Paroxysmal atrial fibrillation signal identification method and device, computer equipment and storage medium
By using multi-path convolutional neural networks and multi-granularity fusion technology, an atrial fibrillation signal recognition model is constructed, which solves the problem of insufficient reliability and accuracy in the recognition of paroxysmal atrial fibrillation signals in existing technologies. It achieves efficient and comprehensive recognition of atrial fibrillation signals and supports real-time monitoring by wearable devices.
Patent Information
- Application Number
- CN202410474603.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-19
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2044-04-19
AI Technical Summary
Existing technologies, especially those using wearable single-lead electrocardiogram (ECG) devices, struggle to meet clinical diagnostic requirements in terms of reliability and accuracy when identifying paroxysmal atrial fibrillation signals, and also require sophisticated data processing.
A multi-path convolutional neural network is used, with each path having a receptive field corresponding to atrial fibrillation signals of various granularities. Through feature extraction and multi-granularity fusion, an atrial fibrillation signal recognition model is constructed, including a residual connection module, a multi-granularity fusion module, and a temporal feature extraction module, thereby improving recognition accuracy.
It achieves efficient and comprehensive identification of paroxysmal atrial fibrillation signals, improves the reliability and accuracy of identification, can identify atrial fibrillation signals of different granularities, outputs identification results of low-medium-high risk levels, and supports real-time monitoring by wearable devices.
Smart Images

Figure CN118277831B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the technical field of signal processing, and in particular to a method, apparatus, computer device and storage medium for identifying paroxysmal atrial fibrillation signals. Background Technology
[0002] An electrocardiogram (ECG) is a tool that records changes in the electrical activity of the heart during a heartbeat, playing a crucial role in examining and assessing a patient's cardiac condition. Typically, this requires patients to travel to a hospital for evaluation by a trained and experienced cardiologist, which is both costly and inconvenient. Therefore, automated monitoring and diagnostic tools are particularly important in clinics, community health centers, and family health programs. Although ECG technology has made significant progress in signal filtering, detection, and classification over the past few decades, accurate classification of ECG signals, especially those from paroxysmal atrial fibrillation, remains challenging due to noise interference and variability in patient symptoms.
[0003] In the field of ECG signal classification for atrial fibrillation, traditional methods mainly consist of two core steps: feature extraction and classification model training. Feature extraction in this process is primarily performed in the time or frequency domain, focusing on features such as signal amplitude, time intervals, and advanced statistical properties. These feature extraction techniques typically encompass filter bank techniques, Kalman filtering, principal component analysis, frequency analysis methods, wavelet transform, and statistical methods. For classification techniques, common methods include support vector machines, artificial neural networks, hidden Markov models, and hybrid expert systems. Artificial neural networks, in particular, have become the foundation of many studies due to their excellent model-building capabilities. For example, M. Engin designed a fuzzy neural network method for ECG signal classification, which utilizes autoregressive model coefficients, higher-order cumulants, and the variance of wavelet transform as features. LYShyu et al. developed a novel method combining wavelet transform and fuzzy neural networks to detect atrial premature beats, reducing computational complexity by using the same wavelet for P-wave detection and atrial premature beat classification. I. Guler and EDUbeyli recommend a fusion neural network model for classifying atrial fibrillation ECG signals. This model uses statistical features based on discrete wavelet transform as input to the first layer of the network and uses the output of this layer to train subsequent networks.
[0004] While the above methods have achieved good results, they also share some common drawbacks:
[0005] (1) Artificially generated features rely on expert knowledge and experience and require careful design and evaluation. In addition, the classifier must have appropriate model capabilities to process these features.
[0006] (2) Current methods for identifying paroxysmal atrial fibrillation signals using machine learning require long ECG signals and have high data processing requirements. For dynamic ECG signals collected by wearable single-lead devices, the reliability and accuracy of these methods are still difficult to meet the needs of clinical diagnosis. Summary of the Invention
[0007] This invention provides a method, apparatus, computer device, and storage medium for identifying paroxysmal atrial fibrillation signals, in order to address the shortcomings in the reliability and accuracy of existing technologies for identifying paroxysmal atrial fibrillation signals.
