Pulse field ablation dynamic regulation and control system and device

Abstract

The invention is suitable for the technical field of medical equipment, and provides a pulse field ablation dynamic regulation and control system and device. In the embodiment, the target tissue to be subjected to ablation treatment can be subjected to ablation treatment through the set pulse parameters, then the multi-modal data of the target tissue are collected in the treatment process, and then the real-time target pulse parameters in the treatment process are determined through the multi-modal data; and then corresponding pulse energy is generated according to the target pulse parameter for treatment. The problem of insufficient data utilization during pulse field ablation treatment is solved. And in the treatment process, the acting area of the pulse energy is adjusted in real time according to the field intensity gradient distribution generated by the pulse energy and the three-dimensional model containing the bioelectric characteristics of the target tissue, so that the pulse energy acts on the target tissue according to an expected target. The problem that regulation and control real-time performance is insufficient during pulse field ablation treatment is solved, and accuracy and safety of electric pulse ablation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical equipment, in particular to a pulse field ablation dynamic regulation system and device. BACKGROUND

[0002] Pulse field ablation (PFA) as a new technology for arrhythmia treatment, its core lies in the dynamic regulation of electric field parameters to achieve efficient and safe tissue ablation. The PulseSelect system of Medtronic represents the current international advanced level, which uses adaptive waveform technology to adjust the pulse width (50-100μs) and voltage (1500-2000V / cm) according to the tissue impedance in real time, and has achieved remarkable results in clinical practice, with a transmural lesion rate of up to 99.2%. This achievement has promoted the widespread application of PFA technology in the field of arrhythmia treatment, however, there are still many limitations in the existing technology.

[0003] Data utilization is insufficient: most existing systems only rely on a single parameter of tissue impedance to adjust the electric field parameters, and cannot fully perceive the complex characteristics of the tissue. The heterogeneity of the tissue, such as fibrosis and fat infiltration, will cause significant changes in its electrical, mechanical and thermal properties, and it is difficult to accurately predict the dynamic response of the tissue to the electric field by relying only on impedance feedback. In addition, the existing technology lacks comprehensive analysis of multi-modal data such as tissue elastic modulus and temperature distribution, and cannot construct a complete tissue characteristic model, which limits the accuracy of electric field parameter optimization.

[0004] Regulation real-time is insufficient: the commonly used regulation algorithm, such as PID control algorithm, has a long response time, usually more than 50ms. While the tissue electroporation phenomenon in the PFA process occurs in a very short time (10-100μs level change), this regulation lag makes the electric field parameters unable to adjust in time with the changes in tissue state, affecting the stability and consistency of the ablation effect. During the ablation process, the electrical properties of the tissue will change rapidly with the occurrence of electroporation, if the regulation system cannot respond in time, it may lead to excessive or insufficient local electric field intensity, excessive intensity is easy to cause excessive tissue damage and complications, and insufficient intensity cannot achieve the expected ablation effect.

[0005] Safety risk is high: the existing field intensity distribution model is mostly a static model, which does not fully consider the individual differences of patients and the dynamic changes of the tissue during the ablation process. In clinical practice, sensitive areas such as the esophagus and diaphragmatic nerve have a high probability of thermal injury, and clinical trial data shows that the probability of esophageal thermal injury is more than 15%. For obese patients, the increase in tissue thickness and fat content will change the propagation characteristics of the electric field; while the tissue of thin-walled patients is more sensitive to the electric field, the existing technology is difficult to achieve personalized dose adaptation, and cannot effectively reduce the risk of complications while ensuring the ablation effect. SUMMARY

[0006] Therefore, the application provides a pulse field ablation dynamic regulation system and device to solve the problems of insufficient data utilization and insufficient real-time regulation during pulse field ablation treatment, and to monitor sensitive areas and thermal damage risks during treatment, thereby improving the accuracy and safety of electric pulse ablation.

