Method and system for locating based on cardiac fibrillation electrophysiological simulation

Abstract

A positioning method and system based on cardiac atrial fibrillation electrophysiology simulation, the system of the present application is through: constructing a cardiac electrophysiology model, wherein the atrial fibrillation trigger focus is set to obtain the typical atrial fibrillation electrocardiogram signal; P wave extraction is performed on the electrocardiogram signal, the P wave time domain feature is obtained, and the dimension is reduced through principal component analysis method to construct a data set; a machine learning model is trained based on the data set, and the atrial region corresponding to the classification result is divided; the multi-lead signal classification result is summarized, the grade rating is carried out, and the region where the atrial fibrillation trigger focus is located is positioned. The present application can obtain various simulation atrial fibrillation electrocardiogram signals, efficiently and accurately identify the P wave band, save a large amount of calculation time and calculation cost under the premise of retaining sample information, and effectively improve the classification effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical information processing, in particular to a positioning method and system based on cardiac atrial fibrillation electrophysiology simulation. BACKGROUND

[0002] Atrial fibrillation (AF) is the most common supraventricular tachyarrhythmia, accompanied by uncoordinated atrial electrical activation and ineffective atrial contraction, which is prone to form thrombus and cause serious complications such as pulmonary embolism and limb embolism. Accurate positioning of AF ectopic trigger focus is crucial. At present, there is a lack of algorithms in the field of medical information processing that can intelligently locate AF ectopic trigger focus. Accurate simulation of human heart activity has important practical significance for pathological research of the heart, which can achieve the effect of prediction and intervention, and provide training samples that can replace clinical data for related processing algorithms. Based on the simulation data set, the time domain features of the AF trigger focus located in different regions of the atrium corresponding to the electrocardiogram waveform are extracted through machine learning method, which can realize intelligent positioning of AF ectopic trigger focus.

[0003] The above information disclosed in the background section is only for the purpose of enhancing the understanding of the background of the present application and therefore can include information that does not constitute prior art that is already known to those of ordinary skill in the art. SUMMARY

[0004] In view of the problems in the prior art, the present application proposes a positioning method based on cardiac atrial fibrillation electrophysiology simulation, which can obtain various simulation atrial fibrillation electrocardiogram signals, efficiently and accurately identify the P wave band, save a large amount of calculation time and cost under the premise of retaining sample information, and effectively improve the classification effect.

[0005] The present application is realized by the following technical solutions:

[0006] A positioning system based on cardiac atrial fibrillation electrophysiology simulation, comprising:

[0007] A first unit for constructing a cardiac electrophysiology model, wherein an AF trigger focus is set to obtain an electrocardiogram signal of a typical AF;

[0008] A second unit for P wave extraction of the electrocardiogram signal, obtaining P wave time domain features and dimension reduction through principal component analysis method, and constructing a data set;

[0009] A third unit for training a machine learning model based on the data set and dividing the atrial region corresponding to the classification result;

[0010] A fourth unit for summarizing the multi-lead signal classification results, performing grade rating, and positioning the region where the AF trigger focus is located.

[0011] Preferably,

[0012] The cardiac electrophysiological model comprises a cardiac conduction system and preset atrial fibrillation trigger foci in each region of the atrium.

[0013] Preferably,

[0014] The action potential equation of the cardiac conduction system is:

[0015]

[0016]

[0017]

[0018] Wherein, V e is an extracellular potential, V i is an intracellular potential, u is a recovery variable for controlling the refractory period of the cell; σ e and σ i are extracellular and intracellular conductivities, respectively, V m is a transmembrane potential, V m = V i -V e ; a, b, c1, c2, d, e, k, A and B are two-domain model parameters, i ion，l is a cell membrane ion flow; l takes the value of 1 or 2; is a Hamiltonian operator.

