Processing method, device and processing equipment of QRS complex anomaly classification model

Abstract

The application provides a processing method and device of a QRS complex anomaly classification model and processing equipment. The QRS complex anomaly classification model is constructed by introducing a QRS complex anomaly classification mechanism based on 4 leads or 8 leads. Experimental results show that the detection accuracy of different types of abnormal QRS complexes is significantly improved in the 8-lead format. In addition, due to the simplification of data, especially in the 4-lead format, the application cost in terms of data processing time, data application time and data storage time is obviously improved, and the personnel workload in clinical work is also significantly reduced, so the application value is better.

Description

Technical Field

[0001] This application relates to the field of medical technology, specifically to methods, apparatus, and equipment for processing QRS complex anomaly classification models. Background Technology

[0002] The QRS complex in an electrocardiogram (ECG) signal represents ventricular depolarization and is a key component of cardiac electrical activity. Specifically, the QRS complex reflects changes in the depolarization potential and time of the left and right ventricles. The first downward wave is the Q wave, the upward wave is the R wave, and the following downward wave is the S wave. It is closely related to the occurrence of heart failure, sudden death, and structural heart disease. Therefore, accurate detection of ECG signals with abnormal QRS complex morphology is crucial for appropriate treatment, while misinterpretation may lead to adverse outcomes.

[0003] Currently, ECG signal detection based on artificial intelligence (AI) has been extensively studied, mainly focusing on ECGs with abnormal QRS complex morphology. However, the inventors of this application have found that there is basically no research focused on the refined detection of widespread QRS complex abnormalities.

[0004] Furthermore, diagnostic guidelines for ECG abnormalities rely on 12-lead ECGs, and related algorithms are also developed in the form of 12-lead ECGs. While this provides comprehensive information on ECG changes, it also significantly increases the time required for data processing, application, and storage, resulting in higher application costs.

[0005] In other words, existing detection schemes for ECG signals with abnormal QRS complex morphology have limited precision and high application costs. Summary of the Invention

[0006] This application provides a method, apparatus, and processing equipment for a QRS complex anomaly classification model. By introducing a QRS complex anomaly classification mechanism based on 4-lead or 8-lead, the constructed QRS complex anomaly classification model has been experimentally demonstrated to significantly improve the detection accuracy of different types of abnormal QRS complexes in the 8-lead format. In addition, due to data simplification, especially in the 4-lead format, significant improvements have been made in several aspects of application costs, such as data processing time, data application time, and data storage time. It also significantly reduces the workload of clinical staff, thus having good application value.

[0007] Firstly, this application provides a method for processing a QRS group anomaly classification model, the method comprising:

[0008] Acquire sample ECG signals of different types of QRS complexes. The sample ECG signals include ECG signals of normal QRS complexes and ECG signals of abnormal QRS complexes. Specifically, the sample ECG signals are ECG signals of a preset lead format, which is either a 4-lead format or an 8-lead format.

[0009] Configure corresponding annotations for the sample ECG signals to indicate the corresponding QRS group type;

[0010] Based on the labeled sample ECG signals, a QRS group anomaly classification model is trained. The QRS group anomaly classification model is a deep learning model architecture. The QRS group anomaly classification model is used to perform corresponding QRS group anomaly classification processing based on the ECG signals input to the model with a preset lead format.

[0011] Secondly, this application provides a processing apparatus for a QRS group anomaly classification model, the apparatus comprising:

[0012] The acquisition unit is used to acquire sample ECG signals of different types of QRS complexes. The sample ECG signals include ECG signals of normal QRS complexes and ECG signals of abnormal QRS complexes. Specifically, the sample ECG signals are ECG signals of preset lead formats, which are either 4-lead or 8-lead formats.

[0013] The annotation unit is used to configure corresponding annotations for the sample ECG signal to annotate the corresponding QRS group type;

[0014] The training unit is used to train a QRS cluster anomaly classification model based on labeled sample ECG signals. The QRS cluster anomaly classification model is a deep learning model architecture. The QRS cluster anomaly classification model is used to perform corresponding QRS cluster anomaly classification processing based on the ECG signal input to the model with a preset lead format.

[0015] Thirdly, this application provides a processing device, including a processor and a memory, wherein a computer program is stored in the memory, and when the processor invokes the computer program in the memory, it executes the method provided by the first aspect of this application or any possible implementation of the first aspect of this application.

