A heart electrophysiology dynamic modeling method and system based on big data analysis

Abstract

The application discloses a kind of based on big data analysis cardiac electrophysiology dynamic modeling method and system, specifically related to cardiac three-dimensional modeling technical field, according to the real-time ECG signal and / or real-time electrophysiological signal of current patient by big data model, give initial electric stimulation signal, input initial electric stimulation signal in cardiac mathematical model, simulate the response of cardiac tissue to initial electric stimulation signal, output predicted arrhythmia ECG signal, whether arrhythmia is induced according to predicted arrhythmia ECG signal, if not, according to real-time ECG signal and real-time electrophysiological signal, initial electric stimulation signal or cardiac mathematical model is revised, to obtain recommended electric stimulation signal, and after recommended electric stimulation signal is applied to cardiac tissue, according to real-time ECG signal and real-time electrophysiological signal, cardiac mathematical model and recommended electric stimulation signal are revised, realize according to the specificity of current patient, recommend suitable cardiac electric stimulation strategy, shorten operation time.

Description

Technical Field

[0001] This invention relates to the field of cardiac three-dimensional modeling technology, specifically to a method and system for dynamic modeling of cardiac electrophysiology based on big data analysis. Background Technology

[0002] Electrophysiological testing of the heart is a precise and invasive method for evaluating cardiac electrical function. It involves inserting mapping electrode catheters into the heart to measure electrical activity in different areas to determine the type and location of cardiac lesions. Typically, weak electrical stimulation is also applied to the cardiac tissue through the mapping electrode catheters to induce arrhythmias, thereby locating and treating the lesions. When applying electrical stimulation to cardiac tissue, appropriate electrical stimulation parameters, including pulse width, energy, application location, and cycle, must be set according to the patient's cardiac condition. Inappropriate parameter settings may result in failure to induce arrhythmias or induction of ventricular fibrillation, causing further damage to the patient's heart. Because different patients have different cardiac conditions, the setting of electrical stimulation parameters during surgery currently relies on the surgeon's personal experience. Different electrical stimulation strategies are selected based on the individual patient's cardiac condition, and the electrical stimulation parameters are continuously changed during the operation, gradually inducing arrhythmias through repeated trials. This places high demands on the surgeon's personal experience and knowledge.

[0003] While some electrophysiological examination and treatment systems can perform electrical stimulation using pre-programmed procedures, the complexity and diversity of patients' physical conditions mean that these pre-programmed procedures are usually applied in combination. For example, some surgical cases employ a strategy of pre-programmed stimulation followed by graded incremental stimulation, with subsequent stimulation methods adjusted based on the cardiac electrophysiological signals after stimulation, such as increasing pre-programmed stimulation, changing the baseline circumference of pre-stimulation, or altering the stimulation site. This undoubtedly increases the surgical time and raises the surgical risks. Summary of the Invention

[0004] To address the technical issue that electrical stimulation strategies rely on the personal experience of healthcare professionals, this application provides a method and system for dynamic modeling of cardiac electrophysiology based on big data analysis. The method for dynamic modeling of cardiac electrophysiology based on big data analysis includes the following steps:

[0005] The initial electrical stimulation signal is given based on the patient's real-time ECG signal and / or real-time electrophysiological signal using a big data model;

[0006] The initial electrical stimulation signal is input into the cardiac mathematical model to simulate the response of cardiac tissue to the initial electrical stimulation signal and output a predicted arrhythmia ECG signal.

[0007] Based on the predicted arrhythmia ECG signal and the current patient's historical arrhythmia ECG signals, the initial electrical stimulation signal is determined to be correct. Based on the determination result, the initial electrical stimulation signal or the cardiac mathematical model is corrected, and a recommended electrical stimulation signal is output.

[0008] The recommended electrical stimulation signal is applied to the heart tissue of the current patient, and the cardiac mathematical model and the recommended electrical stimulation signal are corrected based on real-time ECG signals and real-time electrophysiological signals.

[0009] Specifically, the big data model provides the initial electrical stimulation signal through the following steps:

[0010] The big data model selects matching heart tissue samples based on the disease type given by the patient during the initial diagnosis.

[0011] A historical ECG vector is constructed based on the current patient's historical ECG signal, and each component of the historical ECG vector represents the frequency and amplitude of each lead in the historical ECG signal.

[0012] The ECG signals corresponding to the preliminarily screened heart tissue samples are extracted one by one, and the sample ECG vector is constructed. Each component of the sample ECG vector represents the frequency and amplitude of each lead in the sample ECG signal.

[0013] A loss function is constructed based on the historical ECG vector and the sample ECG vector;

[0014] The initial electrical stimulation signal is obtained by iterating through the loss function using a convolutional neural network method.

[0015] Specifically, the mathematical model of the heart tissue is established through the following steps:

[0016] Based on the current patient's cardiac tissue image data, a three-dimensional model of the heart is established, and an initial mathematical model is established based on the three-dimensional model of the heart.

