Myoelectric interference removal method based on improved ICEEMDAN

Abstract

The present invention relates to a method for removing myoelectric interference based on improved ICEEMDAN, and belongs to the technical field of vital sign signal processing. The method first performs ICEEMDAN decomposition on the noisy signal and calculates the sample entropy of each component, selects a suitable component for lifting wavelet threshold denoising, and then reconstructs the denoised component and the unprocessed component to complete the denoising process. The present invention can improve the denoising effect of the signal and retain the authenticity of the signal to a greater extent.

Description

Technical Field

[0001] The invention belongs to the technical field of vital sign signal processing and relates to a method for removing myoelectric interference based on improved ICEEMDAN. Background Art

[0002] Electrocardiogram (ECG) contains a wealth of information about heart activity, which is crucial for predicting and diagnosing heart-related diseases. Therefore, ECG measurement is widely used in medical instruments and equipment such as electrocardiographs and electrocardiograph monitors, but it is easily interfered by noise during the measurement and recording process. Myoelectric signals are high-frequency interference caused by muscle contraction. Their frequency is within the ECG frequency band and can easily mix into ECG signals to cause interference. In order to obtain high-quality ECGs and ensure the accuracy of ECG recordings, it is extremely important to suppress myoelectric interference and study its suppression methods.

[0003] At present, ECG electromyographic interference removal algorithms are mainly represented by wavelet transform and empirical mode transform (EMD) and its variants. Traditional wavelet transform relies on Fourier transform, which occupies a lot of system resources and has slow running speed. Therefore, wavelet transform came into being. It has the characteristics of wavelets while breaking the limitations, and has a good effect in signal and image denoising. The advantage of EMD is that it can directly decompose the signal without the need for prior analysis and processing, so it is widely used in the field of denoising. However, EMD cannot accurately provide the physical meaning of each order of intrinsic mode functions (IMF), and there are problems of pseudo-mode and endpoint effects. Reference "Hong Jie, Wang Lu, Shu Junyong, et al. Study on surface electromyographic signal denoising based on EMD wavelet threshold [J]. Journal of Chongqing University of Technology (Natural Science), 2015, 29(08): 124-128" Using the EMD wavelet threshold denoising method, although it combines the advantages of the two methods, the endpoint effect and modal aliasing problems of EMD itself cannot be avoided, and the removal of electromyographic interference is not ideal. Algorithms such as Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and empirical wavelet transform improve the signal decomposition effect, and are therefore also used in ECG signal processing.

[0004] Reference "Xu Yang, Luo Mingzhang, Li Tao. Research on ECG denoising algorithm based on CEEMDAN and wavelet threshold [J]. Modern Electronic Technology, 2018, 41(07): 45-48+53." CEEMDAN wavelet threshold is used for signal denoising, but the problem of pseudo-modality still exists. Reference "Lu Lirong, Wang Jian, Niu Xiaodong. ECG electromyographic interference denoising based on VMD and wavelet threshold [J]. Journal of Sensor Technology, 2020, 33(06): 867-873." The combination of variational mode decomposition (VMD) and other denoising algorithms suppresses electromyographic interference, but since the parameters affecting VMD decomposition are mainly selected based on experience, the effect of different signal decompositions cannot be guaranteed. Although there is an optimized VMD algorithm with adaptive parameter selection, the optimization method is very time-consuming and the optimization effect is unstable, and it is not universal.

[0005] Therefore, in order to improve the removal effect of ECG electromyographic interference, the present invention proposes an electromyographic interference removal method based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN). Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide a method for removing electromyographic interference based on improved ICEEMDAN. Aiming at the problem that the electromyographic interference removal effect in the existing ECG signal is not good, the noisy signal is firstly decomposed by ICEEMDAN and the sample entropy of each component is calculated, and the appropriate component is selected for lifting wavelet threshold denoising, and then the denoising component and the unprocessed component are reconstructed to complete the denoising process, which improves the denoising effect of the signal and retains the authenticity of the signal to a greater extent.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] A method for removing electromyographic interference based on improved ICEEMDAN specifically includes the following steps:

[0009] S1: The ICEEMDAN algorithm is used to adaptively decompose the ECG signal mixed with myoelectric interference, generating several IMFs and residual components, whose frequencies are from high to low; where ICEEMDAN stands for improved adaptive noise complete set empirical mode decomposition;

[0010] S2: Calculate the sample entropy of each group of IMF components. The larger the sample entropy value, the fewer effective components in the signal, and the signal is dominated by noise. Conversely, the smaller the sample entropy, the fewer noise components. After decomposition, the electromyographic interference noise is mainly concentrated in the high-frequency part, and the sample entropy can be used to filter the IMF that needs to be denoised;

[0011] S3: Process the IMF selected by S2 through the lifting wavelet threshold algorithm, and do not process the remaining components;

[0012] S4: Reconstruct the components processed in S3 and the unprocessed components to complete denoising and obtain a denoised ECG signal.

[0013] Further, in step S1, the ICEEMDAN algorithm specifically includes the following steps:

[0014] S101: E1[ω (i) ] is added to the signal s, that is

[0015] s (i) =s+β0E1[ω (i) ] (1)

[0016] Where s={s1,s2,...,s n} is an ECG sequence containing myoelectric interference, with a length of n; ω (i) represents the i-th group of Gaussian white noise added; β0=ε0std(s) / std(E1[ω (i) ]), ε0 represents the noise amplitude parameter, std(·) represents the signal standard deviation estimation operator;

[0017] S102: Calculate s by EMD (i) The local mean of , and calculate the first residual r1 and the first IMF value

[0018] r1= <M(s (i) )> (2)

[0019]

[0020] Among them, M(·) means finding the local mean of the signal, and the operator <·> means taking the mean;

[0021] S103: Calculate the value of the second IMF using local mean decomposition

[0022]

[0023] r2= <M{r1+β1E2[ω (i) ]}> (5)

[0024] S104: According to the formula Calculate the value of the kth IMF in,

[0025] r k = <M{r k-1 +β k-1 E k [ω (i) ]}>,k=3,...,K (6)

[0026] Among them, r k represents the kth residual, β k =ε k std(r k ), ε k represents the noise amplitude parameter, E k (·) means finding the kth IMF component of a signal EMD decomposition, K represents the number of IMF components;

[0027] S105: until the calculation and decomposition are completed, all IMFs and residual components are obtained.

[0028] Furthermore, in step S2, the sample entropy mainly uses the probability of generating a new pattern in the signal to measure the complexity of the time series. The larger the sample entropy, the greater the probability of generating a new pattern and the greater the complexity of the sequence. The calculation of sample entropy does not depend on the length of the data. The specific steps of calculating sample entropy include:

[0029] S201: The time series Y={y(1), y(2), ..., y(N)} of length N is constructed into an m-dimensional vector as follows:

[0030] Y(i)={y(i),y(i+1),...,y(i+m-1)} (7)

[0031] Where, i=1,2,...,N-m+1;

[0032] S202: Define the largest difference between the corresponding elements of Y(i) and Y(j) as the distance d[Y(i),Y(j)], (i≠j), that is,

[0033]

[0034] S203: Given a threshold r (r>0), calculate B i The ratio to the total number of vectors Nm is:

[0035]

[0036] Among them, B i represents the number of d[Y(i),Y(j)]<r;

[0037] S204: average all the results obtained by formula (9), that is:

[0038]

[0039] S205: increase the dimension m by 1, and repeat S201 to S204;

[0040] S206: Theoretically, the sample entropy of this sequence can be expressed as:

[0041]

[0042] However, N cannot be infinite, so the estimated value of sample entropy is:

[0043]

[0044] Where In(·) represents the natural logarithm.

[0045] Further, in step S3, the lifting wavelet threshold algorithm specifically includes the following steps:

[0046] S301: lifting the wavelet using a lifting wavelet scheme;

[0047] S302: multi-layer wavelet decomposition;

[0048] S303: Setting a threshold and performing threshold processing using a soft threshold;

[0049] S304: Wavelet reconstruction to obtain a denoised signal.