[0016] Fourthly, this application provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to perform the method provided in the first aspect of this application or any possible implementation thereof.

[0017] From the above, it can be concluded that this application has the following beneficial effects:

[0018] For the detection of ECG signals with abnormal QRS complex morphology, this application introduces a QRS complex abnormality classification mechanism based on 4-lead or 8-lead. The QRS complex abnormality classification model constructed in this way has been experimentally demonstrated to significantly improve the detection accuracy of different types of abnormal QRS complexes in the 8-lead format. In addition, due to the simplification of data, especially in the 4-lead format, the application costs in terms of data processing time, data application time, and data storage time have also been significantly improved. At the same time, it has also significantly reduced the workload of clinical staff, thus having good application value. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a processing method for the QRS group anomaly classification model of this application.

[0021] Figure 2 This is a schematic diagram of a scenario involving the model architecture of this application;

[0022] Figure 3 This is a schematic diagram illustrating an example of the ROC curve evaluation metric in this application.

[0023] Figure 4 This is a schematic diagram of an example of the confusion matrix evaluation index in this application;

[0024] Figure 5 This is a schematic diagram of a processing device for the QRS group anomaly classification model of this application;

[0025] Figure 6 This is a schematic diagram of one type of processing equipment used in this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. The naming or numbering of steps appearing in this application does not imply that the steps in the method flow must be performed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical purpose, as long as the same or similar technical effect is achieved.

[0028] The module division described in this application is a logical division. In practical applications, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the coupling or direct coupling or communication connection between modules shown or discussed may be through some interfaces, and the indirect coupling or communication connection between modules may be electrical or other similar forms, none of which are limited in this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed in multiple circuit modules. Some or all of the modules may be selected to achieve the purpose of the solution in this application according to actual needs.

[0029] Before introducing the processing method of the QRS group anomaly classification model provided in this application, we will first introduce the background content involved in this application.

[0030] The processing method, apparatus, and computer-readable storage medium for the QRS complex anomaly classification model provided in this application can be applied to processing equipment. By introducing a QRS complex anomaly classification mechanism based on 4-lead or 8-lead, the constructed QRS complex anomaly classification model has been experimentally demonstrated to significantly improve the detection accuracy of different types of abnormal QRS complexes in the 8-lead format. In addition, due to the simplification of data, especially in the 4-lead format, the application costs in terms of data processing time, data application time, and data storage time have also been significantly improved. At the same time, it has also significantly reduced the workload of clinical staff, thus having good application value.

[0031] The processing method for the QRS anomaly classification model mentioned in this application can be executed by a processing device for the QRS anomaly classification model, or by different types of processing devices such as servers, physical hosts, or user equipment (UE) that integrate the processing device for the QRS anomaly classification model. The processing device for the QRS anomaly classification model can be implemented in hardware or software. The UE can specifically be a smartphone, tablet, laptop, desktop computer, or personal digital assistant (PDA) or other terminal device. The processing devices can be configured in a device cluster.

[0032] It is understood that the processing device that executes the processing method of the QRS group anomaly classification model of this application, or the processing device that carries the processing application service corresponding to the QRS group anomaly classification model of this application, only needs to have the corresponding data processing capabilities, considering that the focus of the solution of this application is on the configuration or training of the model.

[0033] If the processing equipment is still needed to be involved in the actual application of the model, it needs to be combined with clinical applications. It can be deployed on equipment that is specifically responsible for diagnosis based on ECG signals or QRS complex anomaly analysis, or it can be deployed on equipment that is also responsible for other tasks (such as other clinical diagnoses or even routine office work).

[0034] Of course, the processing equipment can also be divided into a first part responsible for model training and a second part responsible for model application, depending on the training and application stages of the model. For example, the backend server can train the model and then send the model to the front end (the doctor's side) through a remote service for specific application, which is also possible.

[0035] Therefore, the specific type of processing equipment and the form of equipment deployment can be flexibly adjusted according to actual needs, and this application does not impose specific limitations.

[0036] The following section introduces the processing method of the QRS group anomaly classification model provided in this application.

[0037] First, refer to Figure 1 , Figure 1 The diagram illustrates a processing method for the QRS group anomaly classification model of this application. The processing method for the QRS group anomaly classification model provided in this application may specifically include the following steps S101 to S103:

[0038] Step S101: Obtain sample ECG signals of different types of QRS complexes. The sample ECG signals include ECG signals of normal QRS complexes and ECG signals of abnormal QRS complexes. Specifically, the sample ECG signals are ECG signals of a preset lead format, which is either a 4-lead format or an 8-lead format.

