Method for predicting recurrence risk after atrial fibrillation ablation and memory

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an atrial fibrillation ablation postoperative recurrence risk prediction method and a memory, and the method comprises the steps: S1, collecting intraoperative intracavitary electrocardiograms at a plurality of moments for a to-be-predicted patient; and S2, sequentially inputting the intraoperative intracavity electrocardiogram into a risk prediction model according to a time sequence to obtain a risk prediction result for the to-be-predicted patient. Intraoperative intracavity electrocardiograms of a patient to be predicted at multiple moments are collected and then input into a risk prediction model, graph related features in the electrocardiograms are extracted and predicted, effective recognition and prediction of multi-guide related features commonly used in the diagnosis process are achieved, and the risk prediction accuracy is improved. And the postoperative recurrence risk of the to-be-predicted patient can be predicted more accurately.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method and memory for predicting the risk of recurrence after atrial fibrillation ablation. Background Technology

[0002] Atrial fibrillation (AF) is a common rapid cardiac arrhythmia with significant adverse consequences. In clinical practice, catheter ablation has been widely used as a treatment for AF. The CABANA trial showed that catheter ablation is more effective than drug therapy in reducing AF recurrence. However, regrettably, some patients who undergo catheter ablation still experience AF recurrence. AF recurrence may be due to the complex pathogenesis of AF, individual differences during treatment, and the combined effects of various factors, including the postoperative cardiac electrophysiological state. Therefore, further research and exploration are needed to find more effective methods to predict and prevent AF recurrence.

[0003] In existing technologies, there are methods for processing and predicting electrocardiogram (ECG) signals based on artificial intelligence (AI) technology to determine the risk of atrial fibrillation (AF). For example, Chinese patent CN202010211091.9 discloses an AF prediction device and apparatus based on deep learning. The device includes: an information extraction module for acquiring multi-lead raw ECG data and obtaining the ECG rhythm information corresponding to each lead; an information processing module for using filters to determine the R-point positions in the ECG rhythm information and the heartbeat cycles and waveform information corresponding to each R-point position to construct a waveform information set; a noise reduction autoencoder for performing feature extraction and dimensionality reduction on the waveform information set corresponding to each heartbeat cycle to obtain multi-dimensional feature data corresponding to the waveform information set, and calculating the mean and variance of each dimension of feature data; and a result prediction module for comparing the mean and variance of the multi-dimensional feature data corresponding to different heartbeat cycles with an adaptive threshold for the heartbeat cycle in chronological order to determine whether AF occurs in the heartbeat cycle.

[0004] However, in actual implementation, the inventors found that such technical solutions usually rely on time-related models to process ECG signal sequences, such as RNN and LSTM models. They mainly focus on signal features in continuous sequences, such as the time-related features of ECG signals in a certain conduction chamber. They do not pay enough attention to the multi-lead features that are often important in the actual diagnosis of atrial fibrillation, resulting in poor diagnostic results or weak interpretability. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, a method for predicting the risk of recurrence after atrial fibrillation ablation is provided.

[0006] The specific technical solution is as follows:

[0007] A method for predicting the risk of recurrence after atrial fibrillation ablation includes:

[0008] Step S1: Collect intraoperative intracavitary electrocardiograms at multiple time points for the patient to be predicted;

[0009] Step S2: Input the intraoperative intracavitary electrocardiograms into the risk prediction model in chronological order to obtain the risk prediction results for the patient to be predicted.

[0010] On the other hand, step S1 includes:

[0011] Step A1: During the surgical procedure, the current image is captured by the electrocardiogram monitor at preset time intervals as the intraoperative intracavitary electrocardiogram;

[0012] The electrocardiogram monitor is connected to a coronary sinus electrode.

[0013] On the other hand, step S1 includes:

[0014] Step B11: Obtain the intraoperative electrocardiogram data sequence of the patient to be predicted;

[0015] Step B12: Draw a full-frame electrocardiogram according to the described intraoperative electrocardiogram data sequence;

[0016] Step B13: Obtain the intraoperative intracavitary electrocardiogram by extracting the full-frame electrocardiogram according to the preset time interval and display window.

[0017] On the other hand, the risk prediction model includes:

[0018] The first convolutional layer receives a single frame of the intraoperative intracavitary electrocardiogram and extracts it to obtain a first feature map;

[0019] A max pooling layer, which is connected to the first convolutional layer, halves the data of the first feature map to obtain a second feature map;

[0020] Multiple convolutional modules connected in sequence, each convolutional module downsamples the input image of the previous stage and outputs it to the next stage;

[0021] The convolutional modules also include a skip structure, which outputs the input image from the previous stage to the convolutional module in the next stage.