[0019] Preferably,

[0020] When l takes the value of 1, the cell membrane ion flow i ion，1 in the atrial fibrillation trigger focus is defined as:

[0021]

[0022] When l takes the value of 2, the cell membrane ion flow i ion，2 in the cardiac conduction system is defined as:

[0023]

[0024] Preferably,

[0025] In the second unit, the P wave of the electrocardio signal is extracted based on wavelet packet analysis. The wavelet packet is a function set associated with the scale function and the wavelet mother function. For any positive integer n, it is expressed in binary form as:

[0026]

[0027] Wherein ε i represents the value of each bit in binary, taking 0 or 1; 2 i-1 represents the base of each bit,

[0028] Wavelet packet W n is expressed as:

[0029]

[0030]

[0031] wherein h0 and h1 are both orthogonal scale function weighting coefficients, h1(k) = (-1) k h0(1-k); ω is a frequency variable of the signal,

[0032] The wavelet packet coefficient d at the node (j+1, p) is:

[0033]

[0034]

[0035] m is an integer,

[0036] The P-wave corresponding frequency band is extracted by reconstructing the wavelet packet decomposition coefficients, and then the P-wave corresponding waveband is identified.

[0037] Preferably,

[0038] The P-wave time domain features include 13-dimensional features such as mean value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude factor, waveform factor, impact factor, margin factor, and energy,

[0039] The second unit reduces the dimension by principal component analysis, sorts according to the contribution rate of the principal component, selects the first five principal components and adds the lead number as the sixth-dimensional feature to construct a data set.

[0040] Preferably,

[0041] The third unit obtains the penalty factor C and the Gaussian kernel function parameter y of the support vector machine through the particle swarm optimization algorithm, constructs a PSO-SVM model to enhance the classification effect of the atrial fibrillation trigger focus, and the velocity iteration formula of the particle swarm optimization algorithm contains three parameters of inertia weight w and learning factors c1 and c2, and the expression is:

[0042]

[0043] c1 = (e1-e2)g / g max +e2

[0044] c2 = (f1-f2)g / g max +f2

[0045] wherein g is the current iteration number; a, b, and d are parameters, a is the maximum value of the inertia weight, b is the minimum value of the inertia weight, d is a parameter, and the value is 2; gmax The maximum value represents the number of iterations; e1, e2, f1, and f2 are parameters, where e1 takes the value 1, e2 takes the value 2, f1 takes the value 1, and f2 takes the value 2.

[0046] Preferred,

[0047] The fourth unit determines the location of atrial fibrillation trigger foci by summarizing the classification results of all leads in the case and setting a confidence threshold.

[0048] Compared with the prior art, the present invention has the following advantages:

[0049] This invention can acquire various simulated atrial fibrillation ECG signals, efficiently and accurately identify the P-wave segment, and save significant computation time and cost while preserving sample information, effectively improving classification results. Specifically, this invention: by establishing an electrophysiological simulation model, using a dual-domain model to set the action potential, and adjusting the location of the atrial fibrillation trigger foci and the activation potential frequency, can acquire various simulated atrial fibrillation ECG signals; through a P-wave extraction method based on wavelet packet analysis, the P-wave segment can be identified efficiently and accurately; through PCA dimensionality reduction, significant computation time and cost can be saved while preserving sample information; and based on a PSO-SVM data classification model, the optimal particle points can be intelligently found as model parameters, effectively improving classification results. Attached Figure Description

[0050] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0051] In the attached diagram:

[0052] Figure 1 This is a flowchart illustrating a localization method based on electrophysiological simulation of atrial fibrillation provided by the present invention.

[0053] Figure 2 This is a schematic diagram of the electrophysiological simulation model of an embodiment of the present invention;

[0054] Figure 3 This is an embodiment of the atrial fibrillation electrocardiogram waveform of the present invention;

[0055] Figure 4 This is a diagram illustrating the P-wave characteristic peak recognition effect achieved by this invention.

[0056] Figure 5 is a single cycle complete P wave waveform extraction effect drawing of implementation;

[0057] Figure 6 is a P wave time domain feature PCA scatter plot matrix of the embodiment of the application;

[0058] Figure 7 is a PSO-SVM training set result schematic diagram of the embodiment of the application;

[0059] Figure 8 is a PSO-SVM verification set result schematic diagram in the atrial fibrillation trigger focus positioning method based on cardiac electrophysiological simulation of the embodiment of the application.