[0050] Further, in step S303, the soft threshold function is:

[0051]

[0052] Among them, sgn(·) is the sign function, λ is the threshold, Where σ is the median value of the absolute values ​​of the high-frequency coefficients of each layer of wavelet arranged in ascending order / 0.6745.

[0053] The beneficial effect of the present invention is that: for the problem of removing electromyographic interference, the method of the present invention first decomposes the noisy signal into a series of components through ICEEMDAN, uses sample entropy to screen out suitable components, and reconstructs the denoised signal with the unprocessed components after lifting wavelet soft threshold denoising. The present invention can adaptively decompose the signal, improve the decomposition effect in each EMD improved algorithm, improve the stability of the decomposition ability of different signals, improve the denoising effect of the signal, and retain the authenticity of the signal to a greater extent.

[0054] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0056] Figure 1 It is a denoising flow chart of the myoelectric interference removal method based on improved ICEEMDAN of the present invention;

[0057] Figure 2 The original ECG and ECG with 20dB electromyography interference;

[0058] Figure 3 are IMF and residual components;

[0059] Figure 4 is the sample entropy curve;

[0060] Figure 5 To compare the reconstructed ECG No. 107 and the original ECG No. 107;

[0061] Figure 6 Comparison chart of reconstructed ECG No. 119 and original ECG No. 119. DETAILED DESCRIPTION

[0062] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0063] See also Figure 1 to Figure 6For ECG signals with electromyographic interference, using ICEEMDAN to decompose the signal can reduce pseudo-modes and better separate signals and noise. After the signal is decomposed, a series of IMFs can be obtained, and each group of IMFs contains both useful signals and noise. Therefore, removing a certain group of IMFs alone will result in the loss of useful signals. Lifting the wavelet threshold is not ideal for processing nonlinear and non-stationary signals. Based on the advantages and disadvantages of the two methods, the present invention improves the ICEEMDAN algorithm. First, ICEEMDAN is used to adaptively decompose the noisy ECG signal to generate a series of IMF components. In order to screen out the noise-dominated IMF components, the sample entropy value of each component is calculated separately, and then the screened components are denoised using the lifted wavelet threshold, thereby minimizing signal distortion while suppressing electromyographic interference. Figure 1 As shown, the specific steps of the algorithm are as follows:

[0064] 1) ICEEMDAN is used to adaptively decompose ECG signals mixed with myoelectric interference to generate several IMFs and residual components, whose frequencies are from high to low.

[0065] 2) Calculate the sample entropy of each group separately. The larger the sample entropy value, the less effective components in the signal, and the signal is dominated by noise. Conversely, the smaller the sample entropy, the less noise components. After decomposition, the electromyographic interference noise is mainly concentrated in the high-frequency part, and the sample entropy can be used to screen the IMF that needs to be processed.

[0066] 3) The selected IMFs are processed by lifting the wavelet threshold, and the remaining components are not processed.

[0067] By reconstructing the processed components and the unprocessed components in 3), denoising can be completed to obtain a relatively pure ECG.

[0068] In order to measure the denoising effect of the algorithm, the present invention adopts three standards, signal-to-noise ratio, root mean square error and autocorrelation coefficient, to evaluate the noise component contained in the denoised signal, the discrete degree between the denoised signal and the original ECG signal, and the correlation degree before and after denoising.

[0069] Signal-to-Noise Ratio:

[0070]

[0071] Root Mean Square Error:

[0072]

[0073] Autocorrelation coefficient:

[0074]

[0075] Among them, x and x′ represent the original ECG sequence without electromyographic interference and the denoised and reconstructed ECG sequence, respectively. and are the means of sequences x and x', respectively. x(i) and x'(i) represent the i-th element of the original ECG sequence and the denoised ECG sequence, respectively. The larger the SNR, the smaller the RMSE, and the larger the AC, the stronger the denoising ability.

