A fetal electrocardiogram detection method based on kernel independent component analysis

Through the method based on nuclear independent component analysis, spheroidization processing and cost function minimization, the difficulty of detecting fetal ECG signals under maternal ECG interference is solved, and the clear extraction and accurate detection of fetal ECG signals is achieved, and the detection is improved robustness and success rate.

CN114732417BActive Publication Date: 2024-11-22JIANGSU UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210249288.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-14
Publication Date
2024-11-22
Estimated Expiration
2042-03-14

AI Technical Summary

Technical Problem

During the transmission process, the fetal ECG signal is severely disturbed by the maternal ECG signal and the signal is weak. Conventional signal filtering methods cannot extract clear and stable fetal ECG signal, which makes it difficult to judge the fetal intrauterine status.

Method used

Using a method based on kernel independent component analysis, the fetal ECG signal spheroidization process, calculation of the center Gram matrix and minimizing cost function are used to extract the fetal ECG signal, and the nuclear independent criterion and covariance operator of the regenerated nuclear Hilbert space are used. Spectrum, adapt to different sources of distribution and improve detection success rate.

Benefits of technology

It effectively eliminates maternal electrocardiogram interference, improves the detection accuracy and success rate of fetal electrocardiogram signals, is suitable for different application occasions, and improves the robustness of fetal electrocardiogram detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114732417B_ABST
    Figure CN114732417B_ABST
Patent Text Reader

Abstract

The present invention discloses a fetal electrocardiogram detection method based on kernel independent component analysis, adopts a signal separation method of kernel independent component analysis, and separates fetal electrocardiogram signals from the electrocardiogram signals of the abdominal wall of pregnant women based on this method. The present invention adopts kernel independent component analysis instead of independent component analysis, which can adapt to different signal source distributions, so that the algorithm is more robust to various application scenarios. On the other hand, kernel independent component analysis provides a new solution to the signal separation problem, which is not based on a single nonlinear function, but on a candidate nonlinear function space. The regeneration characteristics of the regenerating kernel Hilbert function space are introduced into the construction process of the cost function, and the approximate value of the independent cost function is obtained by effectively searching the entire regenerating kernel Hilbert function space, thereby improving the robustness of the fetal electrocardiogram signal extraction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of maternal and infant medical health, and specifically is a fetal electrocardiogram detection method based on kernel independent component analysis. Background Art

[0002] With the rapid development of modern electronic information technology, electronic fetal monitoring has rapidly become popular and has become an essential means of obstetric examination. Changes in fetal heart function are a manifestation of the regulatory function of the central nervous system. The fetal central nervous system is the organ with the least reserve capacity for the deterioration of the intrauterine environment, that is, it lacks the most tolerance to hypoxia, and once it is damaged, it will leave lifelong sequelae. Fetal electrocardiogram can promptly reflect fetal intrauterine distress. As the gestational age increases, its manifestation becomes more obvious. For example, abnormal changes in the ST segment of the fetal electrocardiogram are accurate and reliable objective indicators for judging fetal hypoxia and prognosis. Continuous fetal electrocardiogram monitoring can assist physicians in more accurately judging the intrauterine status of the fetus. The true instantaneous heart rate can also be obtained from the fetal electrocardiogram, and the quantitative parameters of fetal heart rate variability can be obtained.

[0003] Although transabdominal fetal ECG contains rich information and is safe and convenient, it is not widely used in clinical practice. The fetal ECG signal must pass through amniotic fluid, uterine myometrium, abdominal muscles and fat tissues to be transmitted to the body surface. It is already very weak and is seriously interfered by the maternal ECG. The fetal ECG amplitude is only about 10-50μV, while the maternal ECG can reach 800μV; about 10%-30% of the fetal ECG in the time domain waveform is mixed with the maternal ECG, and the two also overlap mostly in the frequency domain; coupled with interference such as maternal breathing, uterine contraction, 50Hz power frequency, non-Gaussian noise, etc., the entire abdominal wall ECG signal has strong non-stationarity; there is also the problem of nonlinear distortion in the ECG signal transmission process. Faced with the above problems, it is impossible to extract clear and stable fetal ECG signals using conventional signal filtering methods. Summary of the invention

[0004] In view of the above-mentioned technical problems, the present invention designs a fetal electrocardiogram detection method based on kernel independent component analysis.