[0060] The application will be further explained in conjunction with the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0061] The specific embodiments of the application will be described below in conjunction with the accompanying drawings. Although specific embodiments of the application are shown in the drawings, it should be understood that the application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided so that the application can be more thoroughly understood and the scope of the application can be fully conveyed to those skilled in the art.

[0062] It should be noted that certain terms are used in the specification and claims to refer to certain components. Those skilled in the art will understand that the same component can be referred to by different terms. The specification and claims do not distinguish components based on the difference in terminology, but rather on the difference in function. As used throughout the specification and claims, "comprise" or "include" is an open term, which should be interpreted as "including but not limited to". The subsequent description describes preferred embodiments of implementing the application, but the description is for the purpose of illustrating the general principles of the specification, and is not intended to limit the scope of the application. The scope of protection of the application is defined by the appended claims.

[0063] To facilitate understanding of the embodiments of the application, the following will be further explained and described with specific embodiments as examples in conjunction with the accompanying drawings, and each drawing does not constitute a limitation on the embodiments of the application.

[0064] As shown in Figures 1 to 8 , the application discloses a positioning method and system based on cardiac atrial fibrillation electrophysiological simulation, wherein,

[0065] The system comprises:

[0066] A first unit for constructing a cardiac electrophysiological model, wherein a atrial fibrillation trigger focus is set to obtain an electrocardiogram signal of a typical atrial fibrillation;

[0067] a second unit for P-wave extraction of the electrocardio signal, obtaining P-wave time domain features and dimension reduction by principal component analysis, and constructing a data set;

[0068] a third unit for training a machine learning model based on the data set, and dividing the atrial region corresponding to the classification result;

[0069] a fourth unit for summarizing the multi-lead signal classification results, performing grade rating, and positioning the region where the atrial fibrillation trigger site is located.

[0070] The method comprises the following steps:

[0071] Step S1, constructing a cardiac electrophysiological model, wherein a typical atrial fibrillation electrocardio signal is obtained by setting an atrial fibrillation trigger site;

[0072] Step s2, P-wave extraction of the electrocardio signal, obtaining P-wave time domain features and dimension reduction by principal component analysis, and constructing a data set;

[0073] Step S3, training a machine learning model based on the data set, and dividing the atrial region corresponding to the classification result;

[0074] Step S4, summarizing the multi-lead signal classification results, performing grade rating, and positioning the region where the atrial fibrillation trigger site is located.

[0075] In another embodiment,

[0076] The cardiac electrophysiological model comprises a cardiac conduction system and a preset atrial fibrillation trigger site in each region of the atrium.

[0077] In another embodiment,

[0078] The action potential equation of the cardiac conduction system is:

[0079]

[0080]

[0081]

[0082] wherein V e is the extracellular potential, V i is the intracellular potential, u is a recovery variable for controlling the refractory period of the cell; σ e and σ i are the extracellular and intracellular conductivities, respectively, V m is the transmembrane potential, V m = V i -V e ; a, b, c1, c2, d, e, k, A and B are two-domain model parameters, iion，l is the ion current across the cell membrane; l takes the value 1 or 2; is a Hamiltonian operator.

[0083] In another embodiment,

[0084] is the ion current across the cell membrane in the atrial fibrillation trigger site when l takes the value 1. ion，1 is defined as:

[0085]

[0086] is the ion current across the cell membrane in the cardiac conduction system when l takes the value 2. ion，2 is defined as:

[0087]

[0088] In another embodiment,

[0089] The P wave of the electrocardio signal is extracted based on wavelet packet analysis. The wavelet packet is a function set associated with the scale function and the wavelet mother function. For any positive integer n, it can be expressed in binary form:

[0090]

[0091] wherein ε i represents the value of each bit in binary, taking 0 or 1; 2 i-1 represents the base of each bit.

[0092] The wavelet packet W n is expressed as:

[0093]

[0094]

[0095] wherein h0 and h1 are both orthogonal scale function weighting coefficients, h1(k) = (-1) k h0(1-k); ω is the frequency variable of the signal.

[0096] The wavelet packet coefficient d at the node (j+1, p) is:

[0097]

[0098]

[0099] m is an integer.