[0007] The first aspect of the application provides a pulse field ablation dynamic regulation system, which comprises a multi-modal perception module, an AI decision module and a dynamic execution module. The multi-modal perception module is used to collect multi-modal data generated during pulse field ablation, and to extract and fuse features of the multi-modal data to obtain a multi-modal feature vector reflecting the multi-modal feature of the tissue. The AI decision module is used to determine a time series prediction model based on a hybrid architecture of LSTM and Transformer, and to take the multi-modal feature vector as an input parameter of the time series prediction model to obtain a target pulse parameter corresponding to the multi-modal feature vector. The dynamic execution module is used to generate a corresponding pulse energy according to the target pulse parameter, to apply the pulse energy to the target tissue for pulse field ablation, to synchronously monitor the field intensity gradient distribution of the pulse energy, and to make the pulse energy act on the target tissue according to the expected target through the field intensity gradient distribution and a three-dimensional model containing the bioelectric characteristics of the target tissue.

[0008] Optionally, the multi-modal perception module comprises a data acquisition module and a feature vector generation module. The data acquisition module comprises an impedance spectrometer for obtaining the impedance spectrum of the target tissue, an ultrasonic transducer for obtaining the ultrasonic elasticity data of the target tissue, and a contact pressure micro-electro-mechanical system array for obtaining the contact pressure data of the target tissue, and the multi-modal data is composed of the impedance spectrum, the ultrasonic elasticity data and the contact pressure data. The feature vector generation module performs deep fusion on the multi-modal data based on a cross-modal neural network of feature-level fusion to obtain a high-dimensional feature representation capable of comprehensively representing dielectric properties and mechanical properties, and determines the high-dimensional feature representation as the multi-modal feature vector.

[0009] Optionally, the data acquisition module further comprises a temperature sensor for obtaining the temperature of the target tissue, and the multi-modal data further comprises the temperature of the target tissue.

[0010] Optionally, the system further comprises a thermal damage risk detection module configured to predict whether the target tissue is at risk of thermal damage based on the temperature and the pulse energy, and instruct the dynamic execution module to stop generating the pulse energy if so.

[0011] Optionally, the system further comprises a dangerous region processing module configured to monitor the region affected by the pulse energy in real time, and adjust the target pulse parameters if the distance between the affected region and a preset dangerous region is less than a dangerous threshold.

[0012] The second aspect of the present application provides a device for dynamically regulating pulsed field ablation, comprising: a multi-modal sensing unit configured to collect multi-modal data generated during pulsed field ablation, and to extract and fuse features of the multi-modal data to obtain a multi-modal feature vector reflecting tissue characteristics of the multi-modal features; an AI decision unit configured to determine a time series prediction model based on a hybrid architecture of LSTM and Transformer, and to use the multi-modal feature vector as an input parameter of the time series prediction model to obtain target pulse parameters corresponding to the multi-modal feature vector; a dynamic execution unit configured to generate corresponding pulse energy according to the target pulse parameters, and to apply the pulse energy to a target tissue undergoing pulsed field ablation, and to monitor the field intensity gradient distribution of the pulse energy in real time, and to cause the pulse energy to act on the target tissue according to the intended target through the field intensity gradient distribution and a three-dimensional model containing biological electrical characteristics of the target tissue.

[0013] Optionally, the multi-modal sensing unit comprises a data acquisition unit and a feature vector generation unit; The data acquisition unit comprises an impedance spectrometer configured to obtain impedance spectrum of the target tissue, an ultrasonic transducer configured to obtain ultrasonic elasticity data of the target tissue, and a contact pressure micro-electro-mechanical system array configured to obtain contact pressure data of the target tissue, and the multi-modal data is composed of the impedance spectrum, the ultrasonic elasticity data, and the contact pressure data; The feature vector generation unit performs deep fusion on the multi-modal data based on a cross-modal neural network of feature-level fusion to obtain a high-dimensional feature representation capable of comprehensively representing dielectric properties and mechanical properties, and determines the high-dimensional feature representation as the multi-modal feature vector.

[0014] Optionally, the data acquisition unit further comprises a temperature sensor configured to obtain temperature of the target tissue, and the multi-modal data further comprises the temperature of the target tissue.

[0015] Optionally, the device further comprises: a thermal damage risk detection unit configured to estimate whether the target tissue is at risk of thermal damage by the temperature and the pulse energy, and instruct the dynamic execution unit to stop generating the pulse energy if so.

