in formula (4) mn (τ)s n (t-τ) Extract all motor unit action potential sequences decomposed under the contractile force signal of the previous adjacent level to form a motor unit action potential sequence matrix S0. For example, if we want to decompose the surface electromyography signal of 50% of the maximum voluntary contractile force, we need to extract the motor unit action potential sequence matrix after decomposing the electromyography signal of the previous level, that is, 40% of the maximum voluntary contractile force.

[0055] Step 6: Execute the Fast ICA decomposition method from Step 3 to decompose the surface electromyography signal under a certain medium-to-high level of contractile force.

[0056] Step 7, according to a in formula (4) mn (τ)s n The (t-τ) term is used to extract the action potential sequence matrix of the kinematic unit under the medium-to-high level contractile force signal.

[0057] Step 8: Next, make the action potential sequences extracted from each medium-to-high level contraction force and its adjacent lower level contraction force signals as close as possible. Then, subtract and sum the action potential sequence matrices of two adjacent levels. The formula for subtraction is as follows:

[0058]

[0059] Specifically, equation (5) can be written as:

[0060]

[0061] Where, x mn This represents the action potential sequence of the nth moving unit in matrix S0 within channel m. In the m-channel, the matrix The action potential sequence of the nth motor unit

[0062] Step 9: Iteratively calculate steps 5, 6, and 7. After each decomposition of the surface electromyography signal for medium-to-high grade contractile forces using Fast ICA, a motor unit sequence matrix can be extracted. Using different sequence matrices The electromyographic signal decomposition result corresponding to the smallest error e(m) is selected by subtracting the previous level of contractile force S0.

[0063] Step 10: Repeat steps 5, 6, 7, 8, and 9 to sequentially decompose the surface electromyographic signals under 50%, 60%, 70%, 80%, and 100% of the maximum voluntary contractile force.

[0064] Through the above steps, electromyographic signals at different contractile force levels are decomposed step by step. To verify the effectiveness of the above method, the decomposition results of the method of the present invention are compared with those of the traditional FastICA method. Figure 2 and Figure 3 This presentation shows the decomposition results of a certain channel's electromyography (EMG) signal under 60% MVC in a certain subject, including motor units, motor unit firing time series, and residual signals, synthesizing data from 8 subjects. Figure 4 The average number of kinematic units decomposed by the FastICA method and the method of this invention under 70% MVC contraction force was statistically analyzed, along with the average accuracy. Here, the accuracy refers to the overlap rate of the kinematic unit firing sequences in the 128 channels that collect the contraction force signal at this level, divided into odd and even channels. The experimental results show that the method of this invention decomposes the number of kinematic units and the accuracy to a certain extent, and the residual signal after decomposition is less.

[0065] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A surface electromyography method based on FastICA and contractile force, characterized in that: Including step 1: Acquiring high-density surface electromyography signals under different levels of contractile force; Step 2: Preprocess the surface electromyography signals; Step 3: Decompose the surface electromyographic signals of each level of contractile force using FastICA; Step 4: Determine the contraction force level. When the contraction force level is low, directly output the surface electromyography signal decomposition results. When the contraction force level is medium to high, it is necessary to extract the motor unit action potential sequence under 40% MVC and medium to high contraction force at the same time as decomposition. The low level is 10% to 40% MVC, and the medium to high level is 50% to 100% MVC. Step 5: After decomposing the electromyographic signals using FastICA, extract the motor unit action potential sequence matrix; Step 6: Perform the FastICA decomposition method from Step 3 to decompose the surface electromyography signals under medium to high levels of contractile force; Step 7: Extract the action potential sequence matrix of motor units under high-level contractile force signals based on Step 5. Step 8: Based on the characteristics of high-density surface electromyography (EMG) signals during constant force contraction, the signals generated by the same motor unit at different release times have certain similarities between two adjacent contraction force levels. Since FastICA has a high accuracy rate in decomposing surface EMG signals of low-level contraction forces, in order to improve the accuracy of surface EMG signal decomposition of medium and high-level contraction forces, the action potential sequences extracted from each medium and high-level contraction force and its previous level contraction force signal are made as close as possible. The difference between the action potential sequence matrices of motor units under two adjacent levels is then calculated. Step 9: Iterate through steps 6, 7, and 8. Each time FastICA is used to decompose the surface electromyography signals of medium- to high-level contractile forces, a motor unit sequence matrix can be extracted. After each decomposition, the current motion unit sequence matrix is... The difference between the motion sequence identity matrix S0 under the contractile force of the previous level is calculated using the following formula: Finally, the electromyographic signal decomposition result corresponding to the smallest error e is selected; Step 10: Repeat steps 5, 6, 7, 8, and 9 to sequentially decompose the surface electromyographic signals of medium to high contractile force levels.

2. The surface electromyography method based on FastICA and contractile force according to claim 1, characterized in that: Step 3 includes the first step: performing mean removal and whitening processing on the denoised electromyographic signals; Step 2: Initialize the decomposition vector W j (i), take ||W j (0)||=1, i is the number of independent components to be extracted, i=1,...,k, j is the source signal number to be decomposed, let j=1; Step 3: Iteratively calculate the p+1 independent components of W. Where G is a non-polynomial function, here we take G(x) = log(cosh(x)); Step 4: Orthogonalize independent components, calculation formula Step 5: Normalization, the calculation formula is as follows: w j (p+1)=w j (p+1) / ||w j (p+1)|| 2 ; (3) Step 6: If w j If the convergence fails, return to step 3, where W = [w1, w2, ..., w k ] T .

3. The surface electromyography method based on FastICA and contractile force according to claim 1, characterized in that: The decomposed surface electromyographic signal can be represented as: where a mn (τ) represents the waveform of the nth motor unit in the m channels, τ represents the delay of the firing sequence, s n (t - τ) represents the firing time series of the nth motor unit, a mn (τ)s n (t - τ) represents the motor unit action potential sequence of the nth motor unit in the m channels. Before decomposing the surface electromyogram signal of the high contraction force level, according to a in formula (4) mn (τ)s n (t - τ) extracts all the motor unit action potential sequences decomposed under the previous adjacent level of contraction force signal to form the motor unit action potential sequence matrix S0.

4. The surface electromyography method based on FastICA and contractile force according to claim 3, characterized in that: Step 7 is based on formula (4) in step 5, where a mn (τ)s n The (t-τ) term is used to extract the action potential sequence matrix of motor units under medium-to-high-level contractile force signals.

Images