[0100] The P wave corresponding frequency band is extracted by reconstructing the wavelet packet decomposition coefficients, and then the P wave corresponding wave band is identified.

[0101] In another embodiment,

[0102] The P-wave time domain features include 13-dimensional features such as mean value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude factor, waveform factor, impact factor, margin factor, and energy,

[0103] The system and the method thereof are characterized in that the dimensionality reduction is performed through principal component analysis, the principal components are sorted according to the contribution rates of the principal components, the first five principal components are selected, and the lead number is added as the sixth-dimensional feature to construct a data set.

[0104] In another embodiment,

[0105] For training the machine learning model based on the data set, the atrial region corresponding to the classification result is divided:

[0106] The application obtains the penalty factor C and the Gaussian kernel function parameter y of the support vector machine through the particle swarm optimization algorithm, constructs a PSO-SVM model to enhance the classification effect of the atrial fibrillation trigger, and the velocity iteration formula of the particle swarm algorithm includes three important parameters of inertia weight w and learning factors c1 and c2, and the expression is:

[0107]

[0108] c1=(e1-e2)g / g max +e2

[0109] c2=(f1-f2)g / g max +f2

[0110] wherein g is the current iteration number; a, b and d are parameters, a is the maximum value of the inertia weight, b is the minimum value of the inertia weight, d is a parameter and takes the value of 2; g max represents the maximum value of the iteration number; e1, e2, f1 and f2 are parameters, e1 takes the value of 1, e2 takes the value of 2, f1 takes the value of 1, and f2 takes the value of 2.

[0111] Specifically, the PSO-SVM training process is as follows:

[0112] (1) initialize the particle swarm position and velocity. The position of the particle represents the value of the kernel function parameter and the penalty parameter C of the SVM; (2) for each particle, use its position to train the SVM model, and calculate the accuracy on the training set as the fitness value of the particle; (3) find the global optimal particle (global optimal particle); (4) update the velocity and position of each particle according to the current optimal position and the individual historical optimal position; (5) use the updated new position to train the SVM again, and calculate the new fitness; (6) judge whether the stopping condition is met (reach the maximum iteration number or meet the accuracy requirement). If not, return to step 2 for iteration; (7) output the global optimal position, and the corresponding SVM model parameters.

[0113] To deal with the possible non-convergence of training, the following measures are taken:

[0114] (1) Adjust the learning factors c1 (set to 1.0) and c2 (set to 1.3) of PSO to control the influence of individual optimum and population optimum; (2) Increase the particle swarm size (set to 5) to avoid local optimum; (3) Appropriately increase the maximum number of iterations (set the maximum number of evolution to 120); (4) Normalize the training data; (5) Appropriately reduce the C value range of SVM (0.1-100); (6) Run multiple times to get the best result.

[0115] In another embodiment,

[0116] All lead classification results of the cases are summarized, a confidence threshold is set, and the area where the atrial fibrillation trigger focus is located is determined.

[0117] In another embodiment,

[0118] By establishing an electrophysiological simulation model, setting the action potential using a dual-domain model, and adjusting the position of the atrial fibrillation trigger focus and the frequency of the activation potential, a variety of simulated atrial fibrillation electrocardiogram signals can be obtained; by using the P-wave extraction method based on wavelet packet analysis, P-wave bands can be efficiently and accurately identified; by using the PCA dimension reduction method, a large amount of calculation time and cost can be saved while preserving sample information; the data classification model based on PSO-SVM can intelligently find the optimal particle point as the model parameter, effectively improving the classification effect.

[0119] In one embodiment, the P-wave of an electrocardiogram is extracted based on wavelet packet analysis. By reconstructing the wavelet packet decomposition coefficients, the frequency band corresponding to the P-wave can be extracted, and the P-wave band can be identified.

[0120] In one embodiment, the P-wave time domain features are extracted, the dimension is reduced by principal component analysis, and the lead number is added to construct the data set.

[0121] In one embodiment, the penalty factor C and the Gaussian kernel function parameter y of the support vector machine are obtained by the particle swarm optimization algorithm, and the PSO-SVM model is constructed to enhance the classification effect of the atrial fibrillation trigger focus.