[0017] Based on the initial mathematical model, simulate the ECG signal when the heart rhythm is normal to obtain the training ECG signal. Construct an error term based on the current patient's historical normal ECG signal when the heart rhythm is normal and the training ECG signal. Construct a reward function based on the error term. Use a reinforcement learning algorithm to train the initial mathematical model based on the reward function to obtain the cardiac mathematical model.

[0018] Specifically, the initial mathematical model includes a single-cell reaction model and a conduction system model;

[0019] The single-cell response model is based on the HH model, and the conduction system model is based on a finite element mesh model. The electrical diffusion rate of each node in the mesh model is calculated based on the distance of the cardiac tissue from one or more anatomical structures.

[0020] Specifically, when the initial electrical stimulation signal is incorrect, the following steps determine whether to correct the initial electrical stimulation signal or the cardiac mathematical model:

[0021] A cardiac mathematical model was used to simulate the predicted normal ECG and electrophysiological signals of the current patient, and a predicted normal state vector was constructed.

[0022] A real-time state vector is constructed based on the real-time ECG signal and the real-time electrophysiological signal;

[0023] Subtract the real-time state vector from the predicted normal state vector to obtain the state error vector;

[0024] If the magnitude of the state error vector is less than the waveform accuracy threshold, the initial electrical stimulation signal given by the big data model is corrected to obtain a recommended electrical stimulation signal; otherwise, the cardiac mathematical model is corrected.

[0025] Specifically, the initial electrical stimulation signal is corrected through the following steps:

[0026] An error term is constructed based on the predicted arrhythmia ECG signal and the historical arrhythmia ECG signal. A reward function is constructed based on the error term and the initial electrical stimulation signal. A reinforcement learning algorithm is used to optimize the initial electrical stimulation signal and output a recommended electrical stimulation signal.

[0027] Specifically, the cardiac mathematical model is modified through the following steps:

[0028] An error term is constructed based on the predicted normal state vector and the real-time state vector, and a reward function is constructed based on the error term. A reinforcement learning algorithm is then used to correct the cardiac mathematical model.

[0029] Furthermore, the method for dynamic modeling of cardiac electrophysiology also includes the following steps:

[0030] The recommended electrical stimulation signal is applied to the heart tissue of the current patient, and real-time ECG and real-time electrophysiological signals are acquired.

[0031] Based on the real-time ECG signal and the real-time electrophysiological signal, it is determined whether an arrhythmia has occurred. If so, the recommended electrical stimulation signal is recorded into the big data model. If not, the cardiac mathematical model and the recommended electrical stimulation signal are corrected a second time based on the real-time ECG signal and the real-time electrophysiological signal.

[0032] Specifically, the cardiac mathematical model and the recommended electrical stimulation signal are corrected a second time using the following method:

[0033] The recommended electrical stimulation signal is input into the cardiac mathematical model, and the predicted arrhythmia ECG signal and predicted arrhythmia electrophysiological signal are output to construct the predicted arrhythmia vector.

[0034] Acquire the current patient's real-time ECG and electrophysiological signals, and construct a real-time state vector;

[0035] An error term is constructed based on the predicted arrhythmia vector and the real-time state vector, and a reward function is constructed based on the error term. A reinforcement learning algorithm is then used to perform a secondary correction on the cardiac mathematical model.

[0036] The recommended electrical stimulation signal is modified based on the second-corrected cardiac mathematical model.

[0037] This invention also provides a cardiac electrophysiological dynamic modeling system based on big data analysis to implement the above method, comprising: a cardiac state measurement module, a big data analysis module, and a cardiac mathematical model. The cardiac state measurement module is used to acquire the real-time ECG signal and real-time electrophysiological signal of the current patient. The big data analysis module is used to provide an initial electrical stimulation signal based on the real-time ECG signal and / or the real-time electrophysiological signal. The cardiac mathematical model is used to simulate the response of cardiac tissue to the initial electrical stimulation signal based on the initial electrical stimulation signal and output a predicted arrhythmia ECG signal. The big data analysis module is also used to determine whether the initial electrical stimulation signal is correct based on the predicted arrhythmia ECG signal and the patient's historical arrhythmia ECG signals during arrhythmia, and to correct the initial electrical stimulation signal or the cardiac mathematical model based on the comparison results to output a recommended electrical stimulation signal.

[0038] Furthermore, the big data analysis module is also used to determine whether arrhythmia occurs based on the acquired real-time ECG signal and real-time electrophysiological signal after the recommended electrical stimulation signal is applied to the cardiac tissue, and to select whether to input the recommended electrical stimulation signal into the big data model or to modify the cardiac mathematical model and / or the recommended electrical stimulation signal based on the real-time ECG signal and real-time electrophysiological signal.

