Pulse electric field ablation system electrode attaching detection method based on support vector machine

Abstract

The invention discloses a pulsed electric field ablation system electrode attachment detection method based on a support vector machine, and relates to the technical field of cardiac ablation, and the method comprises the following specific steps: 1, data set collection and preprocessing: recording the current frequency, the impedance amplitude difference and the phase difference between every two adjacent electrodes under different power-on frequencies, step 2, support vector machine SVM training and testing: training the support vector machine SVM through the data set to obtain an SVM nonlinear classifier; and step 3, performing electrode attaching detection. According to the method, the support vector machine SVM is trained through the multiple groups of training set sample data and the corresponding labels, the trained SVM nonlinear classifier is obtained, new data are detected by means of the trained SVM nonlinear classifier, and due to the fact that the SVM has high accuracy for the nonlinear classification problem, the accuracy of the SVM is improved. The contact state of the electrode and the tissue can be accurately judged according to a predicted value output by the trained model, and failure or misjudgment of a model detection result caused by nonlinear change of human tissue impedance is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of cardiac ablation, and in particular to a method for detecting electrode contact of a pulse electric field ablation system based on a support vector machine. Background Art

[0002] Many types of arrhythmias, i.e. conditions in which the normal rhythm of the heart is disrupted, are often treated by epicardial or endocardial ablation (e.g., radiofrequency (RF) ablation, cryoablation, ultrasound ablation, laser ablation, microwave ablation, etc.). Among them, ablation of tissue around the ostium of the pulmonary veins can be performed using a device having a distal end that can be expanded into a substantially circular structure, such as a pulmonary vein ablation catheter. However, the success of any cardiac ablation procedure depends largely on the quality of contact between the ablation electrode and the target tissue.

[0003] The most commonly used method for evaluating the contact between electrodes and tissues is impedance measurement. In pulsed field ablation systems, since ablation can be performed through catheters, these catheters can deliver electrical energy to the tissues and also have the inherent ability to transmit electrical information to remote impedance measurement devices. Due to the differences in blood and heart tissue, ablated tissue and non-ablated tissue, and differences in tissue formation at the tip of the pulsed field ablation catheter at different frequencies, there may be very different tissue impedances, so it is convenient to evaluate the contact status between electrodes and tissues based on impedance difference measurements.

[0004] Although the electrode contact detection algorithm disclosed in the publication number CN115530969A uses variable frequency power-on to measure the human body impedance curve, and then removes abnormal values ​​from the curve and takes the average value as the target value, if the target value is greater than a certain set value of the basic impedance, it is considered that the two electrodes are in contact with the tissue, otherwise they are not in contact. However, the algorithm uses an empirical threshold to determine whether the electrode is in contact with the tissue. This empirical threshold is relatively fixed. When the blood and heart tissue of the human body change due to changes in the external environment, the impedance curve of the human body will change, making the target value invalid, thereby causing errors in the judgment of contact.

[0005] There is also a US patent US10182742B2 that discloses a method and system for evaluating electrode-tissue contact before ablation. The system measures the impedance amplitude difference and phase difference of a given electrode under high-frequency and low-frequency power supply, respectively, and then applies a linear model to these two differences for classification. However, when the blood and heart tissue of the human body undergo complex changes due to changes in the external environment, the impedance of the human body will undergo nonlinear changes, and the linear model will fail and make incorrect judgments. Summary of the invention

[0006] The purpose of the present invention is to provide a pulsed electric field ablation system electrode adhesion detection method based on support vector machine, so as to solve the technical problem that the prior art pulsed electric field ablation system electrode and tissue adhesion state detection method relies on fixed empirical values ​​or linear analysis detection, which easily leads to erroneous judgment results.