[0122] In one embodiment, the PSO-SVM model is trained using the constructed data set, and the classification results of each lead signal are output.

[0123] In one embodiment, all lead classification results of the cases are summarized, a confidence threshold (such as 75%) is set, and the area where the atrial fibrillation trigger focus is located is determined.

[0124] The application will be further described in conjunction with the drawings and specific embodiments.

[0125] Figure 1 A flowchart of a method for locating a focal point of atrial fibrillation based on cardiac electrophysiological simulation according to an embodiment of the present application is shown in FIG. 1, which comprises the following steps: Figure 1

[0126] S1, using COMSOL Multiphysics 6.1 version multi-physics simulation software to construct an electrophysiological simulation model of the heart, setting a focal point of atrial fibrillation in different regions of the atrium, and obtaining the electrophysiological simulation model for simulating atrial fibrillation; 6.1 version multi-physics simulation software to construct an electrophysiological simulation model of the heart, setting a focal point of atrial fibrillation in different regions of the atrium, and obtaining the electrophysiological simulation model for simulating atrial fibrillation;

[0127] Specifically, a model structure diagram is shown in FIG. 2, which shows, from left to right, a complete electrophysiological simulation model (including a simplified torso, lungs, and heart), a complete heart model, and a distribution diagram of focal points of atrial fibrillation (elliptical region, i.e., a pre-set focal point of atrial fibrillation) in the atrium. Figure 2

[0128] The complete mesh includes 71085 domain units, 14970 boundary units, 1491 edge units, and 434 vertex units, with a minimum unit quality of 0.005137 and an average unit quality of 0.6135.

[0129] The geometric parameters of the focal point of atrial fibrillation include: geometric shape (ellipsoid with a half-axis a of 3 mm, a half-axis b of 3 mm, and a half-axis c of 1.5 mm) and geometric position (e.g., the center point coordinates of the first focal point of atrial fibrillation in region 1 are (-2.8 cm, -2.6 cm, 3.1 cm)).

[0130] Specifically, the calculation region of the electrophysiological model of the heart includes a complete cardiac conduction system and pre-set focal points of atrial fibrillation in different regions of the atrium.

[0131] Specifically, the action potential equation of the cardiac conduction system is:

[0132]

[0133]

[0134]

[0135] wherein V e is an extracellular potential, V i is an intracellular potential, u is a recovery variable for controlling the refractory period of a cell; σ e and σ i are extracellular and intracellular conductivities, respectively, V m is a transmembrane potential, V m = V i -V e ​​; a, b, c1, c2, d, e, k, A and B are the parameters of the two-domain model, i ion，l is the ion current of the cell membrane; l is 1 or 2; is the Hamiltonian operator;

[0136] is the ion current of the cell membrane in the atrial fibrillation trigger focus when l is 1 ion，1 is defined as:

[0137]

[0138] is the ion current of the cell membrane in the cardiac conduction system when l is 2 ion，2 is defined as:

[0139]

[0140] Specifically, by using the action potential equations (1)-(4) to define the electrical activity of the atrial fibrillation trigger focus, using equations (1)-(3) and equation (5) to define the electrical activity of the cardiac conduction system, the generation and conduction of cardiac electrical activity are simulated by using the built-in partial differential equation module of COMSOL Multiphysics. By defining the surface of the heart as a current source through the built-in current module of COMSOL Multiphysics, the mapping of the cardiac electrical signal to the body surface is realized.

[0141] Figure 3 It is shown that when the atrial fibrillation trigger focus spontaneously generates electrical excitation, the standard twelve-lead electrocardiogram signals of atrial fibrillation are obtained, including VI, VII, VIII, aVR, aVL, aVF, a total of 6 bipolar lead signals and V1, V2, V3, V4, V5, V6, a total of 6 monopolar lead signals. The RR interval of the lead signal is absolutely irregular (such as the first RR interval is 740 ms, and the second RR interval is 600 ms), which is a clear characteristic of atrial fibrillation.

[0142] The electrophysiological simulation model sets up 64 lead sites in addition to the standard 12 leads, which are used to obtain more rich multi-lead signal waveforms to construct the data set.