[0039] It is understandable that when configuring / training the model, it is necessary to configure the corresponding training samples. For the QRS group anomaly classification model to be constructed in this application, the training samples are specifically sample ECG signals (or sample ECG signal data).

[0040] The sample ECG signal includes two types of ECG signals: ECG signals with normal QRS complexes and ECG signals with abnormal QRS complexes. The ECG signals with abnormal QRS complexes can be further classified into different types of abnormal QRS complexes.

[0041] The sample ECG signal can be a historical ECG signal collected clinically, or an ECG signal obtained by secondary processing based on a historical ECG signal collected clinically. It can also be an ECG signal generated manually or automatically under a preset generation strategy. All of these are possible in practice to meet the application needs in different situations.

[0042] Furthermore, if the sample ECG signal involves real patient data, it is understandable that, considering patient privacy, anonymization settings can be used in the specific operation, retaining only the core data content of the signal.

[0043] It is worth noting that in existing solutions, diagnostic guidelines for ECG abnormalities rely on 12-lead ECGs, and the relevant algorithms are also based on 12-lead ECGs. However, this application focuses on 4-lead and 8-lead ECGs, corresponding to sample ECG signals in 4-lead or 8-lead formats.

[0044] Under the overall objective of detecting different types of abnormal ECG signals, including abnormal QRS complexes, although the existing 12-lead approach seems to provide comprehensive ECG change information, this application found that it does not significantly contribute to the detection of ECG signals with abnormal QRS complexes. In fact, it increases the time and workload of clinical staff. This, coupled with the increased time for data processing, application, and storage, leads to higher application costs, which in practice affects the application value of the ECG signal detection approach for abnormal QRS complexes.

[0045] As an exemplary embodiment, the leads involved in the 4-lead format of this application may specifically include:

[0046] Leads I, II, V1, and V5.

[0047] On the other hand, as an exemplary embodiment, the leads involved in the 8-lead format of this application may specifically include:

[0048] Leads I, II, V1, V2, V3, V4, V5, and V6.

[0049] It is understood that these leads are included in the scope of 12 leads. Therefore, the 4-lead or 8-lead involved in this application can be understood as a simplification of the existing 12-lead. Considering that each lead itself is not the focus of improvement in this application and belongs to the prior art, the focus of this embodiment is to provide a specific application scheme that can involve 4 or 8 lead types, and no further detailed description will be given.

[0050] Furthermore, regarding the signal characteristics of the ECG signal itself, as an exemplary embodiment, the sample ECG signal here and the ECG signal input to the subsequent model have a signal length of 10 seconds and a sampling frequency of 500Hz.

[0051] It is understood that the embodiments described here are specific signal formats of the ECG signals involved in the scheme of this application, and specific implementation schemes are given.

[0052] Step S102: Configure corresponding annotations for the sample ECG signals to annotate the corresponding QRS group types;

[0053] It is understandable that after obtaining the initial training samples, there is also the process of labeling. The purpose of labeling the training samples is to provide the model with the true prediction values, so as to provide a reference for the theoretical prediction values ​​in the subsequent specific model training process, and guide the model to optimize the model parameters based on the prediction values ​​and the loss function. Considering that this belongs to the scope of conventional model training, we will not go into specific details here.

[0054] In practice, annotation can be performed manually or by using automated annotation tools with configured automatic annotation logic; both methods are acceptable.

[0055] The annotation results, or labels, involve QRS group types, including normal QRS group types and different abnormal QRS group types.

[0056] As an exemplary embodiment, the QRS group types involved in this application may specifically include normal type, LBBB type, RBBB type, AVPR type, WPW-A type and WPW-B type.

[0057] Among them, LBBB type is left bundle branch block, RBBB type is right bundle branch block, AVPR type is artificial ventricular pacing rhythm (APVR) type, WPW-A type is Wolff-Parkinson-White syndrome type-A (WPW-A) type, and WPW-B type is Wolff-Parkinson-White syndrome type-B (WPW-B) type.