[0076] Comparative simulation verification:

[0077] In order to verify the effectiveness of the method of the present invention in removing ECG electromyographic interference, ECG signals No. 109, 107 and 100 in the "MIT-BIHArrhythmia Database" were selected as original ECG signals, and Gaussian white noise was added to simulate electromyographic interference. The number of sampling points of the signal is 4096, and the sampling frequency is 360Hz. In order to better test and evaluate the denoising method, this embodiment uses three different signal-to-noise ratios of 5dB, 10dB, and 20dB random Gaussian white noise to simulate electromyographic interference, and superimposes ECG signals No. 100, 107 and 109 to generate a variety of signal-noise mixed ECGs, and compares traditional wavelet threshold denoising, EMD denoising, CEEMDAN-wavelet threshold denoising, VMD-wavelet threshold denoising with the method of this paper, and measures and evaluates the effect of the denoising method from three aspects: SNR, RMSE and AC.

[0078] First, taking ECG No. 107 with 20dB electromyographic interference as an example, the experimental results of several methods are compared in detail.

[0079] 1) 107 signal electromyography interference denoising experiment

[0080] The original ECG No. 107 and 20dB electromyographic interference (Gaussian white noise) signal-noise mixed ECG waveform is as follows Figure 2 As shown, Figure 2 (a) is the original ECG, Figure 2 (b) is an ECG with 20dB electromyographic interference. After the signal-noise mixed ECG is decomposed by ICEEMDAN, 10 IMFs and 1 residual component are obtained, as shown in Figure 3 shown.

[0081] Calculate the sample entropy for each component, such as Figure 4As shown. The sample entropy parameters used in this experiment are: embedding dimension is 2, similarity tolerance is 0.15sd, where sd is the sequence standard deviation (embedding dimension is generally 1 or 2, similarity tolerance is usually 0.1sd~0.25sd). According to the relationship between the two, it can be seen that the noise is mainly concentrated in the first and second components, and the noise-dominated components are denoised by lifting the wavelet threshold. This experiment uses the db4 wavelet to complete the 4-layer decomposition. The selected first and second components are denoised by lifting the wavelet soft threshold, and then the denoised first, second components and the remaining components are reconstructed. The comparison between the ECG No. 107 finally reconstructed by the method of the present invention and the original ECG No. 107 is shown in the figure below. Figure 5 (e) as shown.

[0082] 2) Analysis of denoising effect

[0083] 2.1) Comparison of denoising effect of ECG No. 107 with 20dB electromyography interference

[0084] In order to further verify the denoising performance of the proposed method, wavelet threshold, EMD, CEEMDAN-wavelet threshold, VMD-wavelet threshold and the proposed denoising method are used for comparative experiments. Figure 5 As shown, Figure 5 (a) is the comparison between the ECG reconstructed by wavelet threshold and the original ECG. Figure 5 (a) When the wavelet threshold method is used to remove electromyographic interference, signal distortion occurs, the curve is distorted, and the effect is the worst, which destroys the integrity of the signal; Figure 5 (b) is the comparison diagram of EMD reconstructed ECG and original ECG. Figure 5 In (b), the EMD method has a better effect on removing the noise of the baseline part when removing electromyographic interference than the wavelet threshold method, but the peak part is seriously missing and even changes the original signal. This is related to the direct removal of several IMFs with the most noise content after EMD decomposition; Figure 5 (c) is the comparison between the ECG reconstructed by CEEMDAN-wavelet threshold and the original ECG. Figure 5 In (c), the effect of CEEMDAN combined with wavelet threshold to remove EMG interference is better than wavelet threshold denoising and EMD denoising, but there are more missing parts in the peak part and obvious signal distortion at the top of the peak; Figure 5 (d) is the comparison diagram of VMD-wavelet threshold reconstructed ECG and original ECG. Figure 5 In (d), the VMD-wavelet threshold denoising effect is better than the previous methods. The denoised ECG is basically consistent with the original ECG, but each peak part is distorted; Figure 5 (e) is a comparison diagram of the ECG reconstructed by the method of the present invention and the original ECG, Figure 5In (e), the method of the present invention has the best effect in removing myoelectric interference, with almost no peak loss, retaining the characteristic points of the signal, and effectively suppressing the burrs in the baseline part. The comparison of SNR, RMSE and AC values ​​after denoising is shown in Table 1.