[0143] S2, P-wave extraction is performed on the original electrocardiogram signal, typical time domain feature vectors are obtained, and dimensionality reduction is performed through principal component analysis to construct the data set;

[0144] Specifically, the P-wave of the electrocardiogram is extracted based on wavelet packet analysis. The wavelet packet is a function set of scale functions and wavelet mother functions. For any positive integer n, it can be expressed in binary form:

[0145]

[0146] where ε irepresents the value of each bit of binary, taking 0 or 1; 2 i-1 represents the base of each bit.

[0147] Wavelet packet W n Can be expressed as:

[0148]

[0149]

[0150] Wherein h0 and h1 are both orthogonal scale function weighting coefficients, h1(k) = (-1) k h0(1-k)(k is an integer, the value range is negative infinity to positive infinity); ω is the frequency variable of the signal.

[0151] The wavelet packet coefficient d at the node (j+1, p) is:

[0152]

[0153]

[0154] Wherein k and wavelet packet W n The definition of k in the expression is consistent, and m is an integer from negative infinity to positive infinity.

[0155] By reconstructing the wavelet packet decomposition coefficient, the frequency band corresponding to the P wave can be extracted, and then the P wave corresponding wave band is identified.

[0156] Figure 4 The P wave recognition effect of the P wave extraction method based on wavelet packet decomposition on the typical electrocardio signal is shown, and the black dots in the figure are the P wave peaks identified. As can be seen from the figure, the P wave extraction method based on wavelet packet decomposition can accurately identify the P wave characteristic peak of the electrocardio signal. Taking the first trough before the P wave characteristic peak as the starting point and the first trough after the characteristic peak as the ending point, the starting and ending positions of the P wave can be accurately positioned. Figure 5 The extracted single cycle complete P wave waveform is shown.

[0157] After obtaining the complete P wave waveform based on wavelet packet decomposition, the P wave time domain features are extracted, including 13-dimensional features such as mean, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude factor, waveform factor, impact factor, margin factor, and energy. Through principal component analysis, the dimension is reduced, the principal components are sorted according to the contribution rate, the first 5 principal components are selected and the lead number is added as the 6th feature to construct the data set. The goal of PCA is to find a p x m matrix ω (p is the number of principal components, p ≤ m) for an n x m sample matrix X (n samples, m features) to form a new set of variables z1, z2, …, zp. p , the expression is

[0158]

[0159] where ω ij (i = 1, 2,... p, j = 1, 2,... m) are the coefficients of the matrix ω, X j (j = 1, 2,... m) are the column vectors of the sample matrix X.

[0160] such that z i is uncorrelated with z j (i ≠ j; i, j = 1, 2,... m); z1 is the one with the maximum variance among all linear combinations of the columns of X j (j = 1, 2,... m); z2 is the one with the maximum variance among all linear combinations of the columns of X p that are uncorrelated with z1, z2,..., z p-1 (j = 1, 2,... m); and so on. j (j = 1, 2,... m). The new variables z1, z2,..., zp are the first, second,..., p-th principal components of the original sample matrix X. p

[0161] Figure 6 The PCA scatter plot matrix of the 13-dimensional time-domain feature vectors is shown, in which the cumulative contribution rate of the first 5 principal component features is about 96.39%, indicating that selecting the first 5 principal component features can effectively save calculation time and cost on the premise of retaining most of the time-domain information of the samples.

[0162] S3, training a machine learning model based on the data set, and dividing the atrial region corresponding to the classification result;

[0163] Specifically, 2560 samples of 64-lead signals of two heartbeat periods of 7 regions and a total of 20 atrial fibrillation sources (of which 7 regions are 2 atrial fibrillation sources, and the remaining 3 are) are divided into a training set and a validation set, and each sample contains 5 time-domain features after dimension reduction and lead number, a total of 6 features. The training set is all periods of the first atrial fibrillation source of regions 1-6, the first period of the second and third atrial fibrillation sources, and the first period of the second atrial fibrillation source of the first atrial fibrillation source of region 7 and the first period of the second atrial fibrillation source; the rest constitutes the validation set.