[0008] In a first aspect, embodiments of the present invention provide a method for identifying paroxysmal atrial fibrillation signals, comprising:
[0009] Acquire the electrocardiogram data to be identified;
[0010] A model for identifying atrial fibrillation signals is determined; the model includes a multi-channel convolutional neural network, each channel of which has a receptive field corresponding to atrial fibrillation signals of various granularities. The receptive field is used to extract features from the atrial fibrillation signals in the electrocardiogram data to be identified.
[0011] The electrocardiogram data to be identified is input into the atrial fibrillation signal recognition model to obtain the recognition result of the atrial fibrillation signal.
[0012] Secondly, embodiments of the present invention also provide a device for identifying paroxysmal atrial fibrillation signals, comprising:
[0013] The data acquisition module is used to acquire the electrocardiogram data to be identified.
[0014] The model determination module is used to determine the atrial fibrillation signal recognition model; the atrial fibrillation signal recognition model includes a multi-channel convolutional neural network, each channel of the convolutional neural network is set with a receptive field corresponding to atrial fibrillation signals of various granularities, and the receptive field is used to extract features from the atrial fibrillation signals in the electrocardiogram data to be identified;
[0015] The model recognition module is used to input the electrocardiogram data to be recognized into the atrial fibrillation signal recognition model to obtain the recognition result of the atrial fibrillation signal.
[0016] Thirdly, embodiments of the present invention also provide a computer device, the computer device comprising:
[0017] One or more processors;
[0018] Memory, used to store one or more programs.
[0019] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for identifying paroxysmal atrial fibrillation signals as described in any of the first aspects.
[0020] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for identifying paroxysmal atrial fibrillation signals as described in any one of the first aspects.
[0021] In this embodiment, electrocardiogram (ECG) data to be identified is acquired, and an atrial fibrillation (AF) signal recognition model is determined. The AF signal recognition model includes multiple convolutional neural networks (CNNs), each with a receptive field corresponding to AF signals of various granularities. These receptive fields are used to extract features from the AF signals in the ECG data to be identified. The ECG data is then input into the AF signal recognition model to obtain the AF signal recognition result. This method uses multiple convolutional neural networks with different receptive fields to identify AF signals in the ECG data to be identified, enabling the identification of AF signals of different granularities and providing a more comprehensive recognition of AF signals, thus improving the reliability and accuracy of AF signal recognition. Attached Figure Description
[0022] Figure 1 A flowchart illustrating a method for identifying paroxysmal atrial fibrillation signals provided in Embodiment 1 of the present invention;
[0023] Figure 2 This is a schematic diagram of the structure of a paroxysmal atrial fibrillation signal identification device provided in Embodiment 2 of the present invention;
[0024] Figure 3 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. Detailed Implementation
[0025] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.
[0026] Example 1
[0027] Figure 1This is a flowchart of a method for identifying paroxysmal atrial fibrillation signals according to Embodiment 1 of the present invention. This embodiment is applicable to the identification of atrial fibrillation signals from dynamic electrocardiogram signals acquired by a wearable single-lead device. The method can be executed by a paroxysmal atrial fibrillation signal identification device, which can be implemented by software and / or hardware and can be configured in a computer device, such as a personal laptop, desktop computer, server, industrial control computer, all-in-one computer, etc. The specific steps include the following:
[0028] S110. Obtain the electrocardiogram data to be identified.
[0029] Atrial fibrillation, also known as atrial fibrillation, is one of the most common cardiac arrhythmias and is associated with high morbidity and mortality rates from diseases such as stroke, heart failure, myocardial infarction, and cerebral thrombosis. Electrocardiography (ECG), as a non-invasive method for examining and diagnosing cardiac activity, is widely used worldwide. Effective and rapid detection of atrial fibrillation using ECG data is crucial for reducing patient disability and mortality rates.
