Pulse electric field ablation effect evaluation system and device based on artificial intelligence

Abstract

According to the pulse electric field ablation effect evaluation system based on artificial intelligence, the ablation depth can be accurately predicted through deep analysis of intracardiac electric signals and attachment impedance data and by means of XGBoost and other artificial intelligence algorithms, and real-time feedback and optimization suggestions are provided for clinical treatment. The system is helpful for improving the treatment precision and safety, and has important clinical application value.

Description

Technical Field

[0001] The present invention relates to the field of medical electrophysiology, and particularly to an artificial intelligence-based pulsed electric field ablation effect evaluation system and device. Background Art

[0002] The pulsed electric field ablation technology is widely used in the treatment of heart diseases such as arrhythmia. Traditional ablation effect evaluation methods mostly rely on doctors' experience and real-time monitoring data, but lack standardized and quantitative evaluation methods. In actual treatment, accurate monitoring of ablation depth and area is particularly crucial. However, existing technologies usually have difficulty providing a comprehensive and accurate real-time evaluation means. In order to improve the accuracy and safety of treatment, there is an urgent need for an artificial intelligence-based evaluation system that can comprehensively integrate intracardiac electrical signals and dielectric property data to automatically evaluate the ablation effect, predict the ablation depth, and then optimize the treatment plan.

[0003] The present invention designs an artificial intelligence-based pulsed electric field ablation effect evaluation system. This system can collect intracardiac signals (EGM) and dielectric property data (contact impedance value), combine artificial intelligence algorithms, evaluate the ablation effect in real time, predict the ablation depth, and assist clinicians in optimizing cardiac ablation treatment. Summary of the Invention

[0004] The present invention provides an artificial intelligence-based pulsed electric field ablation effect evaluation system and device, which includes a data acquisition module, an ablation depth prediction training module, and an ablation depth prediction evaluation module. The data acquisition module is connected to a multi-electrode ablation catheter to synchronously collect intracardiac electrical signals (EGM signals) and contact impedance value data at multiple points.

[0005] Further, the ablation depth prediction training module uses the previously collected intracardiac electrical signals (EGM signals) and contact impedance values as historical data for model training, including the following steps:

[0006] Step 1: Data cleaning;

[0007] Step 2: Data annotation;

[0008] Step 3: Data preprocessing;

[0009] Step 4: Configure the XGBoost model;

[0010] Step 5: Model training;

[0011] Step 6: Algorithm evaluation and optimization;

[0012] Step 7: Provide the training result.

[0013] Furthermore, the ablation depth prediction and evaluation module uses the intracardiac electrical signal (EGM signal) and the contact impedance value collected in real time as real-time data for predicting the ablation depth, which includes the following steps:

[0014] Step 1: Real-time data input;

[0015] Step 2: Data preprocessing;

[0016] Step 3: Ablation depth evaluation model;

[0017] Step 4: Ablation depth prediction result.

[0018] Furthermore, the data preprocessing includes data denoising, data alignment, normalization, and feature extraction.

[0019] The data cleaning includes removing outliers and filling in missing values from historical data.

[0020] The data annotation annotates historical data to provide the target for supervised learning of the model.

[0021] Furthermore, the input of the XGBoost model structure in the step 4 is where D i represents the collected data, including the EGM data D EGM,i and the contact impedance data feature D R,i ; y i is the label, including the EGM label y EGM,i and the contact impedance data label y R,i , and after data processing, it is represented as x i is the data feature, including the EGM data feature x EGM,i and the contact impedance data feature x R,i ,

[0022] In the model training stage, the XGBoost model structure is constructed through the following steps:

[0023] (1) Initialize the predicted value: Set it to the mean or median of the target variable;

[0024] (2) Calculate the residual: For each sample, calculate the difference between the current predicted value and the true value y i to obtain the residual;

[0025] (3) Calculate the gradient and second derivative: According to the loss function L, calculate the gradient (first derivative) and second derivative of each sample;

[0026] (4) Select the best split point: On each feature, evaluate different split points and select the split point that can maximize the information gain;

[0027] (5) Update the weights of the leaf nodes: Calculate the weights of each leaf node based on the gradients and second-order derivatives to minimize the loss function.

[0028] Furthermore, the model training in step five optimizes the objective function, constructs multiple weak classifiers, and combines them to improve the prediction accuracy; the objective function consists of two parts: the loss function L and the regularization term Ω; the goal is to minimize the following formula:

[0029]

[0030] Where: represents the true value y i and the predicted value error;

[0031] represents the regularization term, which is used to control the complexity of the model and avoid overfitting; where:

[0032] T: represents the number of trees in the model, that is, how many decision trees are trained in total;

[0033] K: represents the number of leaf nodes in a certain tree, that is, how many leaf nodes are there in a single tree;

[0034] W: represents the weight of the leaf node, that is, the predicted value of this leaf;

[0035] γ: the penalty coefficient that controls the number of leaf nodes and prevents the generation of overly deep trees;

[0036] λ: controls the L2 regularization strength of the leaf node weights and avoids overfitting caused by overly large weight values;

[0037] In the t-th iteration, the new predicted value is updated through the following formula:

[0038]

[0039] Where:

[0040] is the prediction result of the first m - 1 trees;

[0041] f t (x i ): is the new predicted value of the m-th tree.