[0022] An average pooling layer is connected to the last stage of the convolutional module. The average pooling layer performs average pooling on the third feature map output by the last stage of the convolutional module to obtain a fourth feature map and outputs it.

[0023] A fully connected layer predicts and outputs the risk probability based on the fourth feature map.

[0024] On the other hand, the convolution module includes:

[0025] The second convolutional layer downsamples the input image according to a preset stride to obtain a downsampled image;

[0026] The third convolutional layer is connected to the second convolutional layer. The third convolutional layer extracts features from the downsampled image and then inputs them into the next level convolutional module.

[0027] On the other hand, step S2 includes:

[0028] Step S21: Use the risk prediction model to predict the risk probability for each frame of the intraoperative intracavitary electrocardiogram;

[0029] Step S22: Calculate the average risk probability by averaging the risk probabilities;

[0030] Step S23: Map the average risk probability to the corresponding scoring interval to obtain the risk prediction result and output it.

[0031] On the other hand, prior to performing step S1, a model training process is also included, which is used to generate the risk prediction model. The model training process includes:

[0032] Step C01: Obtain the patient's electrocardiogram data and electronic medical record, generate a sample electrocardiogram based on the patient's electrocardiogram data, and extract information on recurrence after atrial fibrillation ablation from the electronic medical record;

[0033] Step C02: Add labels to the sample electrocardiograms according to the recurrence information after atrial fibrillation ablation to construct a dataset;

[0034] Step C03: Train the risk prediction model using the dataset.

[0035] On the other hand, after performing step S2, the following is also included:

[0036] Step S3: Return to step S1 to predict and display the risk prediction results in real time.

[0037] A memory storing computer instructions, wherein the prediction method described above is executed when a computer device runs the computer instructions.

[0038] The above technical solution has the following advantages or beneficial effects:

[0039] To address the issue of unsatisfactory prediction performance of existing time-series models, this solution introduces a prediction method based on intraoperative intracardiac electrocardiograms (ECGs). By collecting intraoperative ECGs of the patient at multiple time points and inputting them into a risk prediction model, the model extracts and predicts graphic-related features from the ECGs. This enables effective identification and prediction of commonly used multi-channel correlation features in the diagnostic process, facilitating more accurate predictions of postoperative recurrence risk for the patient. Attached Figure Description

[0040] Embodiments of the invention will be described more fully with reference to the accompanying drawings. However, the drawings are for illustration and explanation only and do not constitute a limitation on the scope of the invention.

[0041] Figure 1 This is an overall schematic diagram of an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of an electrocardiogram according to an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of step A1 in an embodiment of the present invention;

[0044] Figure 4 This is a schematic diagram of step B1 in an embodiment of the present invention;

[0045] Figure 5 This is a schematic diagram of the risk prediction model according to an embodiment of the present invention;

[0046] Figure 6 This is a schematic diagram of the convolution module according to an embodiment of the present invention;

[0047] Figure 7 This is a schematic diagram of step S2 in an embodiment of the present invention;

[0048] Figure 8 This is a schematic diagram of the training process according to an embodiment of the present invention;

[0049] Figure 9 This is a schematic diagram of the training results in an embodiment of the present invention;

[0050] Figure 10 This is a schematic diagram of step S3 in an embodiment of the present invention; Detailed Implementation

[0051] 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.

[0052] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0054] This invention includes:

[0055] A method for predicting the risk of recurrence after atrial fibrillation ablation, such as Figure 1 As shown, it includes:

[0056] Step S1: Collect intraoperative intracavitary electrocardiograms at multiple time points for the patient to be predicted;

[0057] Step S2: Input the intraoperative intracavitary electrocardiograms into the risk prediction model in chronological order to obtain the risk prediction results for the patient to be predicted.

[0058] Specifically, addressing the issue of poor performance of existing time-series models in the prediction process, this embodiment introduces a prediction method based on intraoperative intracardiac electrocardiograms (ICGs). By collecting ICGs of the patient under prediction at multiple time points during surgery and then inputting them into the risk prediction model, the model extracts and predicts graphic-related features from the ICGs. This enables the effective identification and prediction of multi-channel correlation features commonly used in the diagnostic process, facilitating a more accurate prediction of the postoperative recurrence risk for the patient under prediction.