[0030] The ECG data to be identified in this embodiment can be ECG data that has been filtered to remove various interferences (such as power line interference, baseline drift, and muscle contraction noise) from the ECG signal, or it can be ECG data directly acquired from the ECG acquisition device. To reduce the impact of noise in the ECG data on the subsequent atrial fibrillation signal identification model, in one implementation of this embodiment, the original ECG data acquired by the acquisition device can be denoised first to obtain denoised original ECG data. To enable the model to better utilize temporal feature information when processing the ECG data, the ECG signal in the denoised original ECG data can be transformed from a time-domain signal to a frequency-domain signal to obtain the final ECG data to be identified. In another implementation of this embodiment, signal modeling and filtering can be combined to denoise the ECG data to be identified. For example, a method combining wavelet transform and neural networks can be used to eliminate different types of noise in the ECG data to be identified.
[0031] S120. Determine the atrial fibrillation signal recognition model.
[0032] The atrial fibrillation signal recognition model in this embodiment includes a multi-channel convolutional neural network (CNN). Each CNN has a receptive field corresponding to atrial fibrillation signals of various granularities. The receptive field is used to extract features of atrial fibrillation signals from the ECG data to be identified. The receptive field refers to the size of the region on the input image mapped back to a pixel in the feature map output by each layer of the CNN; that is, the size of a pixel in the feature map output by the CNN during convolution operations relative to the original image, and also the region of the input image that the CNN features can see. Multi-granularity refers to analyzing and processing data or signals at different spatial or temporal scales. This embodiment uses a multi-channel CNN, with each CNN having a receptive field of various granularities corresponding to atrial fibrillation signals. This improves the image representation capability of ECG data, covers more types of data resolution when processing the ECG data to be identified, and further enhances the comprehensiveness of feature extraction and the diversity of feature information.
[0033] In this embodiment, the specific method for determining the atrial fibrillation signal identification model includes:
[0034] S1201. Obtain unlabeled electrocardiogram data as training data. The training data includes electrocardiogram signals containing atrial fibrillation signals.
[0035] S1202. Label the ECG signals in the training data with atrial fibrillation signals of various granularities to obtain the labeled training data.
[0036] In one specific implementation, S1202 may include: using a pre-trained signal labeling model to label atrial fibrillation signals of various granularities in the electrocardiogram (ECG) signals in the training data to obtain labeled training data; the signal labeling model includes a coarse-grained feature selection module and a fine-grained feature selection module, the coarse-grained feature selection module is used to extract the temporal context information of the ECG signals, and the fine-grained feature selection module is used to extract the context information related to the atrial fibrillation signals from the ECG signals and extract the position information of the atrial fibrillation signals in the ECG signals.
[0037] In another specific implementation, S1202 may include: filtering electrocardiogram (ECG) signals containing atrial fibrillation signals from the training data; using a sliding window algorithm to extract signal segments from the ECG signals containing atrial fibrillation signals every first preset time period to obtain multiple signal segments; and labeling the multiple signal segments with atrial fibrillation signals of various granularities to obtain labeled training data. The specific details of "using the sliding window algorithm to extract signal segments from the ECG signal containing atrial fibrillation signals every first preset time interval to obtain multiple signal segments" can include: for the ECG signal containing atrial fibrillation signals, marking the signal occurrence point of each atrial fibrillation event; for the ECG signal containing atrial fibrillation signals, starting from the signal occurrence point of each atrial fibrillation event, using a sliding window to slide the ECG signal forward for a first preset time interval to extract a signal segment, obtaining multiple signal segments; for multiple signal segments, locating the end sampling point of each signal segment according to a preset reference heart rate; for multiple signal segments, collecting all heart rate data for a second preset time interval backward from the end sampling point of each signal segment, as the heart rate dataset corresponding to the current signal segment; for multiple signal segments, determining whether there is more than 50% abnormal data in the heart rate dataset of each signal segment, if so, removing the current signal segment, if not, retaining the current signal segment.
[0038] S1203. Construct a multi-path convolutional neural network as the basic model. Each path of the convolutional neural network is set with a receptive field corresponding to a certain granularity of atrial fibrillation signal.