[0058] These abnormal QRS complex types all present as wide QRS complexes. For the WPW-A type, an abnormal atrioventricular connection exists, leading to left ventricular pre-excitation and electrical vectors towards the precordial leads. Therefore, the WPW-A type exhibits positive triangular waves and QRS waves in all precordial leads, features often confused with the RBBB type. Similarly, for the WPW-B type, an abnormal atrioventricular connection between the right atrium and right ventricle leads to right ventricular pre-excitation. Therefore, the WPW-B type presents negative QRS waves in lead V1 and positive triangular waves and QRS waves in leads V4-V6, often misdiagnosed as the LBBB type. The AVPR type also exhibits a similar wide waveform to the LBBB type, further increasing the difficulty of diagnosing these ECG abnormalities, and exacerbating the shortcomings of the existing diagnostic methods mentioned earlier.

[0059] In this case, the output of the subsequent QRS group anomaly classification model can be the probability value (which can be represented by P) of the ECG signal input to the model belonging to the above six QRS group types.

[0060] Step S103: Based on the labeled sample ECG signals, train a QRS group anomaly classification model. The QRS group anomaly classification model is a deep learning model architecture. The QRS group anomaly classification model is used to perform corresponding QRS group anomaly classification processing based on the ECG signals input to the model in a preset lead format.

[0061] After obtaining the labeled ECG signals of the samples, it is clear that the specific model training process can then proceed. The model training process typically involves the following steps:

[0062] In each round of model training, a sample ECG signal is input into the model, enabling it to perform QRS cluster anomaly classification. This determines the QRS cluster type to which the input sample ECG signal belongs or is compatible with, achieving forward propagation. Then, based on the QRS cluster anomaly classification results output by the model, the loss function is calculated in conjunction with the corresponding annotation results. The model parameters are then optimized based on the loss function calculation results, achieving backpropagation. In this way, when the preset model training requirements such as training time, number of training sessions, or detection accuracy are met, the model training is completed, and a QRS cluster anomaly classification model that can be put into practical use is obtained.

[0063] It is understandable that during model training, the specific model architecture, training architecture, and loss function used can be flexibly configured according to actual needs. Existing solutions can be used, optimized based on existing solutions, or novel self-developed solutions can be adopted. All of these are possible in practice.

[0064] As an exemplary embodiment, this application provides a specific implementation scheme for the specific model architecture of the QRS wavegroup anomaly classification model.

[0065] Specifically, the QRS wavegroup anomaly classification model in this application can be built on the ResNet model architecture.

[0066] ResNet is a network that combines the advantages of Convolutional Neural Networks (CNNs) and solves the gradient vanishing problem as the number of network layers increases. By introducing shortcut connections, it reduces the optimization difficulty, can deepen the number of network layers, and also has better network learning performance.

[0067] refer to Figure 2The diagram illustrates one scenario of the model architecture involved in this application. The QRS wavegroup anomaly classification model, based on the ResNet model architecture, includes 37 layers with 15 residual blocks in the middle. Each residual block consists of two convolutional layers connected via shortcuts. The entire network can still be trained through end-to-end backpropagation. Here, the residual block is called the Residual block, and the convolutional layer is called Conv2D. Furthermore, each convolutional layer is preceded by a Batch Normalization (BN) layer and a ReLU layer. The BN layer normalizes the input to help improve the stability of the model during training, while the ReLU layer introduces non-linearity, enabling the model to learn complex patterns. The BN layer is a batch normalization layer, and the ReLU layer is a non-linear activation layer (Rectified Linear Unit).

[0068] In addition, the max pooling layer compresses the feature map processed by the previous layer to reduce the computational burden. The max pooling layer is also known as Global max polling.

[0069] In addition, the global average pooling layer generates the final features and reduces the risk of overfitting by significantly reducing the number of parameters. The global average pooling layer is also known as Global average pooling.

[0070] In addition, the last layer is the softmax layer, which provides the posterior probability for each class through the softmax function.

[0071] Furthermore, in specific operations, other aspects of model configuration can be involved. For example, the model configuration framework can be implemented based on TensorFlow and run on an NVIDIA Tesla V100 graphics card. In the CNN structure, the kernel size of the main convolution is fixed at (1,5), the pooling size of the max pooling layer is fixed at 2, and the pooling stride is fixed at 2. An effective optimization method called mom can be used to achieve efficient computation. A weighted loss function called cross-entropy loss is used during training. The model hyperparameters are set to {100, 0.00008, 500}, representing the batch size, learning rate, and number of training epochs, respectively.

[0072] Meanwhile, to better understand the improvements in detection performance brought about by the QRS group anomaly classification model constructed in the above-mentioned scheme of this application, the following examples can also be used to aid understanding.