[0059] In practice, the aforementioned prediction method is primarily configured as a software implementation in computer equipment, such as a doctor's workstation. This workstation receives externally input intraoperative intracardiac electrocardiograms (ECGs) via corresponding software interfaces, and predicts and outputs risk prediction results according to the aforementioned processing flow. Intraoperative intracardiac electrocardiography (ECG) refers to the process during surgery where coronary sinus electrodes are placed into the endocardium of the atrium, ventricle, or coronary sinus via a venous route. Electrical signals are recorded using an externally connected multi-electrophysiological recorder, and the obtained local potential maps are plotted to obtain the predicted results. Figure 2 The diagram shown illustrates multiple multi-channel configurations corresponding to different body parts, depending on the specific equipment used. The risk prediction model is a pre-trained artificial intelligence model used for extracting and classifying image features. The classification targets are typically different labels set by doctors as needed, such as high risk, medium risk, and low risk, or numerical risk probabilities.

[0060] Generally, existing technologies typically use multi-channel physiological recorders to directly input time-domain signal sequences into a model for processing. In this process, conventional time-domain processing models, such as RNNs, usually only focus on signal features appearing in a single continuous sequence. Furthermore, the processing time (seconds, minutes) is relatively short compared to the surgical procedure (hours), and features from more distant time points are easily overwritten by subsequent features. If it's necessary to memorize features from longer time points, the model depth increases dramatically. Similarly, if joint processing of multi-channel time-domain signal sequences is required, significant improvements to such models are needed, such as adding more fully connected layers for feature exchange and prediction. All these improvements lead to a surge in model complexity and make effective convergence difficult during subsequent training.

[0061] To address the aforementioned issues, this embodiment directly acquires intraoperative intracavitary electrocardiograms at regular time intervals using a multi-channel electrophysiological recorder and inputs them directly into the risk prediction model for processing. This method "compresses" the originally high-refresh-rate time-domain signal sequence into a two-dimensional image, allowing for the identification of features over a longer time period within a single prediction. Simultaneously, referring to... Figure 2 As shown in the graph, there is a clear data overlap between channels CS 9-10 and CS 7-8. By combining the signal features from other channels, a wider range of ECG features can be extracted, enabling the model to more accurately predict the risk of ventricular fibrillation recurrence. This process is relatively consistent with doctors' diagnostic habits and has strong interpretability.

[0062] In one embodiment, such as Figure 3 As shown, step S1 includes:

[0063] Step A1: During the surgical procedure, the ECG monitor acquires the current image as an intraoperative intracavitary electrocardiogram at preset time intervals;

[0064] The electrocardiogram monitor is connected to the coronary sinus electrode.

[0065] Specifically, in order to achieve better acquisition results, in this embodiment, based on the original electrocardiogram monitor, the currently displayed image is directly read from the video output interface at predetermined time intervals, that is, the real-time refreshed multi-lead electrocardiogram is used as the intraoperative intracavitary electrocardiogram output, and then sent into the model for processing and classification to obtain the predicted risk results, thus achieving better processing timeliness.

[0066] In one embodiment, such as Figure 4 As shown, step S1 includes:

[0067] Step B11: Obtain the intraoperative electrocardiogram (ECG) data sequence of the patient to be predicted;

[0068] Step B12: Draw a full-frame electrocardiogram based on the intraoperative electrocardiogram data sequence;

[0069] Step B13: Obtain the intraoperative intracavitary electrocardiogram by intercepting the full-frame electrocardiogram according to the preset time interval and display window.

[0070] Specifically, to achieve better adaptability to various situations, this embodiment selects to read the intraoperative electrocardiogram (ECG) data sequence of the patient to be predicted for offline scenarios, and then converts the intraoperative ECG data sequence into a full-frame ECG. This full-frame ECG graphically records the ECG signal images that occur during the operation. Subsequently, according to the input image size requirements of the model, a display window is pre-constructed, and then a preset time interval is used as the step size to extract the intraoperative intracavitary ECG from the full-frame ECG.

[0071] This method can be used to organize and predict data after surgery, and can also be used to uniformly transform and process historical electrocardiogram data, making it easier to expand the dataset during model training.

[0072] In one embodiment, such as Figure 5 As shown, the risk prediction model includes:

[0073] The first convolutional layer 1 receives the input single-frame intraoperative intracavitary electrocardiogram and extracts it to obtain the first feature map;

[0074] Max pooling layer 2 connects to the first convolutional layer 1. Max pooling layer 2 halves the data of the first feature map to obtain the second feature map.

[0075] Multiple convolutional modules 3 are connected in sequence. Each convolutional module 3 downsamples the input image of the previous level and outputs it to the next level.