[0007] The technical problem to be solved by the present invention can be achieved by the following technical solutions:

[0008] The method for detecting electrode contact of a pulse electric field ablation system based on a support vector machine comprises the following steps:

[0009] Step 1: Dataset collection and preprocessing: First, the electrode patch of the pulsed electric field ablation system is energized at three frequencies: high frequency, medium frequency, and low frequency; at each energization frequency, the current frequency, the impedance amplitude difference and the phase difference between two adjacent electrodes are recorded as a set of characteristic values, and then the contact state between the electrode patch and the human tissue is recorded as a label, and a set of characteristic values ​​is recorded as a set of data sets, corresponding to a label of the contact state, and multiple sets of data sets are recorded, and the multiple sets of data sets are randomly divided into training set samples and test set samples;

[0010] Step 2: Support vector machine (SVM) training and testing: Standardize all features in the data set, obtain an SVM nonlinear classifier through training set samples and corresponding labels, and then input all test set samples into the trained SVM nonlinear classifier to obtain three classification results;

[0011] Step 3: Electrode contact detection: Test the impedance amplitude difference and phase difference between two adjacent electrodes at high frequency, medium frequency and low frequency in turn, bring the recorded characteristic values ​​into the SVM model to obtain the label prediction value, and judge the electrode contact state according to the obtained label prediction value.

[0012] As a further solution of the present invention: the frequency adopted by the high-frequency power supply is 100kHZ, the frequency adopted by the medium-frequency power supply is 50kHZ, and the frequency adopted by the low-frequency power supply is 10kHZ.

[0013] As a further solution of the present invention: the contact state between the electrode patch and the human tissue is divided into three states, namely: normal contact, no contact, and excessive contact; normal contact is recorded as label 1, no contact is recorded as label 0, and excessive contact is recorded as label 2.

[0014] As a further solution of the present invention: multiple data sets are divided into three data sets according to the three contact states between the electrode patch and the human tissue, namely, a normal contact data set, a non-contact data set and an excessive contact data set, and the quantity ratio of the three data sets is 2:2:1.

[0015] As a further solution of the present invention: the training set samples divided from the multiple data sets account for 80%; the test set samples account for 20%.

[0016] As a further solution of the present invention: the standardization process in step 2 is specifically as follows: first, the characteristic value of the data set is substituted into the standardization formula as follows:

[0017]

[0018] Among them, x is the original data eigenvalue, μ is the mean of all sample data eigenvalues, σ is the standard deviation of all sample data, and z is the data after standardization;

[0019] Now, the standardized training set sample data characteristic value obtained by inputting the training set sample characteristic value into the standardized formula is set to z1, and the corresponding training set label is set to y; the test set sample data characteristic value obtained by inputting the test set sample characteristic value into the standardized formula is set to z2, and the corresponding test set label is set to y_test.

[0020] As a further solution of the present invention: the specific steps of training the SVM nonlinear classifier in step 2 are: the support vector machine SVM adopts a radial basis kernel function, and sends z1 and y to the SVC function in the Python machine learning library sklearn for training to obtain the SVM nonlinear classifier.

[0021] As a further solution of the present invention: all z2 are input into the trained SVM nonlinear classifier, the output result includes the predicted sticking state degree label y_pred, and then y_pred is compared with the test set label y_test to calculate the accuracy.

[0022] As a further solution of the present invention: the radial basis kernel function is a Gaussian kernel function.

[0023] As a further solution of the present invention: if the label prediction value in step three is 1, it is judged that the two electrodes are in a normal tissue contact state; if the label prediction value is 2, it is judged that the two electrodes are in an excessive contact state, and then the next group of adjacent electrodes are tested; if the label prediction value is 0, the two electrodes are in a non-tissue contact state.

[0024] Beneficial effects of the present invention:

[0025] The present invention sets multiple groups of data sets, and each group of data sets corresponds to a label, which is distinguished according to the contact state between the electrode and the tissue. Each group of data sets corresponds to a label, and then the multiple groups of data sets are divided into training set sample data and test set sample data, and the support vector machine SVM is trained by the multiple groups of training set sample data and the corresponding labels to obtain a trained SVM nonlinear classifier, and the new data is detected by relying on the trained SVM nonlinear classifier. Since SVM has high accuracy for nonlinear classification problems, the contact state between the electrode and the tissue can be accurately judged according to the predicted value output by the trained model, thereby avoiding the failure or misjudgment of the model detection result due to the nonlinear change of human tissue impedance. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention will be further described below in conjunction with the accompanying drawings.