[0039] It should be noted that convolutional neural networks (CNNs) can extract complex features from image data and effectively represent images through layer-by-layer convolution and pooling operations. To improve the reliability and accuracy of identifying paroxysmal atrial fibrillation (AF) signals, this application designs a multi-path CNN as the basic model for AF signal recognition. Each CNN has a receptive field corresponding to a specific granularity of AF signal, which enhances the model's ability to process large-scale, high-resolution image data and overcomes the drawbacks of large-scale and high-resolution image data, such as high computational cost, numerous parameters, and long training time. Furthermore, the multi-path CNN in this application can use multiple "convolution-activation-pooling" or "convolution-pooling" layers to extract feature information from the input image data hierarchically. This allows the model to extract features from shallow to deep layers and from local to global perspectives, resulting in more diverse and richer features, enabling better understanding and recognition of the data information in the input image data.
[0040] S1204. Based on the basic model, construct a residual connection module, a multi-granularity fusion module, a temporal feature extraction module, and a classifier to obtain the atrial fibrillation signal recognition model to be trained.
[0041] In this embodiment, building a residual connection module on top of the basic model helps overcome the problems of gradient vanishing and gradient exploding in ultra-deep convolutional layer networks. Building a multi-granularity fusion module facilitates feature fusion of multiple ECG signals containing atrial fibrillation signals of different granularities, improving the model's recognition performance and generalization ability. Building a temporal feature extraction module further extracts effective signal data, allowing the model to classify only the effective signal data before outputting classification results, significantly reducing computational load and improving computational efficiency. In other examples, residual neural networks (ResNet) and long short-term memory networks (LSTM) can be used as prototypes, with the addition of a classifier to construct the atrial fibrillation signal recognition model to be trained.
[0042] S1205. Use the labeled training data to train the atrial fibrillation signal recognition model until the preset convergence condition is met. Then the atrial fibrillation signal recognition model is considered to have been trained and the trained atrial fibrillation signal recognition model is obtained.
[0043] In a specific implementation, S1205 may include the following detailed steps:
[0044] S12051. The labeled training data is input into the multi-path convolutional neural network in the atrial fibrillation signal recognition model to be trained, so that the multi-path convolutional neural network performs convolution and pooling operations on the labeled training data to obtain the first feature map.
[0045] Convolution is used to learn the local features in the training data, while pooling is used to ensure the translation invariance of these local features.
[0046] S12052. Input the first feature map into the residual connection module in the atrial fibrillation signal recognition model to be trained to obtain the second feature map;
[0047] S12053. Use the multi-granularity fusion module in the atrial fibrillation signal recognition model to be trained to fuse the first feature map and the second feature map to obtain fused feature data.
[0048] S12054. Input the fused feature data into the time-series feature extraction module in the atrial fibrillation signal recognition model to be trained to obtain time-series feature data;
[0049] S12055. Input the time-series feature data into the classifier in the atrial fibrillation signal recognition model to be trained, and obtain the recognition output result;
[0050] S12056. Determine whether the recognition output result meets the preset convergence condition;
[0051] S12057. If the convergence condition is met, the atrial fibrillation signal recognition model to be trained is determined to be completed, and the atrial fibrillation signal recognition model after training is obtained.
[0052] S12058. If the convergence condition is not met, continue executing S12051.
[0053] It should be noted that before continuing to execute S12051, the loss value of the recognition output in S12055 needs to be calculated first. The loss value is then used to backpropagate the atrial fibrillation signal recognition model to be trained, and the network parameters (such as weight parameters, bias parameters, etc.) of the model in the previous round of forward propagation training are updated. After all the network parameters in the model have been updated, S12051 is executed with the updated network parameters.
[0054] S130. Input the ECG data to be identified into the atrial fibrillation signal recognition model to obtain the recognition result of the atrial fibrillation signal.
[0055] In this embodiment, the identification results of atrial fibrillation signals can be presented in the form of three risk levels: "low-medium-high," so that users can pay timely attention to and receive medical intervention for paroxysmal atrial fibrillation.