[0076] There is also a skip structure 4 between the convolution modules 3, which also outputs the input image of the previous stage to the convolution module 3 of the next stage;

[0077] Average pooling layer 5 connects to the last convolutional module 3. Average pooling layer 5 performs average pooling on the third feature map output by the last convolutional module to obtain the fourth feature map and output it.

[0078] Fully connected layer 6 predicts and outputs the risk probability based on the fourth feature map.

[0079] Specifically, to achieve better prediction results, this embodiment uses the aforementioned risk prediction model. During processing, the first convolutional layer 1 and convolutional module 3 use corresponding strides and convolutional kernels of different sizes to downsample the input image, halving the number of channels or data, gradually converging to specific image features. Finally, average pooling is performed through the average pooling layer 5, enabling the fully connected layer 6 to predict and output the risk probability. To better preserve the original features and improve prediction accuracy, a skip structure 4 is used to directly halve the channels of the previous level's input image before inputting it into the next level's convolutional module 3. This allows the convolutional module 3 to perform prediction processing based on the image output from the previous level's convolutional module 3 and the original input image it relies on.

[0080] In one embodiment, such as Figure 6 As shown, convolution module 3 includes:

[0081] The second convolutional layer 31 downsamples the input image according to a preset stride to obtain a downsampled image;

[0082] The third convolutional layer 32 is connected to the second convolutional layer 31. The third convolutional layer 32 extracts features from the downsampled image and inputs them into the next level convolutional module 3.

[0083] Specifically, in order to achieve better image feature extraction, in this embodiment, the two input images are downsampled by the second convolutional layer 31 according to a preset stride to obtain downsampled images, and then the downsampled images are extracted and fused by the third convolutional layer 32 before being input into the next level convolutional module 3 to achieve better extraction results.

[0084] In one embodiment, such as Figure 7 As shown, step S2 includes:

[0085] Step S21: Use a risk prediction model to predict the risk probability for each frame of intraoperative intracavitary electrocardiogram;

[0086] Step S22: Average risk probability is obtained by averaging the risk probabilities;

[0087] Step S23: Map the average risk probability to the corresponding scoring interval to obtain the risk prediction result and output it.

[0088] Specifically, to improve the accuracy of the scoring, in this embodiment, after the risk prediction model predicts the risk probability for each frame of intraoperative intracavitary electrocardiogram, it also performs mean processing on the risk probability. This mean processing can be either average value processing or time-related weighted average, for example, the risk probability weight is higher the closer to the current time. This method obtains a more accurate average risk probability and avoids the impact of data fluctuations. Finally, the risk prediction result is obtained and output by mapping the average risk probability to a pre-defined scoring interval.

[0089] In one embodiment, prior to performing step S1, a model training process is further included. This model training process is used to generate a risk prediction model, such as... Figure 8 As shown, the model training process includes:

[0090] Step C01: Obtain the patient's electrocardiogram data and electronic medical record, generate a sample electrocardiogram based on the patient's electrocardiogram data, and extract information on recurrence after atrial fibrillation ablation from the electronic medical record;

[0091] Step C02: Add labels to the sample electrocardiograms according to the recurrence information after atrial fibrillation ablation to construct a dataset;

[0092] Step C03: Train the risk prediction model using the dataset.

[0093] Specifically, to achieve better model training results, in this embodiment, a training set is constructed using the aforementioned method before prediction begins. The patient's electrocardiogram (ECG) data refers to historical ECG images or images obtained by converting signal sequences using the above method, which have the same number of channels and arrangement as those in the actual examination. Furthermore, information on atrial fibrillation recurrence after ablation within a certain period, determined through electronic medical records or manual annotation, is used as labels corresponding to the sample ECGs. A corresponding dataset can be constructed using the above methods, which includes a training set, a test set, and a validation set according to the appropriate training method, and can be used for model training.

[0094] The training results of one set of models are as follows Figure 9 As shown.

[0095] In one embodiment, such as Figure 10 As shown, after executing step S2, the following steps are also included:

[0096] Step S3: Return to step S1 to predict and display the risk prediction results in real time.

[0097] Specifically, to achieve better results, this embodiment also selects to continuously acquire images output by the electrocardiogram monitor during the operation, and process them together with previously cached historical images or historical risk probabilities to predict the current risk in real time and present them to the doctor, which helps to guide the doctor in the current surgical risk during the operation.

[0098] A memory that stores computer instructions, which, when executed by a computer device, perform the aforementioned prediction method.