[0039] The technical effects and advantages of this invention are as follows: Based on big data analysis of the current patient's real-time ECG and electrophysiological signals, a recommended electrical stimulation signal specific to the current patient is obtained. Furthermore, the current patient's cardiac mathematical model is trained based on historical ECG signals, improving its accuracy. Simultaneously, by continuously comparing the predicted ECG and electrophysiological signals output by the cardiac mathematical model with the real-time ECG and electrophysiological signals, errors in the cardiac mathematical model are corrected, achieving real-time updates of the cardiac mathematical model and electrical stimulation signals. This provides timely and appropriate electrical stimulation signals to medical staff, reducing surgical difficulty and shortening surgical time. Simultaneously, the cardiac mathematical model can be dynamically corrected based on the current patient's cardiac feedback, continuously improving its accuracy. Attached Figure Description

[0040] Figure 1 This is an overall flowchart of the method of the present invention.

[0041] Figure 2 This is a flowchart of the process for obtaining and correcting the cardiac mathematical model in the method of the present invention.

[0042] Figure 3 This is a flowchart illustrating the secondary correction of the recommended electrical stimulation signal and the cardiac mathematical model in the method of the present invention.

[0043] Figure 4 This is the display interface of a common electrophysiological examination instrument.

[0044] Figure 5 This is an overall structural diagram of the system disclosed in this invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] With advancements in medical technology, the recording of various indicators of a patient's cardiac tissue (such as disease type, electrocardiogram, cardiac MRI imaging data, cardiac CT scan images, and electrophysiological signals) has become increasingly sophisticated. Based on this data, analyzing the relationship between the heart's true physiological state (e.g., disease type) and various examination signals for disease diagnosis and treatment has become widely used. Simultaneously, biological research on cardiac tissue has provided numerous mathematical models to simulate the electrophysiological activity of cardiac tissue, including the response and conduction of electrical signals by cardiac tissue cells. This makes it possible to accurately analyze the effects of electrical stimulation signals on cardiac tissue.

[0047] A typical cardiac database comprises multiple cardiac tissue samples, each including ECG signals, electrophysiological signals, disease type, electrical stimulation strategy, and other patient information (such as age, sex, physical condition, underlying diseases, etc.). Based on this information, deep learning or reinforcement learning methods can be used to effectively analyze the relationship between ECG signals (electrocardiogram), electrophysiological signals, disease type, and electrical stimulation strategy.

[0048] Based on the above analysis, the present invention provides a method and system for dynamic modeling of cardiac electrophysiology based on big data analysis. The following are specific embodiments of the present invention.

[0049] It should be noted that, in this application, when comparing signals or vectors composed of signals, the term "consistency" should more accurately be defined as an error less than a certain threshold. In practice, even for the same patient with no significant changes in their physical condition, their ECG or electrophysiological signals fluctuate within a small range. Therefore, ECG or electrophysiological signals collected at different times are unlikely to be completely consistent. When conducting comparative analysis, a small range of error between the two components being compared should be allowed. Considering this is a conventional technical approach, the process of comparing the error with a specific threshold has been omitted in some paragraphs to save space.

[0050] Example 1

[0051] refer to Figures 1 to 3 Embodiment 1 of the present invention provides a method for dynamic modeling of cardiac electrophysiology based on big data analysis, comprising the following steps:

[0052] S1. Obtain the current patient's real-time ECG and real-time electrophysiological signals, construct a real-time state vector, and the big data model provides the initial electrical stimulation signal based on the real-time state vector.

[0053] Electrophysiological signals are electrical signals obtained by mapping the electrical activity of cardiac tissues inside the heart. Guided by X-rays or a three-dimensional mapping system, several thin, flexible electrode catheters are inserted through blood vessels (usually the femoral vein or subclavian vein) into specific locations in the heart (atria, ventricles, coronary sinus, etc.). Electrophysiological signals from different parts of the heart are recorded through these catheters (intracardiac electrocardiography), and weak electrical stimulation (programmed electrical stimulation) is delivered through the catheters to induce arrhythmias that the patient usually experiences. Simultaneously, a three-dimensional mapping system is used to construct cardiac electrical activation sequence maps (activation maps) and voltage maps, precisely locating the origin or key pathways of the arrhythmia.

[0054] Figure 4 This image shows the interface of a common electrophysiological testing instrument, displaying both a three-dimensional model of the heart and the position of the mapping electrode catheter within the heart. The right side of the image shows the waveforms of each lead in the ECG signal, as well as the electrophysiological signals measured by multiple electrodes on the mapping electrode catheter. It is evident that both the ECG and electrophysiological signals are waveforms of voltage changing over time, and ECG signals typically have multiple leads (e.g., a common 12-lead ECG), with a certain correlation between the waveforms of each lead. Similarly, electrophysiological signals also have multiple mapping electrodes, each acquiring a voltage waveform, and there is also a certain correlation between the waveforms of each mapping electrode. Furthermore, there is a correlation between electrophysiological signals and ECG signals. Therefore, when using big data to find suitable electrical stimulation signals for a current patient, it is best to refer to both real-time ECG signals and real-time electrophysiological signals simultaneously.