[0042] Furthermore, during each split, the optimal split point is determined through the gain formula:

[0043]

[0044] Where:

[0045] g i : The first-order derivative of the loss function with respect to the predicted value;

[0046] h i : The second-order derivative of the loss function with respect to the predicted value;

[0047] λ, γ: The parameters in the regularization term respectively.

[0048] Furthermore, the final model combines the predicted values of all trees through weighted summation to obtain the final output:

[0049]

[0050] where f m (x i ) represents the prediction result of the m-th tree, and x i is the input feature value.

[0051] Furthermore, step seven is to provide the training result of the trained model to the ablation depth evaluation model of the ablation depth prediction evaluation module for subsequent prediction of real-time data.

[0052] The present invention also provides an apparatus for evaluating the ablation effect based on artificial intelligence, including:

[0053] EGM data acquisition module, which is used to acquire signal data from the heart through a petal catheter;

[0054] Apposition impedance acquisition module, used in cooperation with the data acquisition module, to acquire the apposition impedance data of the tissue in real time and provide electrical characteristics for evaluating tissue damage and ablation depth;

[0055] Data preprocessing module, which preprocesses the intracardiac signal and dielectric property data, including signal denoising and feature extraction; the signal denoising unit is used to remove power frequency noise and electromyogram interference; the feature extraction unit extracts the amplitude, frequency, and waveform features of the intracardiac signal, as well as the change rate and gradient features of the apposition impedance value;

[0056] Artificial intelligence evaluation module, based on the XGBoost algorithm module, which analyzes the EGM signal data and apposition impedance data, extracts features related to ablation depth from the two data, evaluates the ablation effect in real time and predicts the ablation depth, and predicts the ablation depth according to the input data through the trained regression model to assist the doctor in making real-time treatment adjustments. Brief Description of the Drawings

[0057] Figure 1 Shown is the architecture diagram of the ablation depth prediction system of the present invention;

[0058] Figure 2The figure shows a schematic diagram of an ablation catheter for obtaining EGM data and contact impedance data according to the present invention;

[0059] Figure 3 The figure shows a schematic diagram of the data acquisition module according to the present invention

[0060] Figure 4 The figure shows a schematic diagram of the model structure of XGBoost according to the present invention. Detailed implementation manners

[0061] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0062] Please refer to Figures 1-4 , the present invention provides an artificial intelligence-based pulsed electric field ablation effect evaluation system, which includes a data acquisition module, an ablation depth prediction training module, and an ablation depth prediction evaluation module. The data acquisition module is connected to a multi-electrode ablation catheter to synchronously collect intracardiac electrical signals (EGM signals) and contact impedance value data at multiple points. This data acquisition module is used to collect signal data from the heart through a petal catheter.

[0063] The ablation depth prediction training module uses the previously collected intracardiac electrical signals (EGM signals) and contact impedance values as historical data for model training, including the following steps:

[0064] Step 1: Data cleaning;

[0065] Step 2: Data annotation;

[0066] Step 3: Data preprocessing;

[0067] Step 4: Configure the XGBoost model;

[0068] Step 5: Model training;

[0069] Step 6: Algorithm evaluation and optimization;

[0070] Step 7: Provide training results.

[0071] During the treatment process, the data acquisition module synchronously collects intracardiac electrical signals (EGM) and dielectric property data at multiple points through a multi-electrode ablation catheter. The collected signals are transmitted to the data processing platform through a serial bus. The ablation depth prediction evaluation module uses the real-time collected intracardiac electrical signals (EGM signals) and contact impedance values as real-time data for ablation depth prediction, including the following steps:

[0072] Step 1, Real-time data input;

[0073] Step 2, Data preprocessing;

[0074] Step 3, Ablation depth evaluation model;

[0075] Step 4, Ablation depth prediction result.

[0076] The collected signal data is first passed through a signal denoising unit to remove power frequency noise and electromyogram interference. Subsequently, the feature extraction unit extracts time-domain, frequency-domain, and time-frequency domain features (such as amplitude, frequency, waveform morphology, etc.) from the intracardiac electrical signals, and extracts the change rate and gradient features of the contact impedance data. Finally, the time alignment unit aligns the data of each electrode in time to ensure data synchronization.