[0056] In this embodiment, an atrial fibrillation (AF) signal recognition model is determined by acquiring the electrocardiogram (ECG) data to be identified. The AF signal recognition model includes multiple convolutional neural networks (CNNs), each with a receptive field corresponding to AF signals of various granularities. These receptive fields are used to extract features from the AF signals in the ECG data. The ECG data is then input into the AF signal recognition model to obtain the AF signal recognition result. This method uses multiple convolutional neural networks with different receptive fields to identify AF signals in the ECG data, enabling the identification of AF signals at different granularities and providing a more comprehensive recognition of AF signals, thus improving the reliability and accuracy of AF signal recognition.
[0057] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.
[0058] Example 2
[0059] Figure 2 This is a structural block diagram of a paroxysmal atrial fibrillation signal identification device provided in Embodiment 2 of the present invention, which may specifically include the following modules:
[0060] The data acquisition module 210 is used to acquire electrocardiogram data to be identified;
[0061] The model determination module 220 is used to determine the atrial fibrillation signal recognition model; the atrial fibrillation signal recognition model includes a multi-channel convolutional neural network, each channel of the convolutional neural network is set with a receptive field corresponding to atrial fibrillation signals of various granularities, and the receptive field is used to extract features from the atrial fibrillation signals in the electrocardiogram data to be identified;
[0062] The model recognition module 230 is used to input the electrocardiogram data to be recognized into the atrial fibrillation signal recognition model to obtain the recognition result of the atrial fibrillation signal.
[0063] In one embodiment of the present invention, the data acquisition module 210 to be identified includes:
[0064] The raw data acquisition submodule is used to acquire the raw electrocardiogram data to be identified;
[0065] The noise reduction submodule is used to denoise the original ECG data to be identified, so as to obtain the original ECG data to be identified after noise reduction.
[0066] The signal transformation submodule is used to transform the ECG signal in the denoised original ECG data to be identified from a time domain signal to a frequency domain signal, so as to obtain the ECG data to be identified.
[0067] In one embodiment of the present invention, the model determination module 220 includes:
[0068] The training data acquisition submodule is used to acquire unlabeled electrocardiogram data as training data, including electrocardiogram signals containing atrial fibrillation signals.
[0069] The training data labeling submodule is used to label the electrocardiogram signals in the training data with atrial fibrillation signals of various granularities to obtain the labeled training data;
[0070] The basic model construction submodule is used to build a multi-path convolutional neural network as the basic model. Each convolutional neural network is set with a receptive field corresponding to a certain granularity of atrial fibrillation signal.
[0071] The sub-module for constructing the atrial fibrillation signal recognition model to be trained is used to build a residual connection module, a multi-granularity fusion module, a temporal feature extraction module, and a classifier on the basis of the basic model to obtain the atrial fibrillation signal recognition model to be trained.
[0072] The training submodule is used to train the atrial fibrillation signal recognition model using the labeled training data until a preset convergence condition is met, at which point the training of the atrial fibrillation signal recognition model is determined to be complete, and the trained atrial fibrillation signal recognition model is obtained.
[0073] In one embodiment of the present invention, the training submodule includes:
[0074] The first feature map acquisition unit is used to input the labeled training data into the multi-path convolutional neural network in the atrial fibrillation signal recognition model to be trained, so that the multi-path convolutional neural network performs convolution and pooling operations on the labeled training data to obtain the first feature map.
[0075] The second feature map acquisition unit is used to input the first feature map into the residual connection module in the atrial fibrillation signal recognition model to be trained, so as to obtain the second feature map.
[0076] The fusion feature data acquisition unit is used to perform feature fusion on the first feature map and the second feature map using the multi-granularity fusion module in the atrial fibrillation signal recognition model to be trained, so as to obtain fusion feature data;
[0077] The temporal feature extraction unit is used to input the fused feature data into the temporal feature extraction module in the atrial fibrillation signal recognition model to be trained, so as to obtain temporal feature data;
[0078] The identification output result acquisition unit is used to input the time series feature data into the classifier in the atrial fibrillation signal identification model to be trained, and obtain the identification output result;
[0079] A convergence judgment unit is used to determine whether the recognition output result meets the preset convergence condition;
[0080] The training completion determination unit is used to determine that the atrial fibrillation signal recognition model to be trained has been trained if the convergence condition is met, and to obtain the atrial fibrillation signal recognition model after training.