[0055] Current technologies for diagnosing cardiac diseases based on ECG or electrophysiological signals typically extract specific waveforms (such as QRS complexes) and analyze their duration, peak values, and intervals to determine whether the waveforms are normal and to diagnose the disease type. This requires machine learning or other artificial intelligence methods to analyze and classify the waveform characteristics and disease types. This process is not only time-consuming but also prone to misdiagnosis. Therefore, when analyzing and processing ECG and electrophysiological signals, time-domain signals should be converted to frequency-domain signals for filtering and to improve analysis speed.

[0056] Specifically, in step S1, after periodically acquiring a segment of real-time ECG signal and real-time electrophysiological signal, frequency domain analysis can be performed, and a real-time state vector can be constructed based on the analysis results. For example, Fourier transform can be performed on the data of each lead in the real-time ECG signal to extract the amplitude and frequency of the first two orders, and the same operation can be performed on each waveform of the real-time electrophysiological signal. Then, a real-time state vector can be constructed based on the extracted amplitude and frequency.

[0057]

[0058] in, This represents the first-order amplitude of the first lead in the real-time ECG signal. This represents the first-order frequency of the first lead in the real-time ECG signal. This represents the second-order amplitude of the first lead in the real-time ECG signal. This represents the second-order amplitude of the first lead in the real-time ECG signal. This represents the first-order amplitude of the first electrode in a real-time electrophysiological signal. This represents the first-order frequency of the first electrode in real-time electrophysiological signals.

[0059] The dimension of the real-time state vector is related to the number of leads and electrodes involved in the calculation of the real-time ECG signal. In particular, the position of the mapping electrode catheter also affects the obtained real-time electrophysiological signal. When performing measurements and diagnoses, medical staff will select the number and position of the mapping electrode catheters inserted into the heart as needed. Therefore, the position information of each electrode should also be included in the real-time state vector to make the analysis results more accurate.

[0060] Considering that the heart rate is generally 80-130 beats per minute, the sampling duration of each real-time ECG signal and real-time electrophysiological signal can be set to 2 to 4 seconds to obtain a complete waveform of 2 to 3 heartbeats, balancing calculation speed and effectiveness.

[0061] The process of generating initial electrical stimulation signals based on real-time state vectors using big data models can be performed before surgery to reduce the amount of data computation and processing time during the procedure. This includes the following steps:

[0062] S11. The big data model selects matching heart tissue samples based on the disease type given by the patient during the initial diagnosis.

[0063] S12. Based on the patient's historical ECG signals, find matching heart tissue samples from the initially screened heart tissue samples, obtain the electrical stimulation strategy corresponding to the heart tissue sample, and set the electrical stimulation strategy as the initial electrical stimulation signal.

[0064] In particular, arrhythmias are often paroxysmal. Therefore, the ECG signals acquired during initial examinations may include both ECG signals from when the heart rhythm is normal and historical ECG signals from when arrhythmias occur. When analyzing large datasets, ECG signals of the same type should be compared. For example, during surgery, electrical stimulation may be used to induce arrhythmias in cardiac tissue. The real-time ECG signals acquired before this should be those from when the heart rhythm is normal. Therefore, during large-scale data analysis, these signals should be compared with historical ECG signals from when the heart rhythm is normal. Of course, arrhythmias may also occur paroxysmically during surgery. If the acquired historical ECG signals are those from when arrhythmias occur, they should be compared with ECG signals from when arrhythmias occur in the database. Therefore, the above examples are not intended to limit the invention, but are merely illustrative.

[0065] The process of matching cardiac tissue samples described above can be achieved by constructing historical ECG vectors. Matching is determined by constructing a difference vector and calculating its modulus, thus reducing computational load and improving processing speed. Specifically, it includes the following steps:

[0066] S13. Construct a historical ECG vector based on the current patient's historical ECG signal. Each component of the historical ECG vector represents the frequency and amplitude of each lead in the historical ECG signal.

[0067] S14. Extract the ECG signals of the preliminarily screened heart tissue samples one by one, and construct the sample ECG vector. The information represented by each component of the sample ECG vector is the same as that of the historical ECG vector.

[0068] S15. Subtract the sample ECG vector from the historical ECG vector to obtain the matching difference vector. Calculate the magnitude of the matching difference vector and compare it with the matching accuracy threshold. If the magnitude is less than the matching accuracy threshold, the historical ECG signal matches the heart tissue sample.

[0069] The matching accuracy threshold needs to be determined based on the number of heart tissue samples in the big data model. When the number of heart tissue samples is small, the matching accuracy threshold should be increased appropriately to reduce the matching difficulty. For example, it can be set to 1.5-5, which can balance accuracy and matching success rate for most application scenarios.

[0070] Big data analytics can also be combined with deep learning methods for data interpolation to infer electrical stimulation signals. Specifically, this includes the following steps:

[0071] S16. Construct a loss function based on historical ECG vectors and sample ECG vectors;

[0072] S17. Using a convolutional neural network method, the loss function is iterated to perform interpolation operations on all electrical stimulation signals corresponding to the initially screened heart tissue samples to obtain the initial electrical stimulation signal.