[0016] Optionally, the device further comprises: a dangerous area processing unit configured to monitor the action area of the pulse energy in real time, and adjust the target pulse parameter if the distance between the action area and a preset dangerous area is less than a dangerous threshold.

[0017] In the embodiments provided in the present application, for the target tissue to be ablation treated, the target tissue can be first ablation treated by the set pulse parameter, then the multi-modal data of the target tissue is collected during the treatment process, and the multi-modal feature vector is obtained by feature extraction and fusion of the multi-modal data; then the multi-modal feature vector is input into the pre-trained time series prediction model, and the real-time target pulse parameter during the treatment process is determined, and the corresponding pulse energy is generated by the target pulse parameter for treatment, and the action area of the pulse energy is adjusted in real time according to the field strength gradient distribution generated by the pulse energy and the three-dimensional model containing the biological electric characteristics of the target tissue during the treatment process, so that it acts on the target tissue according to the expected target. This solves the problems of insufficient data utilization and insufficient real-time regulation and control during pulse field ablation treatment, and improves the precision and safety of electric pulse ablation. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The system module diagram provided in the embodiments of the present application is provided; Figure 2 The device structure diagram provided in the embodiments of the present application is provided; Figure 3 The internal structure schematic diagram of the computer equipment provided in the embodiments of the present application is provided. DETAILED DESCRIPTION

[0019] The exemplary embodiments will be described in detail herein below with reference to the drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0020] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0021] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0022] This application provides a dynamic control system for pulse field ablation to improve the accuracy and safety of electrical pulse ablation.

[0023] The technical solutions of this application will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0024] like Figure 1 The diagram shown is a block diagram of a pulse field ablation dynamic control system provided in this application. The implementation process and functional effects of each module are described below.

[0025] Module 1, Multimodal Perception Module. This module is used to collect multimodal data generated during pulsed field ablation, and to extract and fuse features from the multimodal data to obtain a multimodal feature vector that reflects the multimodal tissue characteristics.

[0026] In this embodiment, a sensor array can be constructed to collect various characteristic data. This array may include a high-frequency impedance spectrometer, a miniaturized ultrasonic transducer, and a contact pressure microelectromechanical system (MEMS) array. The high-frequency impedance spectrometer can accurately measure the impedance characteristics of the target tissue at different frequencies, obtaining the corresponding impedance spectrum. The miniaturized ultrasonic transducer can monitor the elastic changes of the target tissue in real time, reflecting its mechanical properties. The contact pressure MEMS array can accurately sense the contact pressure between the electrode and the tissue, ensuring a uniform electric field distribution. Therefore, through the aforementioned sensor array, multimodal data including impedance, ultrasonic elasticity, and contact pressure can be collected.

[0027] In another embodiment, the above-mentioned sensor array can also contain a sensor for collecting temperature, such as a distributed optical fiber temperature sensor. This sensor can achieve high-precision real-time monitoring of the temperature of the ablation area, preventing thermal injury.

[0028] When feature extraction is performed on the above-mentioned multi-modal data, it can be achieved through a cross-modal neural network. The specific process is as follows: The ultrasound elastic modulus is associated with the dielectric spectrum to model a new tissue feature model, and the associated model formula is as follows: , wherein is the elastic modulus, also known as the Young's modulus, representing the elastic mechanical properties of the tissue, which is determined by the above-mentioned elastic change; is the radio frequency conductivity, reflecting the electrical conduction properties of the tissue, which is determined by the above-mentioned impedance spectrum inversion calculation; is the contact pressure density, which reflects the contact state between the electrode and the tissue, which is determined by the above-mentioned MEMS sensor measurement; α∈[7.5,9.0], β∈[0.10,0.15], k∈[0.80,0.85], ρ0=15±2g / mm2; α, β, γ, δ, k, ρ0α, β, γ, δ, k, ρ0 are calibration coefficients, which are determined by experiments on the ex vivo part of the target tissue.