[0164] The penalty factor C and the Gaussian kernel function parameter y of the support vector machine are obtained by the particle swarm optimization algorithm, and a PSO-SVM model is constructed to enhance the classification effect of the atrial fibrillation trigger. The velocity iteration formula of the particle swarm algorithm contains three important parameters of inertia weight w and learning factors c1 and c2, and the expression is:

[0165]

[0166] ​wherein g is the current iteration number; a, b, d are parameters, a is the maximum value of the inertia weight, b is the minimum value of the inertia weight, d is a parameter, and the value is 2; gmax represents the maximum value of the iteration number; e1, e2, f1, f2 are parameters, e1 takes the value 1, e2 takes the value 2, f1 takes the value 1, and f2 takes the value 2.

[0167] Specifically, the PSO-SVM training process is as follows:

[0168] (1) Initialize the particle swarm position and velocity. The position of the particle represents the value of the kernel function parameter and the penalty parameter C of the SVM; (2) for each particle, use its position to train the SVM model, and calculate the accuracy on the training set as the fitness value of the particle; (3) find the global optimal position (global optimal particle); (4) update the speed and position of each particle according to the current optimal position and individual historical optimal position; (5) use the updated new position to train the SVM again, and calculate the new fitness; (6) judge whether the stopping condition is met (reach the maximum iteration number or meet the accuracy requirement). If not, return to step 2 and continue iteration; (7) output the global optimal position, which corresponds to the SVM model parameters.

[0169] In view of the possible training non-convergence, the following measures are taken:

[0170] (1) Adjust the learning factors c1 (set to 1.0) and c2 (set to 1.3) of PSO to control the influence of individual optimal and group optimal; (2) increase the particle swarm size (set to 5) to avoid local optimum; (3) appropriately increase the maximum iteration number (set the maximum evolution number to 120); (4) normalize the training data; (5) appropriately reduce the C value range of SVM (0.1-100); (6) run multiple times to get the best result.

[0171] Figure 7 、 Figure 8 The PSO-SVM training set and validation set classification results based on the simulation data set are shown, wherein the overall accuracy of the training set is 97.65%, and the overall accuracy of the validation set is 79.57%. It is shown that the trained model can basically realize the classification of the region where the 7 types of atrial fibrillation trigger sites are located.

[0172] The classification results and the corresponding relationship of the atrial region are as follows:

[0173]

[0174] S4, aggregate the multi-lead signal classification results, perform grade rating, and locate the region where the atrial fibrillation trigger site is located.

[0175] Specifically, all lead classification results of the cases are summarized, a confidence threshold is set to 75%, and the region where the atrial fibrillation trigger focus is located is determined. The classification results of 64 leads of each case (i.e. different atrial fibrillation sources) in the verification set are counted. In 13 cases (2 cases in each of 1-6 regions, and 1 case in region 7), the classification results of most (more than 75%) leads of 11 cases are consistent with the true values, indicating that by summarizing the classification results of 64 leads, the accuracy of locating the region where the atrial fibrillation trigger focus is located is about 84.6% (11 / 13).

[0176] In summary, the system and method described in the present application: by establishing an electrophysiological simulation model, setting the action potential using a dual-domain model, and adjusting the position of the atrial fibrillation trigger focus and the frequency of the activation potential, a variety of simulated atrial fibrillation electrocardio signals can be obtained; by P-wave extraction based on wavelet packet analysis, P-wave bands can be accurately identified; by PCA dimension reduction, a large amount of calculation time and cost can be saved on the premise of preserving sample information; and based on the PSO-SVM data classification model, the optimal particle point can be intelligently found as the model parameter, effectively improving the classification effect.

[0177] Although the embodiments of the present application are described above in combination with the drawings, the present application is not limited to the above-mentioned specific embodiments and application fields, and the above-mentioned specific embodiments are only illustrative and guiding, but not limiting. Those skilled in the art can make many forms under the guidance of the present specification and without departing from the scope protected by the claims of the present application, which all belong to the protection of the present application.