[0027] Figure 1 is a schematic diagram of an existing pulsed electric field ablation system;

[0028] Figure 2 It is a graph of impedance amplitude changes between two adjacent electrodes of a catheter measured under different electrode contact conditions in the present invention;

[0029] Figure 3 It is a graph of phase value variation between two adjacent electrodes of a catheter measured under different electrode contact conditions in the present invention;

[0030] Figure 4 It is a schematic diagram of circuits for realizing different frequencies in the present invention.

[0031] Figure 5 It is a flowchart of the electrode sticking detection method of the pulse electric field ablation system based on the support vector machine of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Figure 1It is a pulsed electric field ablation system in the prior art, including a remote controller 20, a pulsed electric field generator 14, an impedance meter 18, surface electrodes 48, an operating handle 22, a distal catheter 28, an annular contact catheter 32, distributed electrodes 38, and a radial axis 30. The operating handle 22 is used to operate the distal catheter 28 to bring the annular contact catheter 32 to the location of human tissue, and the distributed electrodes 38 distributed on the annular contact catheter 32 contact the tissue, and then the treatment operation is performed.

[0034] Since judging whether the electrode is close to the tissue is a nonlinear binary classification problem, and the support vector machine (SVM) is used for linear and nonlinear classification or regression problems, this scheme is based on the support vector machine (SVM) algorithm for judgment. For nonlinear problems, SVM uses a kernel function to map samples into a high-dimensional space. Specifically, each sample is represented as a point in the high-dimensional space, and an attempt is made to find a hyperplane to separate samples of different categories. The selection of the hyperplane is based on the principle of maximizing the interval between the two categories. The support vector machine achieves this by maximizing the minimum interval between the support vector and the hyperplane, and a Gaussian kernel function is used to generate a nonlinear classifier.

[0035] like Figure 5 As shown, the electrode sticking detection method of the pulsed electric field ablation system based on the support vector machine has the following specific steps:

[0036] Step 1: Dataset collection and preprocessing: First, the electrode patch of the pulsed electric field ablation system is energized at three frequencies: high frequency, medium frequency, and low frequency. The preferred frequency for high frequency is 100kHZ; the preferred frequency for medium frequency is 50kHZ; and the preferred frequency for low frequency is 10kHZ. The circuits for different frequencies are as follows: Figure 4 As shown; under each power-on frequency, the current frequency, the impedance amplitude difference and the phase difference between two adjacent electrodes are recorded. Then, 9 characteristic values ​​can be recorded under the three power-on frequencies, and the electrode contact state is recorded, which is divided into normal contact, no contact and excessive contact; and normal contact is recorded as label 1, no contact is recorded as label 0, and excessive contact is recorded as label 2. Then, each set of data has 9 characteristic values ​​and one label (0, 1 or 2);

[0037] It should be noted that Figure 2 The bipolar (between the two electrodes of the catheter) impedance amplitude values ​​measured under different electrode contact conditions are shown to have significant differences at lower frequencies compared to higher frequencies. Figure 3 As shown, the bipolar phase angles measured at different contact levels have significant differences at higher frequencies compared to lower frequencies.

[0038] Multiple sets of data sets are recorded, and the multiple sets of data sets are divided into three sets of data sets according to the three states of contact between the electrode patch and the human tissue, namely, a normal contact data set, a non-contact data set, and an excessive contact data set, and the number ratio of the three data sets is 2:2:1. For example, if a total of 500 sets of data sets are recorded, 200 sets are normal contact between adjacent electrodes and tissues, with a label of 1; 200 sets are non-contact between adjacent electrodes and tissues, with a label of 0; 100 sets are excessive contact between adjacent electrodes and tissues, with a label of 2; and the data sets are randomly shuffled and divided into training sets and test sets, accounting for 80% and 20% respectively. In the case of recording 500 sets of data sets, the number of training set samples is 400 sets, and the number of test set samples is 100 sets.