[0005] The objective of the present invention is achieved through the following technical solution: a fetal electrocardiogram detection method based on kernel independent component analysis, characterized in that it comprises the following steps:

[0006] Step 1: Import multi-channel ECG signal X i (n), i = 1, 2, ..., m, n = 1, 2, ..., N, where i is the channel number and n is the sample number, i.e. there are m channels in total and each channel has N samples;

[0007] Step 2: Multi-channel ECG signal X i (n) Spheroidization is performed to obtain signal Z i (n), so that the signal Zi The covariance matrix of (n) becomes the identity matrix;

[0008] Step 3: From signal Z i (n) Calculate the central Gram matrix G i ,i=1,2,…,m, the signal of each channel calculates its corresponding Gram matrix;

[0009] Step 4: From the central Gram matrix G i Calculate the cost function C(W), where W is the unmixing coefficient;

[0010] Step 5: Minimize the cost function C(W) and find the optimal value of W;

[0011] Step 6: Extract the fetal ECG signal Y through the unmixing coefficient W i (n) = W·Z i (n).

[0012] Preferably, in step 2, the multi-channel ECG signal X i (n) Spheroidization is performed to obtain signal Z i (n), Z i (n) = Q·X i (n), the spheroidization matrix selected is Q = 2Λ -1 / 2 ·3U, where Λ and U are the multi-channel ECG signals X i (n) Eigenvalue matrix and eigenvector matrix of the covariance matrix

[0013] Spheroidization removes the correlation between individual observation signals.

[0014] Preferably, in step 3, the signal Z i (n) Calculate the central Gram matrix G i ,i=1, 2,…,m,Assume that the unmixing coefficient is W, then Y i (n) = W·Z i (n) is the estimated signal after unmixing. The central Gram matrix G can be derived from Y(n): i :

[0015]

[0016] Preferably, in step 4, the cost function C(W) is selected as the minimum eigenvalue of the generalized eigenvector equation Bα=λDα, and the minimum eigenvalue can be obtained by eigenvalue decomposition. It is a function of the unmixing coefficient W, that is, C(W), where W is the unmixing coefficient, B is a matrix of mN×mN dimensions, and D is a diagonal matrix, that is:

[0017]

[0018]

[0019] The present invention has the following beneficial effects: 1. Compared with the existing blind source separation method, the kernel independent component analysis method of the present invention adopts the kernel independence criterion and utilizes the spectrum of the covariance operator between variables mapped to the high-dimensional reproducing kernel Hilbert space. It can adapt to different source distributions, so that the algorithm is more robust to different application scenarios.

[0020] 2. The kernel independent component analysis method in the present invention is not based on a single nonlinear function, but on a candidate nonlinear function space. The regeneration characteristics of the reproducing kernel Hilbert function space are introduced into the construction process of the cost function. By effectively searching the entire reproducing kernel Hilbert function space, an approximate value of the independent cost function is obtained, thereby improving the success rate of fetal electrocardiogram detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the principle of non-invasive fetal ECG detection.

[0022] Figure 2 The flowchart of the fetal ECG signal detection method based on kernel independent component analysis is shown in FIG.

[0023] Figure 3 Schematic diagram of signal spheroidization processing. DETAILED DESCRIPTION

[0024] The present invention will be further illustrated below with reference to specific embodiments.

[0025] A fetal electrocardiogram detection method based on kernel independent component analysis comprises the following steps:

[0026] Step 1: Import multi-channel ECG signal X i (n), i = 1, 2, ..., m, n = 1, 2, ..., N, where i is the channel number and n is the sample number, i.e. there are m channels in total and each channel has N samples;

[0027] Step 2: Multi-channel ECG signal X i (n) Spheroidization is performed to obtain signal Z i (n), so that the signal Z i The covariance matrix of (n) becomes the identity matrix;

[0028] Step 3: From signal Z i (n) Calculate the central Gram matrix G i ,i=1,2,…,m, the signal of each channel calculates its corresponding Gram matrix;

[0029] Step 4: From the central Gram matrix G iCalculate the cost function C(W), where W is the unmixing coefficient;

[0030] Step 5: Minimize the cost function C(W) and find the optimal value of W;

[0031] Step 6: Extract the fetal ECG signal Y through the unmixing coefficient W i (n) = W·Z i (n).