[0073] Electrical stimulation signals are pulsed electrical signals with varying time intervals, durations, and amplitudes. Therefore, in big data models, the data structure of electrical stimulation signals can be organized as vectors or arrays. During interpolation operations, the electrical stimulation signals can be organized into vector form for easier computation and analysis. For example, the following formula illustrates a vector form of an electrical stimulation signal:

[0074]

[0075] in, The location of the applied electrical stimulation is indicated. In different disease diagnoses and treatments, electrical stimulation is applied to different locations. Since these locations are usually fixed, such as the high right atrium, the His bundle region, or the coronary sinus, they can be represented by numbers. Of course, for greater accuracy, the location of the applied electrical stimulation can also be represented using three-dimensional coordinates, in which case the dimension of the electrical stimulation signal vector needs to be changed accordingly. This represents the amplitude of the i-th pulse signal. This represents the duration of the i-th pulse signal segment. If there is no pulse in this segment, then... .

[0076] After obtaining the initial electrical stimulation signal, a mathematical model of the heart tissue needs to be established to verify the correctness of the initial electrical stimulation signal:

[0077] S2. Input the initial electrical stimulation signal into the cardiac mathematical model to simulate the response of cardiac tissue to the initial electrical stimulation signal and output the ECG signal predicting arrhythmia.

[0078] In step S2, the cardiac mathematical model consists of two parts: a single-cell response model, which describes the response of myocardial cells (such as the sinoatrial node, atrioventricular node, His bundle, etc.) to electrical stimulation; and a conduction system model, which describes the transmission process of current in the conduction system (such as Purkinje fibers, ventricular myocytes, etc.).

[0079] The Hdgkin-Huxley model (HH model) is a classic model of the nerve cell membrane. This model considers both the passive electrical properties of the nerve cell membrane and the dependence of sodium and potassium channel conductance on membrane potential and time, thus solving the problem of the mechanism of action potential generation and propagation.

[0080] The following formula is a single-cell response model based on the Hdgkin-Huxley model:

[0081]

[0082] In the formula, This refers to the membrane potential of cardiomyocytes. That is, the change in membrane potential of myocardial cells over time. For the membrane capacitance of cardiomyocytes, This refers to the ion current of myocardial cells. The electrical stimulation current applied to the mapping catheter.

[0083] Based on the HH model described above, the state vector of the single-cell response model at time j can be constructed as follows:

[0084]

[0085] The prior art CN106456269B discloses a real-time simulation system for cardiac electrophysiology. It uses medical image data of patients to generate a patient-specific anatomical model of the heart, and employs the Lattice-Boltzmann Method (LBM-EP) combined with a finite element model of the heart to construct a computational mesh for the heart. By assigning different electrical diffusion rates to each node of the computational mesh, the conduction system of the heart is simulated. Based on this mesh model, the state vector of the conduction system at a certain moment can be constructed.

[0086]

[0087] In the formula, Let be the electric diffusion rate of the i-th grid node at time j.

[0088] Based on the above mathematical model, a specific modification is made to the mathematical model according to the patient's current actual cardiac condition, resulting in a cardiac mathematical model that conforms to the patient's actual condition. Figure 3 Specifically, it includes the following steps:

[0089] S21. Based on the current patient's cardiac image data, establish a three-dimensional cardiac model. Based on the three-dimensional cardiac model, establish an initial mathematical model, which includes a single-cell response model and a conduction system model. The single-cell response model is based on the HH model, and the conduction system model is based on a finite element mesh model. The electrical diffusion rate of each node in the mesh model is calculated based on the distance of the cardiac tissue from one or more anatomical structures.

[0090] S22. Simulate ECG signals when the heart rhythm is normal based on the initial mathematical model to obtain training ECG signals. Construct error terms based on the current patient's historical normal ECG signals when the heart rhythm is normal and training ECG signals. Construct a reward function based on the error terms. Use reinforcement learning algorithm to train the initial mathematical model to obtain a cardiac mathematical model.

[0091] The above steps complete the initial revision of the mathematical model of the patient's current cardiac tissue. Since the initial construction of the cardiac mathematical model is performed preoperatively, and the patient has not yet undergone electrophysiological examination, only ECG signals can be used to revise the cardiac mathematical model.

[0092] In particular, the ECG signal in the above steps should also contain data from multiple leads. The specific number of leads should be the same as that of the real-time ECG signal to ensure a consistent data structure and facilitate subsequent calculations and processing.

[0093] The cardiac mathematical model, after the above steps, has a certain degree of specificity matching with the current patient and can be used to simulate the response of cardiac tissue to electrical stimulation signals, and to determine whether electrical stimulation signals can induce arrhythmias in cardiac tissue.

[0094] S3. Based on the predicted arrhythmia ECG signal and the patient's historical arrhythmia ECG signals during the current arrhythmia, determine whether the initial electrical stimulation signal is correct. If so, use the initial electrical stimulation signal as the recommended electrical stimulation signal. If not, correct the initial electrical stimulation signal or cardiac mathematical model based on the real-time state vector and output the recommended electrical stimulation signal.