Claims

1. A positioning system based on electrophysiological simulation of atrial fibrillation, characterized in that, include: The first unit is used to construct a cardiac electrophysiological model, in which an atrial fibrillation triggering focus is set up to obtain typical atrial fibrillation electrocardiogram signals. The second unit is used to extract the P wave from the electrocardiogram signal, obtain the time-domain features of the P wave, and construct a dataset by dimensionality reduction using principal component analysis. The third unit is used to train a machine learning model based on the dataset and divide the atrial region corresponding to the classification results. The fourth unit is used to summarize the classification results of multi-lead signals, perform grade rating, and locate the area where the atrial fibrillation trigger lesion is located.

2. The positioning system based on electrophysiological simulation of atrial fibrillation according to claim 1, characterized in that, The cardiac electrophysiological model includes the cardiac conduction system and pre-defined atrial fibrillation trigger foci located in various regions of the atrium.

3. A positioning system based on electrophysiological simulation of atrial fibrillation according to claim 2, characterized in that, The action potential equation of the cardiac conduction system is: Among them, V e Extracellular potential, V i σ is the intracellular potential, u is the recovery variable controlling the cell's refractory period; e and σ i These are the extracellular and intracellular electrical conductivities, respectively, V. m V is the transmembrane potential. m =V i -V e a, b, c1, c2, d, e, k, A and B are parameters of the two-domain model, i ion，l This represents the cell membrane ion flow; l takes the value of 1 or 2. For Hamiltonian operators.

4. A positioning system based on electrophysiological simulation of atrial fibrillation according to claim 3, characterized in that, When l is 1, the cell membrane ion flow i within the atrial fibrillation triggering focus ion，1 Defined as: When l is 2, the ion flow i in the cell membrane of the cardiac conduction system ion，2 Defined as:

5. A positioning system based on electrophysiological simulation of atrial fibrillation according to claim 4, characterized in that, In the second unit, the P wave of the electrocardiogram signal is extracted based on wavelet packet analysis. The wavelet packet is a set of functions that are related to the scaling function and the wavelet mother function, and is represented in binary form for any positive integer n: Where ε i This represents the value of each bit in binary, which can be either 0 or 1; 2 i-1 This represents the cardinality of each digit. small wavelet W n Represented as: Where h0 and h1 are both weighting coefficients of the orthogonal scaling function, and h1(k) = (-1) k h0(1-k); ω is the frequency variable of the signal. The wavelet packet coefficient d at node (j+1, p) is: m is an integer By reconstructing the wavelet packet decomposition coefficients, the frequency band corresponding to the P-wave is extracted, and then the corresponding band of the P-wave is identified.

6. A positioning system based on electrophysiological simulation of atrial fibrillation according to claim 5, characterized in that, The time-domain characteristics of P-waves include 13 dimensions, including mean, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude factor, waveform factor, impulse factor, margin factor, and energy. The second unit uses principal component analysis to reduce dimensionality, sorts the principal components according to their contribution rate, selects the top 5 principal components and adds lead numbers as the 6th dimension feature to construct the dataset.

7. A positioning system based on electrophysiological simulation of atrial fibrillation according to claim 1, characterized in that, The third unit uses the particle swarm optimization algorithm to obtain the penalty factor C and Gaussian kernel function parameters y of the support vector machine, and constructs a PSO-SVM model to enhance the classification effect on atrial fibrillation trigger foci. The velocity iteration formula of the particle swarm optimization algorithm includes three parameters: inertial weight w and learning factors c1 and c2, and its expression is as follows: c1=(e1-e2)g / g max +e2 c2=(f1-f2)g / g max +f2 Where g is the current iteration number; a, b, and d are parameters, where a is the maximum value of the inertia weight, b is the minimum value of the inertia weight, and d is a parameter with a value of 2; gmax represents the maximum number of iterations; e1, e2, f1, and f2 are parameters, where e1 takes the value 1, e2 takes the value 2, f1 takes the value 1, and f2 takes the value 2.

8. A positioning system based on electrophysiological simulation of atrial fibrillation according to claim 1, characterized in that, The fourth unit determines the location of atrial fibrillation trigger foci by summarizing the classification results of all leads in the case and setting a confidence threshold.

Images