[0029] The reason for introducing the elastic modulus in this embodiment is as follows: ① Defects of single impedance feedback: For low elastic modulus tissues such as fat layer and esophageal wall, it is difficult to distinguish the reasons for impedance rise, such as poor contact, tissue fibrosis, fat infiltration, etc. For example, when the impedance value is 120Ω, it may be due to insufficient electrode contact force (the catheter needs to be adjusted) or the presence of a fat layer (the voltage needs to be increased), which makes it impossible to decide the adjustment direction only by impedance. Therefore, by determining the change of elastic modulus, it can be determined whether the current is in the fat layer.

[0030] For example, in actual application, when the impedance rises, and both the elastic modulus E and the conductivity σ decrease, it indicates that the current is in the fat layer, and the decision to be taken is to increase the voltage. When the impedance rises, the elastic modulus E is normal, but the conductivity σ decreases, it can be determined that there is insufficient electrode contact force, and the catheter needs to be adjusted.

[0031] ② Insufficient transmurality: Since the fat layer acts as an "electric field insulating layer", it will cause the electric field to attenuate, thereby making it impossible to obtain effective field strength in the deep myocardium. For example, when the impedance value is normal, but the electric field is still attenuated by 30%, if the elastic modulus E also decreases, it can be determined that it is in the fat area, and at this time, the voltage needs to be increased to compensate, thereby solving the problem of decreased transmurality.

[0032] In the above correlation model formula, the mechanical contribution term, the electrical contribution term and the contact contribution term are included respectively. The mathematical form corresponding to the mechanical contribution term is , which can solve the problem of insufficient electric field penetration in fibrosis area. The mathematical form corresponding to the electrical contribution term is , which can solve the problem of electric field shielding effect in fat area. The mathematical form corresponding to the contact contribution term is , which can solve the problem of ablation failure caused by poor contact of catheter. Therefore, the multi-modal feature vector extracted from the above multi-modal data by the correlation model formula is also the optimal solution to the above problems.

[0033] Module two, AI decision module. This module is used to determine a time series prediction model based on a hybrid architecture of LSTM and Transformer, and the multi-modal feature vector is used as an input parameter of the time series prediction model to obtain a target pulse parameter corresponding to the multi-modal feature vector.

[0034] In this embodiment, the LSTM layer is responsible for capturing the impedance dynamic changes at the millisecond level and remembering the short-term fluctuations of the electrical properties of the tissue. The time series prediction model is a lightweight Transformer model, which is composed as follows: Input layer: multi-modal feature vector (dimension ≥ 128); Encoder: 4 layers of self-attention mechanism, embedding dimension 64; Output layer: dynamic pulse parameters (voltage V, pulse width PW, pulse interval t); The input layer receives the multi-modal feature vector (dimension ≥ 128), and the 4-layer self-attention mechanism encoder (embedding dimension 64) is carefully designed. The model can effectively capture the long sequence dependence and spatial features in the data, and deeply analyze the complex correlation between tissue characteristics and electric field parameters. The output layer outputs dynamic pulse parameters, including voltage V, pulse width PW and pulse interval Δt.

[0035] The time series prediction model can be deployed on FPGA (field programmable gate array) and quantized by 8-bit fixed-point operation. This hardware acceleration scheme makes the model inference delay less than 1ms, meeting the strict real-time requirements in the PFA process, and can quickly generate optimized pulse parameter instructions according to multi-modal perception data.

[0036] The embodiment designs a time series prediction model with a hybrid architecture of LSTM and Transformer. LSTM can effectively capture the dynamic changes of tissue impedance in the time dimension, remember short-term fluctuations in electrical properties, and respond quickly to real-time changes in tissue during the ablation process. The self-attention mechanism of Transformer focuses on analyzing the heterogeneity of tissue in the spatial dimension, considering the differences in tissue characteristics in different regions. The two complement each other, achieving simultaneous processing of the spatiotemporal evolution of tissue characteristics, greatly improving the accuracy and timeliness of pulse parameter prediction.

[0037] Module three, dynamic execution module, for generating corresponding pulse energy according to the target pulse parameter, acting the pulse energy on the target tissue for pulse field ablation, synchronously monitoring the field strength gradient distribution of the pulse energy, and through the field strength gradient distribution and the three-dimensional model containing the biological electrical characteristics of the target tissue, making the pulse energy act on the target tissue according to the expected target.