[0095] In step S3, the following steps determine whether it is necessary to correct the initial electrical stimulation signal or to correct the cardiac mathematical model:

[0096] S31. Using a cardiac mathematical model to simulate the predicted normal ECG signal and predicted normal electrophysiological signal of the current patient, construct a predicted normal state vector, subtract the real-time state vector from the predicted normal state vector to obtain the state error vector, determine whether the magnitude of the state error vector is less than the waveform accuracy threshold, if so, then correct the initial electrical stimulation signal given by the big data model to obtain the recommended electrical stimulation signal, if not, then correct the cardiac mathematical model.

[0097] The waveform accuracy threshold is usually set between 0.7 and 1.5, which can balance calculation accuracy and time in most cases. A value that is too small may fail to meet the requirements after multiple iterations, resulting in calculation failure.

[0098] Generally, when the simulated ECG signal for measuring arrhythmia is consistent with the historical ECG signal for measuring arrhythmia, it indicates that the accuracy of the initial electrical stimulation signal and the cardiac mathematical model can simultaneously meet the requirements. Therefore, the initial electrical stimulation signal can be provided to the surgeon as a recommended electrical stimulation signal for reference.

[0099] The comparison method between the predicted arrhythmia ECG signal and the historical arrhythmia ECG signal can also adopt the vector comparison method. By extracting the frequency and amplitude information of the first few orders through frequency analysis, the corresponding vector is constructed to determine whether they are consistent. The specific process is the same as steps S13 to S15, and will not be repeated here.

[0100] If it is determined that the initial electrical stimulation signal cannot induce the current patient's arrhythmia, then it should be determined whether the initial electrical stimulation signal needs to be modified or whether the problem is due to insufficient precision of the cardiac mathematical model.

[0101] The cardiac mathematical model constructed in step S2 is only corrected using ECG signals. However, due to the inherent limitations of ECG signals and the introduction of new errors during the correction process, the cardiac mathematical model may differ from the current cardiac state of the patient. Therefore, further verification and correction of the cardiac mathematical model using electrophysiological signals are necessary. When the normal cardiac activity simulated by the cardiac mathematical model matches the normal cardiac activity of the current patient, it indicates that the cardiac mathematical model is correct, and the initial electrical stimulation signal provided by the big data model needs correction.

[0102] Specifically, the initial electrical stimulation signal is corrected through the following steps:

[0103] S32. Construct an error term based on the predicted arrhythmia ECG signal and the patient's historical arrhythmia ECG signals during the current arrhythmia. Construct a reward function based on the error term and the initial electrical stimulation signal. Optimize the initial electrical stimulation signal using a reinforcement learning algorithm and output a recommended electrical stimulation signal.

[0104] The error term can also be constructed using a vector.

[0105] Reinforcement learning algorithms can be used to change the weight coefficients of each variable based on the correlation between each variable and the error term in the initial electrical stimulation signal, thereby accelerating the calculation process and improving the calculation accuracy.

[0106] The same method can be used to modify the mathematical model of the heart:

[0107] S33. Construct an error term based on the predicted normal state vector and the real-time state vector, and construct a reward function based on the error term. Use a reinforcement learning algorithm to correct the cardiac mathematical model.

[0108] After revising the cardiac mathematical model, it is necessary to simulate the heart's response to the initial electrical stimulation signal again using the cardiac mathematical model to verify whether the initial electrical stimulation signal is correct.

[0109] S34. After correcting the cardiac mathematical model, the initial electrical stimulation signal is input into the corrected cardiac tissue model to simulate the cardiac tissue response to the initial electrical stimulation signal and output the ECG signal predicting arrhythmia.

[0110] S35. Based on the predicted arrhythmia ECG signal and the patient's historical arrhythmia ECG signals during the current arrhythmia, determine whether the initial electrical stimulation signal is correct. If so, use the initial electrical stimulation signal as the recommended electrical stimulation signal. If not, construct an error term based on the predicted arrhythmia ECG signal and the historical arrhythmia ECG signal, construct a reward function based on the error term, and use a reinforcement learning algorithm to correct the initial electrical stimulation signal, outputting the recommended electrical stimulation signal.

[0111] After the recommended electrical stimulation signal is output according to the above steps, it is necessary to determine whether the arrhythmia has been successfully induced based on the real-time ECG and real-time electrophysiological signals. (Refer to...) Figure 3 Specifically, it includes the following steps:

[0112] S4. Apply the recommended electrical stimulation signal to the current patient's cardiac tissue and acquire the current patient's real-time ECG signal and real-time electrophysiological signal. Determine whether the current patient has arrhythmia based on the real-time ECG signal and real-time electrophysiological signal. If yes, input the recommended electrical stimulation signal into the big data model. If no, correct the cardiac mathematical model and the recommended electrical stimulation signal based on the real-time ECG signal and real-time electrophysiological signal.