[0039] Step 2: Support vector machine (SVM) training and testing:

[0040] SVM training: First, all the feature values ​​of the data set are standardized. The standardization formula used is as follows:

[0041]

[0042] Among them, x is the original data eigenvalue, μ is the mean of all sample data eigenvalues, σ is the standard deviation of all sample data, and z is the standardized data.

[0043] Now, the standardized training set sample data characteristic value obtained by inputting the training set sample characteristic value into the standardized formula is set to z1, and the corresponding training set label is set to y. The test set sample data characteristic value obtained by inputting the test set sample characteristic value into the standardized formula is set to z2, and the corresponding test set label is set to y_test.

[0044] If 500 sets of data sets are recorded, the dimension of the training set samples in the standardized data is [400, 9], where 400 represents the number of samples, 9 represents the number of sample features, and the dimension of the training set labels is

[400] . Similarly, the dimension of the test set samples is [100, 9], and the dimension of the test set labels is

[100] .

[0045] Then the radial basis kernel function rbf is used as the kernel function of the support vector machine SVM; the radial basis kernel function is a Gaussian kernel function, and its form is k(‖x-xc‖)=exp{-‖x-xc‖ 2 / (2σ) 2}, where xc is the center of the kernel function and σ is the width parameter of the function, which controls the radial range of the function.

[0046] Then z1 and y are sent to the SVC function in the Python machine learning library sklearn for training to obtain the SVM nonlinear classifier. Here, the Python machine learning library refers to a collection of Python tools and libraries for machine learning and data analysis; sklearn is one of the Python machine learning libraries, used for data mining and data analysis, and SVM is an algorithm in sklearn; the SVC function is a class in the sklearn learning library for support vector machine SVM classification, which implements the SVM classifier;

[0047] Among them, the trained SVM classifier algorithm is:

[0048] model=SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0,

[0049] decision_function_shape='ovc',degree=3,gamma='auto',kernel='rbf',

[0050] max_iter=-1, probability=False, random_state=None, shrinking=True, tol=0.001, verbose=False);

[0051] model.fit(z,y);

[0052] SVM test: Input all z2 into the trained SVM nonlinear classifier, and the output result includes the predicted sticking state label y_pred;

[0053] Where, y_pred = model.predict(z_test);

[0054] Then compare y_pred with the test set label y_test, that is, y_pred is the label simulated by the SVM nonlinear classifier, and y_test is the label originally corresponding to the characteristic value of the test set sample data. The two are compared to calculate the accuracy.

[0055] Step 3, electrode contact detection: The electrodes are recorded as 1, 2,,,, N. The impedance amplitude difference and phase difference between two adjacent electrodes at high frequency, medium frequency and low frequency are tested in turn, and the 9 recorded eigenvalues ​​are brought into the SVM model to obtain the label prediction value. If the label prediction value is 1, it is determined that the two electrodes are in normal contact with the tissue. If the label prediction value is 2, it is determined that the two electrodes are in excessive contact, and then the next test between adjacent electrodes is performed; if the label prediction value is 0, it is determined that the two electrodes are not in contact with the tissue, and the output catheter is not in contact with the tissue. Only when all adjacent electrodes are in normal contact with the tissue, the output catheter is in normal contact with the tissue.