[0099] Those skilled in the art will understand that various aspects, or possible implementations of various aspects, of the present invention can be embodied as systems, methods, or computer program products. Therefore, various aspects, or possible implementations of various aspects, of the present invention can take the form of entirely hardware embodiments, entirely software embodiments (including firmware, resident software, etc.), or embodiments combining software and hardware aspects, all collectively referred to herein as "circuit," "module," or "system." Furthermore, various aspects, or possible implementations of various aspects, of the present invention can take the form of computer program products, which are computer instructions stored in memory.

[0100] The memory can be a computer-readable signal medium or a computer-readable storage medium. Computer-readable storage media include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or apparatuses, or any suitable combination thereof, such as random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, and portable read-only memory (CD-ROM).

[0101] A processor in a computer reads computer instructions stored in memory, enabling the processor to execute the functional actions specified in each step or combination of steps in a flowchart; and to generate means for implementing the functional actions specified in each block or combination of blocks in a flowchart.

[0102] It should be understood that a processor in a computer can be understood as one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components used to execute the aforementioned computer instructions.

[0103] Computer instructions may be executed entirely on the user's local computer, partially on the user's local computer, as a separate software package, partially on the user's local computer and partially on a remote computer, or entirely on a remote computer or server. It should also be noted that in some alternative implementations, the functions indicated by the steps in the flowchart or the blocks in the block diagram may not occur in the order shown in the diagram. For example, depending on the functions involved, two consecutive steps or blocks may actually be executed approximately simultaneously, or these blocks may sometimes be executed in reverse order.

[0104] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the risk of recurrence after atrial fibrillation ablation, characterized in that, include: Step S1: Collect intraoperative intracavitary electrocardiograms at multiple time points for the patient to be predicted; Step S2: Input the intraoperative intracavitary electrocardiograms into the risk prediction model in chronological order to obtain the risk prediction results for the patient to be predicted.

2. The prediction method according to claim 1, characterized in that, Step S1 includes: Step A1: During the surgical procedure, the current image is captured by the electrocardiogram monitor at preset time intervals as the intraoperative intracavitary electrocardiogram; The electrocardiogram monitor is connected to a coronary sinus electrode.

3. The prediction method according to claim 1, characterized in that, Step S1 includes: Step B11: Obtain the intraoperative electrocardiogram data sequence of the patient to be predicted; Step B12: Draw a full-frame electrocardiogram according to the described intraoperative electrocardiogram data sequence; Step B13: Obtain the intraoperative intracavitary electrocardiogram by extracting the full-frame electrocardiogram according to the preset time interval and display window.

4. The prediction method according to claim 1, characterized in that, The risk prediction model includes: The first convolutional layer receives a single frame of the intraoperative intracavitary electrocardiogram and extracts it to obtain a first feature map; A max pooling layer, which is connected to the first convolutional layer, halves the data of the first feature map to obtain a second feature map; Multiple convolutional modules connected in sequence, each convolutional module downsamples the input image of the previous stage and outputs it to the next stage; The convolutional modules also include a skip structure, which outputs the input image from the previous stage to the convolutional module in the next stage. An average pooling layer is connected to the last stage of the convolutional module. The average pooling layer performs average pooling on the third feature map output by the last stage of the convolutional module to obtain a fourth feature map and outputs it. A fully connected layer predicts and outputs the risk probability based on the fourth feature map.

5. The prediction method according to claim 4, characterized in that, The convolutional module includes: The second convolutional layer downsamples the input image according to a preset stride to obtain a downsampled image; The third convolutional layer is connected to the second convolutional layer. The third convolutional layer extracts features from the downsampled image and then inputs them into the next level convolutional module.

6. The prediction method according to claim 4, characterized in that, Step S2 includes: Step S21: Use the risk prediction model to predict the risk probability for each frame of the intraoperative intracavitary electrocardiogram; Step S22: Calculate the average risk probability by averaging the risk probabilities; Step S23: Map the average risk probability to the corresponding scoring interval to obtain the risk prediction result and output it.

7. The prediction method according to claim 1, characterized in that, Before performing step S1, a model training process is also included, which is used to generate the risk prediction model. The model training process includes: Step C01: Obtain the patient's electrocardiogram data and electronic medical record, generate a sample electrocardiogram based on the patient's electrocardiogram data, and extract information on recurrence after atrial fibrillation ablation from the electronic medical record; Step C02: Add labels to the sample electrocardiograms according to the recurrence information after atrial fibrillation ablation to construct a dataset; Step C03: Train the risk prediction model using the dataset.

8. The prediction method according to claim 1, characterized in that, After performing step S2, the method further includes: Step S3: Return to step S1 to predict and display the risk prediction results in real time.

9. A memory storing computer instructions, characterized in that, When the computer device executes the computer instructions, it performs the prediction method as described in any one of claims 1-8.

Images