[0113] Whether an arrhythmia has occurred can be determined by the surgeon's experience, or by comparing the real-time ECG signal with the patient's historical arrhythmia ECG signals using a program.

[0114] If the patient's current cardiac tissue does not induce arrhythmias, it means the recommended electrical stimulation signal is incorrect, and the cardiac mathematical model does not match the actual condition of the patient's heart. In this case, it is necessary to perform a secondary correction to the cardiac mathematical model based on real-time ECG and electrophysiological signals, and then adjust the recommended electrical stimulation signal according to the corrected cardiac mathematical model.

[0115] S41. Input the recommended electrical stimulation signal into the cardiac mathematical model, output the ECG signal and electrophysiological signal of predicted arrhythmia, and construct the predicted arrhythmia vector.

[0116] S42. Obtain the current patient's real-time ECG signal and real-time electrophysiological signal, and construct a real-time state vector;

[0117] S43. Construct an error term based on the predicted arrhythmia vector and the real-time state vector, and construct a reward function based on the error term. Use a reinforcement learning algorithm to perform a secondary correction on the cardiac mathematical model.

[0118] S44. The recommended electrical stimulation signal is corrected based on the second-corrected cardiac mathematical model.

[0119] During subsequent surgeries, steps S4, S41 to S44 are repeatedly executed. Through continuous comparison and correction, the goal of dynamically adjusting the cardiac mathematical model and recommending electrical stimulation signals is achieved, reducing the difficulty of setting various parameters of the electrical stimulation signal and shortening the operation time.

[0120] Example 2

[0121] refer to Figure 5 This application provides a cardiac electrophysiological dynamic modeling system based on big data analysis to implement the above-mentioned dynamic modeling method, including: a cardiac state measurement module, a big data analysis module, and a cardiac simulation module.

[0122] The cardiac status measurement module is used to acquire the current patient's real-time ECG and real-time electrophysiological signals;

[0123] The big data analysis module is used to provide initial electrical stimulation signals based on real-time ECG and real-time electrophysiological signals;

[0124] The cardiac mathematical model is used to simulate the response of cardiac tissue to the initial electrical stimulation signal and output a predictive ECG signal for arrhythmia.

[0125] The big data analysis module is also used to determine whether the initial electrical stimulation signal is correct based on the predicted arrhythmia ECG signal and the patient's historical arrhythmia ECG signals during the current arrhythmia, and to correct the initial electrical stimulation signal or the cardiac mathematical model based on the comparison results.

[0126] The big data analysis module is also used to determine whether arrhythmia occurs after applying the recommended electrical stimulation signal to the heart tissue, based on the acquired real-time ECG and real-time electrophysiological signals. Based on the determination result, it can choose to input the recommended electrical stimulation signal into the big data model or modify the heart mathematical model and the recommended electrical stimulation signal based on the real-time ECG and real-time electrophysiological signals.

[0127] Furthermore, the mathematical model of the heart includes a single-cell response model and a conduction system model.

[0128] Single-cell response models are used to describe the response of cardiomyocytes to electrical stimulation, while conduction system models are used to describe the transmission of current in the cardiac conduction system.

[0129] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic modeling of cardiac electrophysiology based on big data analysis, characterized in that, Includes the following steps: The initial electrical stimulation signal for the current patient is provided through a big data model; The initial electrical stimulation signal is input into the cardiac mathematical model to simulate the response of cardiac tissue to the initial electrical stimulation signal and output a predicted arrhythmia ECG signal. Based on the predicted arrhythmia ECG signal and the current patient's historical arrhythmia ECG signals, the initial electrical stimulation signal is determined to be correct. Based on the determination result, the initial electrical stimulation signal or the cardiac mathematical model is corrected, and a recommended electrical stimulation signal is output. The big data model provides the initial electrical stimulation signal through the following steps: The big data model selects matching heart tissue samples based on the disease type given by the current patient during the initial diagnosis; A historical ECG vector is constructed based on the current patient's historical ECG signal, and each component of the historical ECG vector represents the frequency and amplitude of each lead in the historical ECG signal. The ECG signals corresponding to the preliminarily screened heart tissue samples are extracted one by one, and the sample ECG vector is constructed. Each component of the sample ECG vector represents the frequency and amplitude of each lead in the sample ECG signal. A loss function is constructed based on the historical ECG vector and the sample ECG vector; The initial electrical stimulation signal is obtained by iterating through the loss function using a convolutional neural network method.

2. The method according to claim 1, characterized in that, The cardiac mathematical model was established through the following steps: Based on the current patient's cardiac tissue image data, a three-dimensional model of the heart is established, and an initial mathematical model is established based on the three-dimensional model of the heart. Based on the initial mathematical model, simulate the ECG signal when the heart rhythm is normal to obtain the training ECG signal. Construct an error term based on the current patient's historical normal ECG signal when the heart rhythm is normal and the training ECG signal. Construct a reward function based on the error term. Use a reinforcement learning algorithm to train the initial mathematical model based on the reward function to obtain the cardiac mathematical model.