[0077] Using the preprocessed signal feature data, combined with the labeled ablation depth data, a regression model is trained using the XGBoost algorithm. XGBoost optimizes the parameters of the decision tree through multiple iterations to model the complex non-linear relationship between the input features and the ablation depth. The input of the XGBoost model structure is where D i represents the collected data, including EGM data D EGM,i and the contact impedance data feature D R,i ; y i is the label, including the EGM label y EGM,i and the contact impedance data label y R,i , after data processing, it is represented as x i is the data feature, including the EGM data feature x EGM,i and the contact impedance data feature x R,i ,

[0078] In the model training stage, the XGBoost model structure is constructed through the following steps:

[0079] (1) Initialize the predicted value: Set it to the mean or median of the target variable;

[0080] (2) Calculate the residual: For each sample, calculate the difference between the current predicted value and the true value y i to obtain the residual;

[0081] (3) Calculate the gradient and second derivative: According to the loss function L, calculate the gradient (first derivative) and second derivative of each sample;

[0082] (4) Select the best splitting point: On each feature, evaluate different splitting points and select the splitting point that can maximize the information gain;

[0083] (5) Update the weights of leaf nodes: Calculate the weights of each leaf node based on the gradient and second derivative to minimize the loss function.

[0084] Furthermore, the model training in step (5) optimizes the objective function, constructs multiple weak classifiers, and combines them to improve the prediction accuracy; the objective function consists of two parts: the loss function L and the regularization term Ω; the goal is to minimize the following formula:

[0085]

[0086] Where: represents the true value y i and the predicted value error;

[0087] represents the regularization term, which is used to control the complexity of the model and avoid overfitting; where:

[0088] T: represents the number of trees in the model, that is, how many decision trees are trained in total;

[0089] K: represents the number of leaf nodes in a certain tree, that is, how many leaf nodes are there in a single tree;

[0090] W: represents the weight of the leaf node, that is, the predicted value of this leaf;

[0091] γ: the penalty coefficient for controlling the number of leaf nodes, preventing the generation of overly deep trees;

[0092] λ: controls the L2 regularization strength of the leaf node weights, avoiding overfitting caused by overly large weight values;

[0093] In the t-th iteration, the new predicted value is updated through the following formula:

[0094]

[0095] Where:

[0096] is the prediction result of the first m - 1 trees;

[0097] f t (x i ): is the new predicted value of the m-th tree.

[0098] Furthermore, each time of splitting, the optimal splitting point is judged through the gain formula:

[0099]

[0100] Where:

[0101] gi : The first-order derivative of the loss function with respect to the predicted value;

[0102] h i : The second-order derivative of the loss function with respect to the predicted value;

[0103] λ, γ: are the parameters in the regularization term, respectively.

[0104] Furthermore, the final model combines the predicted values of all trees through weighted summation to obtain the final output:

[0105]

[0106] where f m (x i ) represents the prediction result of the m-th tree, and x i is the input feature value.

[0107] During the actual treatment process, the system receives the intracardiac electrical signals and contact impedance data in real time, and sends them into the trained XGBoost model for ablation depth prediction. The model predicts the depth of the current ablation area based on the real-time data and feeds it back to the doctor.

[0108] According to the predicted ablation depth, the system provides real-time feedback to help the doctor adjust the ablation parameters (such as pulsed electric field strength, frequency, etc.), optimize the treatment plan, and improve the treatment effect.

[0109] The main technical effects of the present invention are as follows:

[0110] (1) Accurate ablation depth prediction: By using intracardiac electrical signals and contact impedance feature data, combined with artificial intelligence algorithms such as XGBoost, it is possible to achieve accurate prediction of the ablation depth, helping the doctor to grasp the treatment progress in real time.

[0111] (2) Automation and real-time performance: The system can automatically process and analyze a large amount of data, provide real-time feedback on the ablation effect, reduce human intervention, and improve the treatment accuracy.

[0112] (3) Multi-channel data synchronous analysis: The system synchronously collects signals at multiple points through a multi-electrode ablation catheter to ensure a comprehensive evaluation of the ablation effect and avoid omission or misjudgment.

[0113] (4) Optimize the treatment plan: The system optimizes the ablation depth according to the real-time feedback results, assists the doctor in adjusting the treatment plan in real time, improves the treatment effect, and reduces the treatment risk.

[0114] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0115] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is limited by the appended claims and their equivalents.

Claims

1. A pulse electric field ablation effect evaluation system based on artificial intelligence, characterized in that: It includes a data acquisition module, an ablation depth prediction training module and an ablation depth prediction evaluation module. The data acquisition module is connected to a multi-electrode ablation catheter to synchronously acquire intracardiac electrical signals (EGM) and close impedance value data at multiple points.