[0038] In this embodiment, the patient's heart tissue can be comprehensively evaluated preoperatively through the multi-modal perception module, and multi-modal data such as impedance spectrum, ultrasonic elasticity data, and contact pressure can be collected. The AI decision-making module formulates a personalized initial pulse parameter scheme for each patient based on these data and the spatiotemporal joint prediction algorithm.

[0039] During surgery, the initial pulse parameter scheme is used to generate the corresponding pulse waveform envelope. After obtaining the target pulse parameter, the high-voltage generator dynamically adjusts the pulse waveform envelope according to the target pulse parameter, which makes the pulse parameter at any time during the operation the optimal pulse parameter calculated based on the current multi-modal feature tissue characteristics, thereby greatly improving the accuracy and safety of electrical pulse ablation.

[0040] This module can use a high-density sensor array to collect real-time electric field strength data in the target area and construct a dynamic field strength gradient distribution map to achieve precise regulation of pulse energy. These real-time data are mapped to a pre-established three-dimensional anatomical model that not only contains spatial structure information of the tissue but also integrates biological electrical property parameters (such as electrical conductivity and dielectric constant) of the target tissue. Through finite element algorithms, the system compares the measured field strength distribution with the expected field strength calculated based on the biological electrical model, identifies areas with insufficient or excessive energy deposition, and dynamically adjusts pulse parameters (such as amplitude, frequency, and phase) or reconfigures the spatial arrangement of the electrode array to achieve optimal matching of the electric field distribution with the target treatment area. The entire process is continuously optimized with millisecond-level closed-loop feedback, combined with impedance monitoring and thermal imaging for multi-modal verification, ultimately ensuring that the pulse energy in the target tissue forms a biological electrical effect that meets the treatment requirements while maximizing the protection of surrounding healthy tissue.

[0041] So far, the functions of each module in the embodiment are completed. Figure 1 So far, the functions of each module in the embodiment are completed.

[0042] In the embodiment, for the target tissue to be ablation treated, the target tissue can be ablation treated by the set pulse parameters first, and then multi-modal data of the target tissue is collected during the treatment process, and then the multi-modal data is subjected to feature extraction and fusion to obtain a corresponding multi-modal feature vector; then the multi-modal feature vector is input into the pre-trained time sequence prediction model to determine a real-time target pulse parameter during the treatment process, and then a corresponding pulse energy is generated by the target pulse parameter for treatment, and the action area of the pulse energy is adjusted in real time according to the field strength gradient distribution generated by the pulse energy and the three-dimensional model containing the biological electric characteristics of the target tissue during the treatment process, so that it acts on the target tissue according to the expected target. This solves the problem of insufficient data utilization and insufficient real-time regulation during pulse field ablation treatment, and improves the accuracy and safety of electric pulse ablation.

[0043] In another embodiment, the system further comprises a thermal damage risk detection module for predicting whether the target tissue has a thermal damage risk by the temperature and the pulse energy, and instructing the dynamic execution module to stop generating pulse energy when determined.

[0044] In the embodiment, different temperature thresholds and pulse energy thresholds can be set according to different tissues such as stomach, intestinal tract, liver, etc., and whether the temperature and pulse energy reach the threshold value is judged in real time during the pulse field ablation of the current treated tissue, and the dynamic execution module is instructed to stop generating pulse energy when it is determined that the threshold value is reached, which realizes the monitoring of the risk of thermal damage during the operation.

[0045] In another embodiment, the system further comprises a dangerous area processing module for monitoring the action area of the pulse energy in real time, and adjusting the target pulse parameter when it is determined that the distance between the action area and the preset dangerous area is less than a dangerous threshold.

[0046] In the embodiment, the point cloud information A of the action area of the pulse energy and the point cloud information B of the dangerous area such as blood vessel wall, heart, etc. can be determined, and the minimum value of the coordinate point distance between the point cloud information A and B is determined, and when the minimum value is less than a preset dangerous threshold, it is determined that the pulse energy is about to act on the dangerous area, and then the pulse sequence is re-planned, and the pulse parameter is recalculated and adjusted to avoid damage to the dangerous area, which realizes the monitoring of the dangerous area during the operation.