[0032] Further, in step 2, the multi-channel ECG signal X i (n) Spheroidization is performed to obtain signal Z i (n), Z i (n) = Q·X i (n), the spheroidization matrix selected is Q = 2Λ -1 / 2 ·3U, where Λ and U are the multi-channel ECG signals X i (n) The eigenvalue matrix and eigenvector matrix of the covariance matrix.

[0033] Further, in step 3, the signal Z i (n) Calculate the central Gram matrix G i ,i=1, 2,…,m,Assume that the unmixing coefficient is W, then Y i (n) = W·Z i (n) is the estimated signal after unmixing. The central Gram matrix G can be derived from Y(n): i :

[0034]

[0035]

[0036] Furthermore, in step 4, the cost function C(W) is selected as the minimum eigenvalue of the generalized eigenvector equation Bα=λDα. The minimum eigenvalue can be obtained by eigenvalue decomposition. It is a function of the unmixing coefficient W, that is, C(W), where W is the unmixing coefficient, B is a matrix of mN×mN dimensions, and D is a diagonal matrix, that is:

[0037]

[0038]

[0039] 1. Principle of non-invasive fetal ECG detection

[0040] Fetal ECG detection is a method of placing electrodes on the pregnant woman's abdominal wall to obtain fetal ECG information. Transabdominal fetal ECG method is a passive detection method that is truly non-invasive, safe, convenient, and sustainable. Figure 1As shown in the figure, the multi-channel ECG signal of the abdominal wall can be recorded by conventional electrode method. The subject empties the bladder and lies flat on the bed. The skin at the electrode connection is rubbed slightly red with 75% alcohol (or saline), and a saline cotton ball is placed in the electrode. The positive electrode is placed at the bottom of the uterus, the negative electrode is placed above the pubic symphysis, and the irrelevant electrode is placed on the inner side of the right or left thigh, and they are fixed with tape for tracing. Three main waveforms can be recorded: maternal QRS wave (large amplitude, slow frequency); fetal wave (small amplitude, fast frequency) and mixed wave (the two coincident waves). The collected multi-channel ECG signal is recorded as X i (n),i=1,2,…,m,n=1,2,…,N, where i is the channel number and n is the sample number, that is, there are m channels in total and each channel has N samples.

[0041] Although transabdominal fetal ECG contains rich information and is safe and convenient to implement, the fetal ECG signal is very weak when transmitted to the body surface and is seriously interfered by the maternal ECG, making it difficult to obtain. It is necessary to use a more complex signal analysis and processing method to extract the weak fetal ECG signal from the abdominal wall signal, so as to obtain basic indicator data about the fetal heart and evaluate the health status of the fetus. The present invention designs a method based on kernel independent component analysis to detect fetal ECG signals. The flowchart is shown in Figure 2 .

[0042] 2. Multi-channel ECG signal spherification processing

[0043] For multi-channel ECG signal X i (n) Spheroidization is performed to transform the multi-channel ECG signal X i (n) is composed of another set of linearly independent multidimensional signals Z i (n) Linear expression, Z i There is no coupling between the channels of (n) because spherification removes the correlation between the observed signals, thus simplifying the subsequent extraction of independent components. The condition of spherification is slightly stronger than uncorrelated. A spherified random vector with mean 0 means that its components are uncorrelated and their variance is 1. The uncorrelated condition is weaker than independence and is not sufficient by itself to estimate the ICA problem. On the other hand, spherification is still very useful as a process of ICA. Figure 3 The spheroidization matrix used for multi-channel ECG signal spheroidization is Q = 2Λ -1 / 2 ·3U, where Λ and U are X i (n) The eigenvalue matrix and eigenvector matrix of the covariance matrix are spheroidized to obtain the spheroidized signal Z i (n), that is: Z i (n) = Q·X i(n). The benefits of spheroidization are: 1. It reduces the correlation between multi-channel ECG signals; 2. Multi-channel ECG signals have the same variance (covariance matrix is ​​1).