[0081] The training unit is configured to continue executing the process of inputting the labeled training data into the atrial fibrillation signal recognition model if the convergence condition is not met, so that the multi-path convolutional neural network performs convolution and pooling operations on the labeled training data to obtain a first feature map.
[0082] In one embodiment of the present invention, the training data labeling submodule includes:
[0083] The model labeling unit is used to label atrial fibrillation signals of various granularities in the electrocardiogram (ECG) signals in the training data using a pre-trained signal labeling model, thereby obtaining labeled training data. The signal labeling model includes a coarse-grained feature selection module and a fine-grained feature selection module. The coarse-grained feature selection module is used to extract the temporal context information of the ECG signals, and the fine-grained feature selection module is used to extract context information related to atrial fibrillation signals from the ECG signals and extract the position information of atrial fibrillation signals in the ECG signals.
[0084] In one embodiment of the present invention, the training data labeling submodule includes:
[0085] A signal filtering unit is used to filter electrocardiogram signals that contain atrial fibrillation signals in the training data;
[0086] The signal segment acquisition unit is used to extract signal segments from the electrocardiogram signal containing atrial fibrillation signal every first preset time period using a sliding window algorithm, thereby obtaining multiple signal segments.
[0087] The signal labeling unit is used to label the multiple signal segments with atrial fibrillation signals of various granularities to obtain labeled training data.
[0088] In one embodiment of the present invention, the signal segment acquisition unit includes:
[0089] The signal occurrence point marking subunit is used to mark the signal occurrence point of each atrial fibrillation event for the electrocardiogram signal containing the atrial fibrillation signal;
[0090] The signal interception subunit is used to intercept a signal segment from the signal occurrence point of each atrial fibrillation event of the electrocardiogram signal containing atrial fibrillation signal by using a sliding window to slide the electrocardiogram signal forward for a first preset time period, thereby obtaining multiple signal segments.
[0091] The end sampling point marker subunit is used to locate the end sampling point of each signal segment according to a preset reference heart rate for the multiple signal segments;
[0092] The heart rate data collection subunit is used to collect all heart rate data for the multiple signal segments, starting from the end sampling point of each signal segment and moving forward for a second preset time period, as the heart rate dataset corresponding to the current signal segment.
[0093] The abnormal data removal subunit is used to determine whether there are more than 50% abnormal data in the heart rate dataset of each of the multiple signal segments. If there are, the current signal segment is removed; if not, the current signal segment is retained.
[0094] The paroxysmal atrial fibrillation signal identification device provided in this embodiment of the invention can execute the paroxysmal atrial fibrillation signal identification method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.
[0095] Example 3
[0096] Figure 3 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. Figure 3 A block diagram of an exemplary computer device 12 suitable for implementing embodiments of the present invention is shown. Figure 3 The computer device 12 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.
[0097] like Figure 3 As shown, the computer device 12 is represented in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and a bus 18 connecting different system components (including system memory 28 and processing unit 16).
[0098] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MA3) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (P3I) bus.
[0099] Computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device 12, including volatile and non-volatile media, removable and non-removable media.
[0100] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 3 Not shown; usually referred to as a "hard drive"). Although Figure 3Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disc drive for reading and writing to a removable non-volatile optical disc (e.g., a 3D-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.
[0101] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.
[0102] Computer device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with computer device 12, and / or with any device that enables computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 22. Furthermore, computer device 12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 20. As shown, network adapter 20 communicates with other modules of computer device 12 via bus 18. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with computer device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0103] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the paroxysmal atrial fibrillation signal identification method provided in the embodiments of the present invention.
[0104] Example 4
[0105] Embodiment 4 of the present invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements each process of the above-described method for identifying paroxysmal atrial fibrillation signals and achieves the same technical effect. To avoid repetition, it will not be described again here.
[0106] Computer-readable storage media may include, for example, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.
[0107] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.