[0047] Through the above embodiments, the safety of electric pulse ablation is further improved.

[0048] As Figure 2As shown, the present application also provides a pulse field ablation dynamic regulation device, which comprises: a multi-modal perception unit 201, configured to collect multi-modal data generated in a pulse field ablation process, and perform feature extraction and fusion on the multi-modal data to obtain a multi-modal feature vector reflecting tissue features of the multi-modal features; an AI decision unit 202, configured to determine a time series prediction model based on a hybrid architecture of LSTM and Transformer, and take the multi-modal feature vector as an input parameter of the time series prediction model to obtain a target pulse parameter corresponding to the multi-modal feature vector; a dynamic execution unit 203, configured to generate a corresponding pulse energy according to the target pulse parameter, apply the pulse energy to a target tissue for pulse field ablation, synchronously monitor a field intensity gradient distribution of the pulse energy, and make the pulse energy act on the target tissue according to an expected target through the field intensity gradient distribution and a three-dimensional model containing biological electrical features of the target tissue.

[0049] In another embodiment, the multi-modal perception unit comprises a data acquisition unit and a feature vector generation unit; The data acquisition unit comprises an impedance spectrometer for acquiring an impedance spectrum of the target tissue, an ultrasonic transducer for acquiring ultrasonic elasticity data of the target tissue, and a contact pressure micro-electro-mechanical system array for acquiring contact pressure data of the target tissue, and the multi-modal data is composed of the impedance spectrum, the ultrasonic elasticity data, and the contact pressure data; The feature vector generation unit performs deep fusion on the multi-modal data based on a cross-modal neural network of feature-level fusion to obtain a high-dimensional feature representation capable of comprehensively representing dielectric properties and mechanical properties, and determines the high-dimensional feature representation as the multi-modal feature vector.

[0050] In another embodiment, the data acquisition unit further comprises a temperature sensor for acquiring a temperature of the target tissue, and the multi-modal data further comprises the temperature of the target tissue.

[0051] In another embodiment, the device further comprises: a thermal damage risk detection unit 204, configured to predict whether the target tissue has a thermal damage risk through the temperature and the pulse energy, and instruct the dynamic execution unit to stop generating pulse energy when it is determined that the target tissue has the thermal damage risk.

[0052] In another embodiment, the device further comprises: a dangerous area processing unit 205, configured to monitor an action area of the pulse energy in real time, and adjust the target pulse parameter when it is determined that a distance between the action area and a preset dangerous area is less than a dangerous threshold.

[0053] The embodiment of the present application provides a kind of pulse field ablation dynamic regulation system, and based on the system, a kind of pulse field ablation dynamic regulation device is provided, by the above-mentioned system and device, the precision and safety of electric pulse ablation can be improved.

[0054] The embodiment also discloses a kind of computer equipment, as shown in Figure 3 The computer equipment includes processor and memory, at least one instruction is stored in the memory, the at least one instruction is loaded and executed by the processor to realize the method on the above-mentioned pulse field ablation dynamic regulation system of any described.

[0055] In addition, in the implementation of the pulse field ablation dynamic regulation device of the above example, the logical division of each program module is only illustrative, and in actual application, the above-mentioned function allocation can be completed by different program modules according to needs, for example, for the configuration requirements of corresponding hardware or the convenience of software implementation, the internal structure of the pulse field ablation dynamic regulation device is divided into different program modules to complete all or part of the functions described above.

[0056] The above only is the preferred embodiment of the present application, and does not limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A pulsed field ablation dynamic control system, characterized in that, The system includes a multimodal perception module, an AI decision-making module, and a dynamic execution module; The multimodal sensing module is used to collect multimodal data generated during the pulse field ablation process, and to extract and fuse the multimodal data to obtain a multimodal feature vector that reflects the multimodal feature organization characteristics. The AI ​​decision module is used to determine a time series prediction model based on a hybrid architecture of LSTM and Transformer, and then uses the multimodal feature vector as the input parameter of the time series prediction model to obtain the target impulse parameter corresponding to the multimodal feature vector. The dynamic execution module is used to generate corresponding pulse energy according to the target pulse parameters, apply the pulse energy to the target tissue for pulse field ablation, simultaneously monitor the field intensity gradient distribution of the pulse energy, and use the field intensity gradient distribution and a three-dimensional model containing the bioelectric characteristics of the target tissue to make the pulse energy act on the target tissue according to the expected target.