[0073] As an example, when evaluating six abnormalities, this application compares the 4-lead (I,II,V1 and V5), 8-lead (I,II,V1-V6) formats with the 12-lead (I,II,III,aVR,aVL,aVF,V1-V6) formats.

[0074] The 8-lead format showed the highest average F1 score of 0.969 (0.943–0.997) and average AUC score of 0.997 (0.975–1.00); in comparison, the 4-lead format showed an average F1 score of 0.952 (0.931–0.969) and an AUC score of 0.996 (0.976–1.00); and the 12-lead format showed an average F1 score of 0.95 (0.927–0.97) and an AUC score of 0.994 (0.971–1.00).

[0075] Meanwhile, the 8-lead format showed the highest mean sensitivity of 0.969 (0.94-0.995) and mean specificity score of 0.994 (0.988-1.00); in comparison, the 4-lead format had a mean sensitivity of 0.952 (0.932-0.964) and mean specificity score of 0.99 (0.971-1.00); and the 12-lead format had a mean sensitivity of 0.95 (0.930-0.965) and specificity of 0.99 (0.968-1.00).

[0076] As can be seen, compared with existing methods that rely on 12-lead ECG, the QRS complex anomaly classification model constructed based on 8 leads in this application has better detection performance in detail. While the QRS complex anomaly classification model constructed based on 4 leads does not show significant differences in improving detection accuracy, it can greatly simplify the amount of data involved in the ECG signal input signal. This can significantly reduce application costs in terms of data processing time, data application time, and data storage time, and can also significantly reduce the workload of clinical staff, which is also very beneficial.

[0077] Furthermore, focusing on the 8-lead format, as another example, the 8-lead QRS complex anomaly classification model achieved an accuracy of 456 out of 480 cases (95.0%) on the test dataset, outperforming the physician assessment accuracy of 432 out of 480 cases (89.9%). The average F1 score of the QRS complex anomaly classification model for the six abnormalities in the 8-lead format was 0.969 (0.943–0.997), exceeding the physician's average score of 0.898 (0.863–0.932).

[0078] Refer to Table 1 below for a set of examples of the evaluation indicators for this application:

[0079] Table 1 - Examples of Evaluation Indicators

[0080]

[0081] The QRS complex anomaly classification model achieved F1 scores higher than the average scores of the six physicians across all six anomaly types. Specifically, the QRS complex anomaly classification model achieved F1 scores of 0.983 (0.958-0.999) for the LBBB type, 0.982 (0.955-0.997) for the RBBB type, 0.981 (0.958-0.997) for the AVPR type, 0.950 (0.91-0.989) for the WPW-A type, and 0.951 (0.913-0.989) for the WPW-B type, demonstrating performance comparable to or even superior to that of physicians.

[0082] The QRS complex anomaly classification model had an average AUC score of 0.997 (0.975-1.00) for these six categories, which was higher than the doctors' AUC score of 0.941 (0.918-0.963). Furthermore, the model's AUC score for each anomaly was higher than the average score of the six doctors.

[0083] The QRS complex anomaly classification model showed an average sensitivity of 0.969 (0.94-0.995) and a specificity of 0.994 (0.988-1.00), which is comparable to the sensitivity of 0.902 (0.858-0.946) and the specificity of 0.980 (0.97-0.99) assessed by physicians.

[0084] See Figure 3 The diagram illustrates an example of the ROC curve evaluation metric used in this application (individual physician performance is represented by dots, and average physician performance by asterisks). Regarding predictive sensitivity, the model performs better when comparing the ROC curves of the QRS complex anomaly classification model with those of six physicians. See also... Figure 4 The diagram shows an example of the confusion matrix evaluation index of this application. The confusion matrix of the QRS complex anomaly classification model shows a distribution trend similar to that of physician performance. These results indicate that the QRS complex anomaly classification model is significantly better than physician assessment in most of the six categories of diagnosis.

[0085] After the model training / configuration is completed, it is easy to understand that the proposed solution can also involve the application of the QRS wavegroup anomaly classification model.

[0086] In this regard, as an exemplary embodiment, after training a QRS group anomaly classification model based on the labeled sample ECG signals, the method of this application may further include:

[0087] Acquire the target ECG signal;

[0088] Input the target ECG signal into the QRS group anomaly classification model;

[0089] Extract the QRS group anomaly classification results corresponding to the target ECG signal from the output of the QRS group anomaly classification model.