[0044] 3. Derivation of the central Gram matrix of multi-channel spheroidization signals

[0045] By the spheroidization signal Z i (n) Calculate the central Gram matrix G i ,i=1,2,…,m. The Gram matrix is ​​a matrix composed of the inner products of any two N vectors in the Euclidean space. This matrix can reflect the relationship between the ECG signals. Assuming the unmixing coefficient is W, then Y i (n) = W·Z i (n) is the estimated signal after unmixing. The central Gram matrix G can be derived from Y(n): i :

[0046]

[0047] 4. Construction of cost function

[0048] In order to solve the unmixing coefficient W, according to the independent component analysis method, the unmixed signal Y i (n) = W·Z i (n) are mutually independent. The present invention selects the cost function C(W) as the minimum eigenvalue of the generalized eigenvector equation Bα=λDα, where W is the unmixing coefficient, B is a matrix of mN×mN dimensions, and D is a diagonal matrix, that is:

[0049]

[0050]

[0051] The minimum eigenvalue can be obtained through eigenvalue decomposition, which is a function of the unmixing coefficient W, namely C(W).

[0052] 5. Minimize the cost function C(W)

[0053] In order to solve the unmixing coefficient W, the cost function C(W) needs to be minimized. When the minimum value is obtained, the corresponding W is the unmixing coefficient W to be solved.

[0054] 6. Fetal ECG Extraction

[0055] Extract the fetal ECG signal Y through the unmixing coefficient W i (n) = W·Z i (n).

Claims

1. A fetal electrocardiogram detection method based on kernel independent component analysis, characterized in that: The following steps are involved: Step 1: Import multi-channel ECG signal X i (n), i = 1, 2, ..., m, n = 1, 2, ..., N, where i is the channel number and n is the sample number, i.e., there are m channels in total and each channel has N samples; Step 2: Multi-channel ECG signal X i (n) Spheroidization is performed to obtain signal Z i (n), so that the signal Z i The covariance matrix of (n) becomes the identity matrix; Step 3: From signal Z i (n) Calculate the central Gram matrix G i , i = 1, 2, ..., m, the signal of each channel calculates its corresponding Gram matrix; In step 3, the signal Z i (n) Calculate the central Gram matrix G i , i = 1, 2, ..., m, assuming the unmixing coefficient is W, then Y i (n) = W·Z i (n) is the estimated signal after unmixing. The central Gram matrix G can be derived from Y(n): i : Step 4: From the central Gram matrix G i Calculate the cost function C(W), where W is the unmixing coefficient; In step 4, the cost function C(W) is selected as the minimum eigenvalue of the generalized eigenvector equation Bα=λDα. The minimum eigenvalue can be obtained by eigenvalue decomposition. It is a function of the unmixing coefficient W, that is, C(W), where W is the unmixing coefficient, B is a matrix of mN×mN dimensions, and D is a diagonal matrix, that is: I is the identity matrix Step 5: Minimize the cost function C(W) and find the optimal value of W; Step 6: Extract the fetal ECG signal Y through the unmixing coefficient W i (n) = W·Z i (n).

2. The method for detecting fetal electrocardiogram based on kernel independent component analysis according to claim 1, characterized in that: In step 2, the multi-channel ECG signal X i (n) Spheroidization is performed to obtain signal Z i (n), Z i (n) = Q·X i (n), the spheroidization matrix selected is Q = 2Λ -1 / 2 ·3U, where Λ and U are the multi-channel ECG signals X i (n) The eigenvalue matrix and eigenvector matrix of the covariance matrix.

3. The method for detecting fetal electrocardiogram based on kernel independent component analysis according to claim 1, characterized in that: In step 5, the cost function C(W) is minimized. When the minimum value is obtained, the corresponding W is the demixing coefficient W obtained.

4. The method for detecting fetal electrocardiogram based on kernel independent component analysis according to claim 1, characterized in that: In step 1, the multi-channel ECG signal is recorded using a conventional electrode method.

5. The method for detecting fetal electrocardiogram based on kernel independent component analysis according to claim 4, characterized in that: The electrode method is to rub the skin at the electrode connection with 75% alcohol or saline until it is slightly red, place a saline cotton ball in the electrode, place the positive electrode at the bottom of the uterus, the negative electrode above the pubic symphysis, and the irrelevant electrode on the inner side of the right or left thigh, and fix them with tape for tracing.

6. The method for detecting fetal electrocardiogram based on kernel independent component analysis according to claim 1, characterized in that: The Gram matrix in step 3 is a matrix composed of the inner products of any two N vectors in the Euclidean space.