Claims
1. A method for identifying paroxysmal atrial fibrillation signals, characterized in that, include: Acquire the electrocardiogram data to be identified; Determine the atrial fibrillation signal recognition model; The electrocardiogram data to be identified is input into the atrial fibrillation signal recognition model to obtain the recognition result of the atrial fibrillation signal; The atrial fibrillation signal identification model includes: Unlabeled electrocardiogram (ECG) data is obtained as training data, including ECG signals containing atrial fibrillation signals. The electrocardiogram signals in the training data are labeled with atrial fibrillation signals of various granularities to obtain the labeled training data; A multi-path convolutional neural network is constructed as the basic model, and each path of the convolutional neural network is set with a receptive field corresponding to a certain granularity of atrial fibrillation signal; Based on the basic model, a residual connection module, a multi-granularity fusion module, a temporal feature extraction module, and a classifier are constructed to obtain the atrial fibrillation signal recognition model to be trained. The atrial fibrillation signal recognition model is trained using the labeled training data until a preset convergence condition is met. Then the atrial fibrillation signal recognition model is determined to be trained and the trained atrial fibrillation signal recognition model is obtained. The process of training the atrial fibrillation signal recognition model using the labeled training data until a preset convergence condition is met, and then determining that the atrial fibrillation signal recognition model has been trained successfully, and obtaining the trained atrial fibrillation signal recognition model, includes: The labeled training data is input into the multi-path convolutional neural network in the atrial fibrillation signal recognition model to be trained, so that the multi-path convolutional neural network performs convolution and pooling operations on the labeled training data to obtain the first feature map. The first feature map is input into the residual connection module of the atrial fibrillation signal recognition model to be trained to obtain the second feature map; The first feature map and the second feature map are fused using the multi-granularity fusion module in the atrial fibrillation signal recognition model to be trained, to obtain fused feature data; The fused feature data is input into the time-series feature extraction module in the atrial fibrillation signal recognition model to be trained to obtain time-series feature data; The time-series feature data is input into the classifier in the atrial fibrillation signal recognition model to be trained to obtain the recognition output result; Determine whether the recognition output result meets the preset convergence condition; If the convergence condition is met, the atrial fibrillation signal recognition model to be trained is determined to be trained and the atrial fibrillation signal recognition model after training is obtained. If the convergence condition is not met, the process of inputting the labeled training data into the atrial fibrillation signal recognition model to be trained continues, so that the multi-path convolutional neural network performs convolution and pooling operations on the labeled training data to obtain the first feature map.
2. The method according to claim 1, characterized in that, The acquisition of the electrocardiogram data to be identified includes: Obtain the raw electrocardiogram data to be identified; The original ECG data to be identified is denoised to obtain the denoised original ECG data to be identified. The ECG signal in the original ECG data to be identified after denoising is transformed from a time domain signal to a frequency domain signal to obtain the ECG data to be identified.
3. The method according to claim 1, characterized in that, The process of labeling the electrocardiogram signals in the training data with atrial fibrillation signals of various granularities to obtain labeled training data includes: A pre-trained signal labeling model is used to label atrial fibrillation signals at multiple granularities in the electrocardiogram (ECG) signals in the training data, resulting in labeled training data. The signal labeling model includes a coarse-grained feature selection module and a fine-grained feature selection module. The coarse-grained feature selection module is used to extract the temporal context information of the ECG signals, and the fine-grained feature selection module is used to extract context information related to atrial fibrillation signals from the ECG signals and extract the position information of atrial fibrillation signals in the ECG signals.
4. The method according to claim 1, characterized in that, The process of labeling the electrocardiogram signals in the training data with atrial fibrillation signals of various granularities to obtain labeled training data includes: Filter the electrocardiogram signals in the training data that contain atrial fibrillation signals; The sliding window algorithm is used to extract signal segments from the electrocardiogram signal containing atrial fibrillation signal every first preset time period to obtain multiple signal segments; The multiple signal segments are labeled with atrial fibrillation signals of various granularities to obtain labeled training data.