2. The artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 1, characterized in that: The ablation depth prediction training module uses the previously collected intracardiac electrical signal EGM signal and the close impedance value as historical data for model training, and includes the following steps: Step 1: Data cleaning; Step 2: Data labeling; Step 3: Data preprocessing; Step 4: Configure the XGBoost model; Step 5: Model training; Step 6: Algorithm evaluation and optimization; Step 7: Provide training results.

3. The artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 1, characterized in that: The ablation depth prediction and evaluation module uses the real-time collected intracardiac electrical signal EGM signal and the close impedance value as real-time data to predict the ablation depth, including the following steps: Step 1: Real-time data input; Step 2: Data preprocessing; Step 3: Ablation depth evaluation model; Step 4: Ablation depth prediction results.

4. The artificial intelligence-based pulse electric field ablation effect evaluation system according to any one of claims 2 or 3, characterized in that: The data preprocessing includes data denoising, data alignment, normalization and feature extraction. The data cleaning includes removing outliers and filling missing values ​​from historical data. The data annotation is to annotate the historical data so as to provide a supervised learning target for the model.

5. The artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 2, characterized in that: The input of the XGBoost model structure in step 4 is Where D i It indicates the collected data, including EGM data D EGM,i and the close impedance data feature D R,i ;y i is a label, including EGM label y EGM,i and attach impedance data label y R,i , after data processing, it is expressed as x j is the data feature, including EGM data feature x EGM,i and the close impedance data feature x R,i , During the model training phase, the XGBoost model structure is constructed through the following steps: (1) Initialize the predicted value: set it to the mean or median of the target variable; (2) Calculate the residual: For each sample, calculate the current prediction value and the true value y i The difference between them is the residual; (3) Calculate gradient and second-order derivative: According to the loss function L, calculate the gradient (first-order derivative) and second-order derivative of each sample; (4) Select the best split point: On each feature, evaluate different split points and select the split point that maximizes the information gain; (5) Update leaf node weights: Based on the gradient and second-order derivative, calculate the weight of each leaf node to minimize the loss function.

6. The artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 5, characterized in that: The model training in step 5 constructs multiple weak classifiers by optimizing the objective function and combines them to improve the prediction accuracy; the objective function consists of two parts: the loss function L and the regularization term Ω; the goal is to minimize the following formula: in: Represents the true value y i and predicted values The error of represents the regularization term, which is used to control the complexity of the model and avoid overfitting; where: T: represents the number of trees in the model, that is, how many decision trees are trained in total; K: represents the number of leaf nodes in a tree, that is, how many leaf nodes there are in a single tree; W: represents the weight of the leaf node, that is, the predicted value of the leaf; γ: The penalty coefficient for controlling the number of leaf nodes to prevent the generation of too deep a tree; λ: controls the L2 regularization strength of the leaf node weights to avoid overfitting caused by excessive weight values; In the tth iteration, the new predicted value is updated by the following formula: in: is the prediction result of the first m-1 trees; f t (x i ): is the new prediction value of the mth tree.

7. The artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 6, characterized in that: Each time splitting, the optimal splitting point is determined by the gain formula: in: g i : is the first-order derivative of the loss function with respect to the predicted value; h i : is the second-order derivative of the loss function with respect to the predicted value; λ, γ: are the parameters in the regularization term.

8. The artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 7, characterized in that: The final model combines the predictions of all trees by weighted summation to get the final output: where f m (x i ) represents the prediction result of the mth tree, x i is the input feature value.

9. The artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 8, characterized in that: The step seven is to provide the training results of the trained model to the ablation depth evaluation model of the ablation depth prediction and evaluation module to facilitate the prediction of subsequent real-time data.

10. An artificial intelligence-based pulse electric field ablation effect evaluation device, applicable to the artificial intelligence-based pulse electric field ablation effect evaluation system according to claim 9, characterized in that: include: An EGM data acquisition module, which is used to collect signal data from the heart through the petal catheter; The close-fitting impedance acquisition module is used in conjunction with the data acquisition module to obtain the close-fitting impedance data of the tissue in real time and provide electrical characteristics for evaluating tissue damage and ablation depth; The data preprocessing module preprocesses the intracardiac signal and dielectric property data, including signal noise reduction and feature extraction. The signal noise reduction unit is used to remove power frequency noise and myoelectric interference. The feature extraction unit extracts the amplitude, frequency, waveform characteristics of the intracardiac signal, as well as the change rate and gradient characteristics of the close impedance value. The artificial intelligence evaluation module is based on the XGBoost algorithm module. This module analyzes the EGM signal data and the proximity impedance data, extracts the features related to the ablation depth in the two data, evaluates the ablation effect in real time and predicts the ablation depth. Through the trained regression model, the ablation depth is predicted according to the input data, assisting doctors in making real-time treatment adjustments.

Images