[0090] The target ECG signal corresponds to the sample ECG signal, which can be either a 4-lead or 8-lead ECG signal. The QRS complex anomaly classification result can output either the most probable QRS complex classification result with its corresponding probability value, or the probability value for each QRS complex anomaly classification result. For example, the model can be configured to output 6 values, showing the probability of the signal belonging to each of the 6 categories, arranged in descending order to clearly and concisely represent the determined category (corresponding to the highest probability value).

[0091] Furthermore, after obtaining QRS complex anomaly results, further data applications can be carried out according to clinical needs, such as result display, result saving, result forwarding, output completion prompts, or corresponding data analysis and processing. Specific data applications can be configured according to actual needs, and this application does not impose specific limitations.

[0092] In conclusion, regarding the above solutions, this application introduces a QRS complex anomaly classification mechanism based on 4-lead or 8-lead models for detecting ECG signals with abnormal QRS complex morphology. Experiments have demonstrated that the constructed QRS complex anomaly classification model significantly improves the detection accuracy for different types of abnormal QRS complexes in the 8-lead format. Furthermore, due to data simplification, especially in the 4-lead format, significant improvements have been made in application costs such as data processing time, data application time, and data storage time. This also significantly reduces the workload for clinical staff, thus demonstrating considerable application value.

[0093] The above is an introduction to the processing method of the QRS group anomaly classification model provided in this application. In order to facilitate better implementation of the processing method of the QRS group anomaly classification model provided in this application, this application also provides a processing device for the QRS group anomaly classification model from the perspective of functional modules.

[0094] See Figure 5 , Figure 5 This is a schematic diagram of a processing device for the QRS group anomaly classification model of this application. In this application, the processing device 500 for the QRS group anomaly classification model may specifically include the following structure:

[0095] The acquisition unit 501 is used to acquire sample ECG signals of different types of QRS complexes. The sample ECG signals include ECG signals of normal QRS complexes and ECG signals of abnormal QRS complexes. Specifically, the sample ECG signals are ECG signals of preset lead formats, which are either 4-lead or 8-lead formats.

[0096] The annotation unit 502 is used to configure corresponding annotations for the sample ECG signal to annotate the corresponding QRS group type;

[0097] Training unit 503 is used to train a QRS cluster anomaly classification model based on labeled sample ECG signals. The QRS cluster anomaly classification model is a deep learning model architecture. The QRS cluster anomaly classification model is used to perform corresponding QRS cluster anomaly classification processing based on the ECG signal input to the model with a preset lead format.

[0098] In one exemplary embodiment, the leads involved in the 4-lead format specifically include:

[0099] Leads I, II, V1, and V5.

[0100] In yet another exemplary embodiment, the leads involved in the 8-lead format specifically include:

[0101] Leads I, II, V1, V2, V3, V4, V5, and V6.

[0102] In yet another exemplary embodiment, the QRS group type specifically includes normal type, LBBB type, RBBB type, AVPR type, WPW-A type and WPW-B type;

[0103] The output of the QRS group anomaly classification model is the probability value of the input ECG signal belonging to one of the six QRS group types.

[0104] In another exemplary embodiment, the QRS wavegroup anomaly classification model, based on the ResNet model architecture, includes 37 layers with 15 residual blocks in the middle, each residual block consisting of two convolutional layers connected by shortcuts.

[0105] Each convolutional layer is preceded by a BN layer and a ReLU layer. The BN layer normalizes the input, and the ReLU layer introduces non-linearity.

[0106] The feature map processed by the previous layer is compressed by the max pooling layer;

[0107] The global average pooling layer generates the final features;

[0108] The last layer is the softmax layer, which provides the posterior probability for each class through the softmax function.

[0109] In yet another exemplary embodiment, the sample ECG signal and the ECG signal input to the model have a signal length of 10 seconds and a sampling frequency of 500Hz.

[0110] In yet another exemplary embodiment, the apparatus further includes an application unit 504 for:

[0111] Acquire the target ECG signal;

[0112] Input the target ECG signal into the QRS group anomaly classification model;

[0113] Extract the QRS group anomaly classification results corresponding to the target ECG signal from the output of the QRS group anomaly classification model.