[0085] Table 1 Comparison of denoising performance under myoelectric interference with different signal-to-noise ratios of ECG No. 107

[0086]

[0087] 2.2) Comparison of ECG denoising effects under myoelectric interference with different signal-to-noise ratios

[0088] In order to test the denoising effect under different signal-to-noise ratios of electromyographic interference, the experiment selected ECG No. 107, No. 100 and No. 109 to test the denoising effect under 5dB, 10dB and 20dB electromyographic interference (random Gaussian white noise). The comparison of SNR, RMSE and AC values ​​after denoising is shown in Table 1, Table 2 and Table 3.

[0089] Table 2 Comparison of denoising performance under myoelectric interference with different signal-to-noise ratios of ECG No. 100

[0090]

[0091]

[0092] Table 3 Comparison of denoising performance under myoelectric interference with different signal-to-noise ratios of ECG No. 109

[0093]

[0094] As can be seen from Tables 1 to 3, under different signal-to-noise ratios of electromyographic interference, compared with wavelet threshold denoising, EMD denoising and CEEMDAN-wavelet threshold denoising, the proposed method has the largest SNR, the smallest RMSE, the largest AC and the best denoising effect after denoising. For ECG No. 107 with 5dB electromyographic interference, the denoising effect of the proposed method is slightly worse than that of VMD-wavelet threshold denoising. VMD uses the alternating direction multiplier method to effectively optimize the variational model, making the model relatively more robust to sampling noise. However, the decomposition of VMD requires the pre-defined modal number, and there are differences in the decomposition modal numbers of different signals, which leads to the need to explore the decomposition modal number when using the VMD method. How to determine an optimal decomposition modal number requires further research. Overall, the proposed method can adaptively decompose the signal without artificially determining the modal number, has the most stable performance under different conditions, has the best denoising performance, has the strongest ability to remove electromyographic interference, and the ECG after denoising is most similar to the original ECG.

[0095] 2.3) Comparison of different ECG denoising effects with 20dB EMG interference

[0096] In order to improve the credibility and accuracy, this experiment selected ECG No. 119, No. 116, and No. 102 as the original signal, and added random Gaussian white noise with a signal-to-noise ratio of 20dB to simulate electromyographic interference. The experiment was repeated, and the denoising effect is shown in Table 4. Figure 6 As shown in the figure, the denoising effect comparison chart of ECG No. 119 is presented for reference, where: Figure 6 (a) is the comparison between the ECG reconstructed by wavelet threshold and the original ECG. Figure 6 (b) is the comparison diagram of EMD reconstructed ECG and original ECG. Figure 6 (c) Comparison between the ECG reconstructed by CEEMDAN-wavelet threshold and the original ECG. Figure 6 (d) is the comparison diagram of VMD-wavelet threshold reconstructed ECG and original ECG. Figure 6 (e) is a comparison chart of the reconstructed ECG and the original ECG by the method of the present invention. It can be seen from the experimental results that, no matter it is ECG No. 119, No. 116 or No. 102, among the five algorithms, the SNR and AC after denoising by the algorithm of this paper are the largest, while the RMSE is the smallest. The effect of removing electromyographic interference by the method of the present invention is the best and has universal applicability.