3. The method according to claim 2, characterized in that, The initial mathematical model includes a single-cell reaction model and a conduction system model; The single-cell response model is based on the HH model, and the conduction system model is based on a finite element mesh model. The electrical diffusion rate of each node in the mesh model is calculated based on the distance of the cardiac tissue from one or more anatomical structures.

4. The method according to claim 1, characterized in that, When the initial electrical stimulation signal is incorrect, the following steps determine whether to correct the initial electrical stimulation signal or the cardiac mathematical model: A cardiac mathematical model was used to simulate the predicted normal ECG and electrophysiological signals of the current patient, and a predicted normal state vector was constructed. A real-time state vector is constructed based on real-time ECG and real-time electrophysiological signals; Subtract the real-time state vector from the predicted normal state vector to obtain the state error vector; If the magnitude of the state error vector is less than the waveform accuracy threshold, the initial electrical stimulation signal given by the big data model is corrected to obtain a recommended electrical stimulation signal; otherwise, the cardiac mathematical model is corrected.

5. The method according to claim 4, characterized in that, The initial electrical stimulation signal is corrected through the following steps: An error term is constructed based on the predicted arrhythmia ECG signal and the historical arrhythmia ECG signal. A reward function is constructed based on the error term and the initial electrical stimulation signal. A reinforcement learning algorithm is used to optimize the initial electrical stimulation signal and output a recommended electrical stimulation signal.

6. The method according to claim 4, characterized in that, The cardiac mathematical model is corrected through the following steps: An error term is constructed based on the predicted normal state vector and the real-time state vector, and a reward function is constructed based on the error term. A reinforcement learning algorithm is then used to correct the cardiac mathematical model.

7. The method according to claim 1, characterized in that, The method further includes the following steps: The recommended electrical stimulation signal is applied to the heart tissue of the current patient, and real-time ECG and real-time electrophysiological signals are acquired. Based on the real-time ECG signal and the real-time electrophysiological signal, it is determined whether an arrhythmia has occurred. If so, the recommended electrical stimulation signal is recorded into the big data model. If not, the cardiac mathematical model and the recommended electrical stimulation signal are corrected a second time based on the real-time ECG signal and the real-time electrophysiological signal.

8. The method according to claim 7, characterized in that, The cardiac mathematical model and the recommended electrical stimulation signal were corrected a second time using the following method: The recommended electrical stimulation signal is input into the cardiac mathematical model, and the predicted arrhythmia ECG signal and predicted arrhythmia electrophysiological signal are output to construct the predicted arrhythmia vector. Acquire the current patient's real-time ECG and electrophysiological signals, and construct a real-time state vector; An error term is constructed based on the predicted arrhythmia vector and the real-time state vector, and a reward function is constructed based on the error term. A reinforcement learning algorithm is then used to perform a secondary correction on the cardiac mathematical model. The recommended electrical stimulation signal is modified based on the second-corrected cardiac mathematical model.

9. A dynamic modeling system for cardiac electrophysiology based on big data analysis, used to implement the method of claim 8, characterized in that, include: Cardiac condition measurement module, big data analysis module, and cardiac mathematical model; The cardiac status measurement module is used to acquire the current patient's real-time ECG signal and real-time electrophysiological signal; The big data analysis module is used to provide the initial electrical stimulation signal for the current patient through a big data model. The cardiac mathematical model is used to simulate the response of cardiac tissue to the initial electrical stimulation signal and output a predicted ECG signal for arrhythmia. The big data analysis module is also used to determine whether the initial electrical stimulation signal is correct based on the predicted arrhythmia ECG signal and the patient's historical arrhythmia ECG signal at the time of the current arrhythmia, and to correct the initial electrical stimulation signal or the cardiac mathematical model based on the comparison results, so as to output a recommended electrical stimulation signal. The big data analysis module provides the initial electrical stimulation signal for the current patient through the following steps: The big data model selects matching heart tissue samples based on the disease type given by the current patient during the initial diagnosis; A historical ECG vector is constructed based on the current patient's historical ECG signal, and each component of the historical ECG vector represents the frequency and amplitude of each lead in the historical ECG signal. The ECG signals corresponding to the preliminarily screened heart tissue samples are extracted one by one, and the sample ECG vector is constructed. Each component of the sample ECG vector represents the frequency and amplitude of each lead in the sample ECG signal. A loss function is constructed based on the historical ECG vector and the sample ECG vector; The initial electrical stimulation signal is obtained by iterating through the loss function using a convolutional neural network method.

10. The system according to claim 9, characterized in that, The big data analysis module is also used to determine whether arrhythmia occurs after applying the recommended electrical stimulation signal to the heart tissue based on the acquired real-time ECG signal and real-time electrophysiological signal, and to select whether to input the recommended electrical stimulation signal into the big data model or to modify the heart mathematical model and / or the recommended electrical stimulation signal based on the real-time ECG signal and real-time electrophysiological signal according to the judgment result.

Images