5. The method according to claim 4, characterized in that, The method involves using a sliding window algorithm to extract signal segments from the electrocardiogram signal containing atrial fibrillation signals at first preset time intervals, resulting in multiple signal segments, including: For the electrocardiogram signals containing atrial fibrillation signals, mark the signal occurrence point for each atrial fibrillation event; For the electrocardiogram (ECG) signal containing atrial fibrillation signals, starting from the signal occurrence point of each atrial fibrillation event, a signal segment is extracted by sliding a window forward for a first preset time period to obtain multiple signal segments; For the multiple signal segments, the end sampling point of each signal segment is located according to a preset reference heart rate; For the multiple signal segments, collect all heart rate data over a second preset time period backward from the end sampling point of each signal segment, and use this as the heart rate dataset corresponding to the current signal segment; For the multiple signal segments, determine whether there are more than 50% abnormal data in the heart rate dataset of each signal segment. If there are, remove the current signal segment; if not, retain the current signal segment.
6. A device for identifying paroxysmal atrial fibrillation signals, characterized in that, include: The data acquisition module is used to acquire the electrocardiogram data to be identified. The model determination module is used to determine the atrial fibrillation signal identification model; The model recognition module is used to input the electrocardiogram data to be recognized into the atrial fibrillation signal recognition model to obtain the recognition result of the atrial fibrillation signal; The model determination module includes: The training data acquisition submodule is used to acquire unlabeled electrocardiogram data as training data, including electrocardiogram signals containing atrial fibrillation signals. The training data labeling submodule is used to label the electrocardiogram signals in the training data with atrial fibrillation signals of various granularities to obtain the labeled training data; The basic model construction submodule is used to build a multi-path convolutional neural network as the basic model. Each convolutional neural network is set with a receptive field corresponding to a certain granularity of atrial fibrillation signal. The sub-module for constructing the atrial fibrillation signal recognition model to be trained is used to build a residual connection module, a multi-granularity fusion module, a temporal feature extraction module, and a classifier on the basis of the basic model to obtain the atrial fibrillation signal recognition model to be trained. The training submodule is used to train the atrial fibrillation signal recognition model under training using the labeled training data until a preset convergence condition is met, at which point the atrial fibrillation signal recognition model under training is determined to be trained and the trained atrial fibrillation signal recognition model is obtained. The training submodule includes: The first feature map acquisition unit is used to input the labeled training data into the multi-path convolutional neural network in the atrial fibrillation signal recognition model to be trained, so that the multi-path convolutional neural network performs convolution and pooling operations on the labeled training data to obtain the first feature map. The second feature map acquisition unit is used to input the first feature map into the residual connection module in the atrial fibrillation signal recognition model to be trained, so as to obtain the second feature map. The fusion feature data acquisition unit is used to perform feature fusion on the first feature map and the second feature map using the multi-granularity fusion module in the atrial fibrillation signal recognition model to be trained, so as to obtain fusion feature data; The temporal feature extraction unit is used to input the fused feature data into the temporal feature extraction module in the atrial fibrillation signal recognition model to be trained, so as to obtain temporal feature data; The identification output result acquisition unit is used to input the time series feature data into the classifier in the atrial fibrillation signal identification model to be trained, and obtain the identification output result; A convergence judgment unit is used to determine whether the recognition output result meets the preset convergence condition; The training completion determination unit is used to determine that the atrial fibrillation signal recognition model to be trained has been trained if the convergence condition is met, and to obtain the atrial fibrillation signal recognition model after training. The training unit is configured to continue executing the process of inputting the labeled training data into the atrial fibrillation signal recognition model if the convergence condition is not met, so that the multi-path convolutional neural network performs convolution and pooling operations on the labeled training data to obtain a first feature map.
7. A computer device, characterized in that, The computer device includes: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method for identifying paroxysmal atrial fibrillation signals as described in any one of claims 1-5.
8. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the method for identifying paroxysmal atrial fibrillation signals as described in any one of claims 1-5.
Citation Information
Patent Citations
Atrial fibrillation signal recognition method, device and equipment
CN108113666A
Electrocardiosignal deep neural network modeling method based on pyramid-shaped convolutional layer
CN111261283A