2. The system according to claim 1, characterized in that, The multimodal perception module includes a data acquisition module and a feature vector generation module; The data acquisition module includes an impedance spectrometer for acquiring the impedance spectrum of the target tissue, an ultrasonic transducer for acquiring the ultrasonic elastic data of the target tissue, and a contact pressure microelectromechanical system array for acquiring the contact pressure data of the target tissue. The impedance spectrum, ultrasonic elastic data, and contact pressure data are then combined to form the multimodal data. The feature vector generation module performs deep fusion of the multimodal data based on a cross-modal neural network with feature-level fusion to obtain a high-dimensional feature representation that can comprehensively characterize dielectric and mechanical properties, and determines the high-dimensional feature representation as the multimodal feature vector.

3. The system according to claim 2, characterized in that, The data acquisition module also includes a temperature sensor for acquiring the temperature of the target tissue, and the multimodal data also includes the temperature of the target tissue.

4. The system according to claim 3, characterized in that, The system also includes a thermal damage risk detection module, which is used to estimate whether the target tissue is at risk of thermal damage based on the temperature and the pulse energy, and instruct the dynamic execution module to stop generating pulse energy when the risk is determined.

5. The system according to claim 1, characterized in that, The system also includes a danger zone processing module, which is used to monitor the area of ​​effect of the pulse energy in real time, and adjust the target pulse parameters when it is determined that the distance between the area of ​​effect and the preset danger zone is less than a danger threshold.

6. A pulsed field ablation dynamic control device, characterized in that, The device includes: A multimodal sensing unit is used to collect multimodal data generated during pulse field ablation, and to extract and fuse features from the multimodal data to obtain a multimodal feature vector that reflects the multimodal feature tissue characteristics. The AI ​​decision unit is used to determine the time series prediction model based on the hybrid architecture of LSTM and Transformer, and then use the multimodal feature vector as the input parameter of the time series prediction model to obtain the target impulse parameter corresponding to the multimodal feature vector. The dynamic execution unit is used to generate corresponding pulse energy according to the target pulse parameters, apply the pulse energy to the target tissue for pulse field ablation, simultaneously monitor the field intensity gradient distribution of the pulse energy, and, through the field intensity gradient distribution and a three-dimensional model containing the bioelectric characteristics of the target tissue, make the pulse energy apply to the target tissue according to the expected target.

7. The apparatus according to claim 6, characterized in that, The multimodal sensing unit includes a data acquisition unit and a feature vector generation unit; The data acquisition unit includes an impedance spectrometer for acquiring the impedance spectrum of the target tissue, an ultrasonic transducer for acquiring the ultrasonic elastic data of the target tissue, and a contact pressure microelectromechanical system array for acquiring the contact pressure data of the target tissue. The impedance spectrum, ultrasonic elastic data, and contact pressure data are then combined to form the multimodal data. The feature vector generation unit performs deep fusion of the multimodal data based on a cross-modal neural network with feature-level fusion to obtain a high-dimensional feature representation that can comprehensively characterize dielectric and mechanical properties, and determines the high-dimensional feature representation as the multimodal feature vector.

8. The apparatus according to claim 7, characterized in that, The data acquisition unit also includes a temperature sensor for acquiring the temperature of the target tissue, and the multimodal data also includes the temperature of the target tissue.

9. The apparatus according to claim 8, characterized in that, The device further includes: A thermal damage risk detection unit is used to predict whether the target tissue is at risk of thermal damage based on the temperature and the pulse energy, and to instruct the dynamic execution unit to stop generating pulse energy when the risk is determined.

10. The apparatus according to claim 6, characterized in that, The device further includes: The danger zone processing unit is used to monitor the area of ​​effect of the pulse energy in real time, and adjust the target pulse parameters when it is determined that the distance between the area of ​​effect and the preset danger zone is less than the danger threshold.

Images