[0097] Table 4 Performance comparison of different ECG denoising algorithms

[0098]

[0099]

[0100] In summary, the method of the present invention is compared with four denoising methods, namely wavelet threshold, EMD, CEEMDAN-wavelet threshold, and VMD-wavelet threshold. The simulation results show that in the denoising process of ECG No. 107 with 5dB electromyographic interference, the denoising performance of the method of the present invention is slightly worse than that of VMD-wavelet threshold denoising. However, since the parameters of VMD decomposition are set by experience, the number of decomposition modes has a great influence on the decomposition effect, so it is necessary to set a suitable number of decomposition modes before decomposition. Overall, the method of the present invention can adaptively decompose the signal, and has the best decomposition effect among the various EMD improved algorithms. The decomposition ability of different signals is more stable, the denoising effect of the signal is better, and the authenticity of the signal is retained to a greater extent.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for removing electromyographic interference based on improved ICEEMDAN, characterized in that: The method specifically comprises the following steps: S1: The ICEEMDAN algorithm is used to adaptively decompose the ECG signal mixed with myoelectric interference to generate several IMFs and residual components; where ICEEMDAN stands for improved adaptive noise complete set empirical mode decomposition; The ICEEMDAN algorithm specifically includes the following steps: S101: E1[ω (i) ] is added to the signal s, that is s (i) =s+β0E1[ω (i) ] (1) Where s={s1,s2,...,s n } is an ECG sequence containing myoelectric interference, with a length of n; ω (i) represents the i-th group of Gaussian white noise added; β0=ε0std(s) / std(E1[ω (i) ]), ε0 represents the noise amplitude parameter, std(·) represents the signal standard deviation estimation operator; S102: Calculate s by EMD (i) The local mean of , and calculate the first residual r1 and the first IMF value r1= <M(s (i) )> (2) Among them, M(·) means finding the local mean of the signal, and the operator <·> means taking the mean; S103: Calculate the value of the second IMF using local mean decomposition r2= <M{r1+β1E2[ω (i) ]}> (5) S104: According to the formula Calculate the value of the kth IMF in, r k = <M{r k-1 +b k-1 E k [oh (i) ]}>,k=3,...,K (6) Among them, r k represents the kth residual, β k =ε k std(r k ), ε k represents the noise amplitude parameter, E k (·) means finding the kth IMF component of a signal EMD decomposition, K represents the number of IMF components; S105: until the calculation and decomposition are completed, all IMFs and residual components are obtained; S2: Calculate the sample entropy of each group of IMF components respectively, and filter the IMF that needs to be denoised by the sample entropy; S3: Process the IMF selected by S2 through the lifting wavelet threshold algorithm, and do not process the remaining components; S4: Reconstruct the processed components and the unprocessed components in S3 to obtain the denoised ECG signal.

2. The improved ICEEMDAN electromyographic interference removal method according to claim 1, characterized in that: In step S2, calculating the sample entropy specifically includes the following steps: S201: The time series Y={y(1), y(2), ..., y(N)} of length N is constructed into an m-dimensional vector as follows: Y(i)={y(i),y(i+1),...,y(i+m-1)} (7) Where, i=1,2,...,N-m+1; S202: Define the largest difference between the corresponding elements of Y(i) and Y(j) as the distance d[Y(i),Y(j)], (i≠j), that is, S203: Given a threshold r (r>0), calculate B i The ratio to the total number of vectors Nm is: Among them, B i represents the number of d[Y(i),Y(j)]<r; S204: average all the results obtained by formula (9), that is: S205: increase the dimension m by 1, and repeat S201 to S204; S206: Theoretically, the sample entropy of this sequence is expressed as: However, N cannot be infinite, so the estimated value of sample entropy is: Where In(·) represents the natural logarithm.

3. The method for removing electromyographic interference of the improved ICEEMDAN according to claim 1, characterized in that: In step S3, the lifting wavelet threshold algorithm specifically includes the following steps: S301: lifting the wavelet using a lifting wavelet scheme; S302: multi-layer wavelet decomposition; S303: Setting a threshold and performing threshold processing using a soft threshold; S304: Wavelet reconstruction to obtain a denoised signal.

4. The method for removing electromyographic interference of the improved ICEEMDAN according to claim 3, characterized in that: In step S303, the soft threshold function is: Among them, sgn(·) is the sign function and λ is the threshold.

Images