[0056] The embodiments of the present invention are described in detail above, but the embodiments of the present invention are not limited thereto and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A pulsed electric field ablation system electrode contact detection method based on support vector machine, characterized in that: The following steps are involved: Step 1: Dataset collection and preprocessing: First, the electrode patch of the pulsed electric field ablation system is energized at three frequencies: high frequency, medium frequency, and low frequency; at each energization frequency, the current frequency, the impedance amplitude difference and the phase difference between two adjacent electrodes are recorded as a set of characteristic values, and then the contact state between the electrode patch and the human tissue is recorded as a label, and a set of characteristic values ​​is recorded as a set of data sets, corresponding to a label of the contact state, and multiple sets of data sets are recorded, and the multiple sets of data sets are randomly divided into training set samples and test set samples; Step 2: Support vector machine (SVM) training and testing: Standardize all features in the data set, obtain an SVM nonlinear classifier through training set samples and corresponding labels, and then input all test set samples into the trained SVM nonlinear classifier to obtain three classification results; Step 3: Electrode contact detection: Test the impedance amplitude difference and phase difference between two adjacent electrodes at high frequency, medium frequency and low frequency in turn, bring the recorded characteristic values ​​into the SVM model to obtain the label prediction value, and judge the electrode contact state according to the obtained label prediction value.

2. The method for detecting electrode contact in a pulsed electric field ablation system based on a support vector machine according to claim 1, characterized in that: The frequency used for high-frequency power supply is 100kHZ, the frequency used for medium-frequency power supply is 50kHZ, and the frequency used for low-frequency power supply is 10kHZ.

3. The method for detecting electrode contact in a pulsed electric field ablation system based on a support vector machine according to claim 1, characterized in that: The contact state between the electrode patch and the human tissue is divided into three states: normal contact, no contact, and excessive contact; normal contact is recorded as label 1, no contact is recorded as label 0, and excessive contact is recorded as label 2.

4. The method for detecting electrode contact in a pulsed electric field ablation system based on a support vector machine according to claim 3, characterized in that: The multiple data sets are divided into three data sets according to the three contact states between the electrode patch and the human tissue, namely, a normal contact data set, a non-contact data set and an excessive contact data set, and the number ratio of the three data sets is 2:2:

1.

5. The method for detecting electrode contact in a pulsed electric field ablation system based on support vector machine according to claim 1, characterized in that: The training set samples in the multi-group data set account for 80%; the test set samples account for 20%.

6. The method for detecting electrode contact in a pulsed electric field ablation system based on a support vector machine according to claim 1, characterized in that: The specific standardization process in step 2 is as follows: first substitute the characteristic value of the data set into the standardization formula as follows: Among them, x is the original data eigenvalue, μ is the mean of all sample data eigenvalues, σ is the standard deviation of all sample data, and z is the data after standardization; Now, the standardized training set sample data characteristic value obtained by inputting the training set sample characteristic value into the standardized formula is set to z1, and the corresponding training set label is set to y; the test set sample data characteristic value obtained by inputting the test set sample characteristic value into the standardized formula is set to z2, and the corresponding test set label is set to y_test.

7. The method for detecting electrode contact in a pulsed electric field ablation system based on support vector machine according to claim 6, characterized in that: The specific steps of training the SVM nonlinear classifier in step 2 are as follows: the support vector machine SVM adopts the radial basis kernel function, and sends z1 and y to the SVC function in the Python machine learning library sklearn for training to obtain the SVM nonlinear classifier.

8. The method for detecting electrode contact in a pulsed electric field ablation system based on support vector machine according to claim 6, characterized in that: Input all z2 into the trained SVM nonlinear classifier. The output result includes the predicted adhesion state label y_pred. Then compare y_pred with the test set label y_test to calculate the accuracy.

9. The method for detecting electrode contact in a pulsed electric field ablation system based on support vector machine according to claim 7, characterized in that: The radial basis kernel function is a Gaussian kernel function.

10. The method for detecting electrode contact in a pulsed electric field ablation system based on support vector machine according to claim 3, characterized in that: In step 3, if the label prediction value is 1, it is judged that the two electrodes are in a normal contact tissue state; if the label prediction value is 2, it is judged that the two electrodes are in an excessive contact state, and then the next group of adjacent electrodes are tested; if the label prediction value is 0, the two adjacent electrodes are in a non-contact tissue state.

Images