[0114] This application also provides a processing device from a hardware architecture perspective, see [link / reference]. Figure 6 , Figure 6 This diagram illustrates a structural schematic of the processing device of this application. Specifically, the processing device may include a processor 601, a memory 602, and an input / output device 603. The processor 601 executes the computer program stored in the memory 602 to implement, for example... Figure 1 The corresponding steps of the processing method for the QRS group anomaly classification model in the embodiment; or, when the processor 601 executes the computer program stored in the memory 602, it implements as follows: Figure 5 Corresponding to the functions of each unit in the embodiment, the memory 602 is used to store the functions executed by the processor 601 as described above. Figure 1 The computer program required for the processing method of the QRS group anomaly classification model in the corresponding embodiment.

[0115] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 602 and executed by processor 601 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a computer device.

[0116] The processing device may include, but is not limited to, processor 601, memory 602, and input / output device 603. Those skilled in the art will understand that the illustrations are merely examples of the processing device and do not constitute a limitation on the processing device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the processing device may also include network access devices, buses, etc., and processor 601, memory 602, input / output device 603, etc., are connected via a bus.

[0117] Processor 601 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the processing device, connecting various parts of the device through various interfaces and lines.

[0118] The memory 602 can be used to store computer programs and / or modules. The processor 601 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 602 and by calling data stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function, etc.; the data storage area may store data created according to the use of the processing device, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0119] When processor 601 executes a computer program stored in memory 602, it can specifically perform the following functions:

[0120] Acquire sample ECG signals of different types of QRS complexes. The sample ECG signals include ECG signals of normal QRS complexes and ECG signals of abnormal QRS complexes. Specifically, the sample ECG signals are ECG signals of a preset lead format, which is either a 4-lead format or an 8-lead format.

[0121] Configure corresponding annotations for the sample ECG signals to indicate the corresponding QRS group type;

[0122] Based on the labeled sample ECG signals, a QRS group anomaly classification model is trained. The QRS group anomaly classification model is a deep learning model architecture. The QRS group anomaly classification model is used to perform corresponding QRS group anomaly classification processing based on the ECG signals input to the model with a preset lead format.

[0123] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the processing device, processing equipment, and its corresponding units of the QRS group anomaly classification model described above can be found in the following reference: Figure 1 The processing method of the QRS group anomaly classification model in the corresponding embodiment will not be described in detail here.

[0124] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0125] Therefore, this application provides a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute the present application. Figure 1 The steps of the processing method for the QRS group anomaly classification model in the corresponding embodiment can be found in the following example. Figure 1 The processing method of the QRS group anomaly classification model in the corresponding embodiment will not be repeated here.

[0126] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0127] Because of the instructions stored in the computer-readable storage medium, the present application can be executed as described above. Figure 1 The steps of the processing method for the QRS group anomaly classification model in the corresponding embodiment can therefore achieve the results of this application. Figure 1 The beneficial effects that the processing method of the QRS group anomaly classification model in the corresponding embodiment can achieve are detailed in the preceding description and will not be repeated here.

[0128] The processing method, apparatus, processing device, and computer-readable storage medium of the QRS group anomaly classification model provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for processing a QRS group anomaly classification model, characterized in that, The method includes: Acquire sample ECG signals of different types of QRS complexes, wherein the sample ECG signals include ECG signals of normal QRS complexes and ECG signals of abnormal QRS complexes, and the sample ECG signals are specifically ECG signals of a preset lead format, which is specifically a 4-lead format or an 8-lead format. Configure corresponding labels for the sample ECG signals to label the corresponding QRS group types; Based on the labeled sample ECG signals, a QRS group anomaly classification model is trained. The QRS group anomaly classification model is a deep learning model architecture. The QRS group anomaly classification model is used to perform corresponding QRS group anomaly classification processing based on the ECG signals input to the model in the preset lead format. The QRS wavegroup anomaly classification model is based on the ResNet model architecture and includes 37 layers with 15 residual blocks in the middle. Each residual block is composed of two Conv2D convolutional layers connected by shortcuts. Each of the convolutional layers is preceded by a BN layer and a ReLU layer. The BN layer normalizes the input, and the ReLU layer introduces non-linearity. The feature map processed by the previous layer is compressed by the max pooling layer; The global average pooling layer generates the final features; The last layer is the softmax layer, which provides the posterior probability for each class through the softmax function.

2. The method according to claim 1, characterized in that, The leads involved in the 4-lead format specifically include: Leads I, II, V1, and V5.

3. The method according to claim 1, characterized in that, The 8-lead format specifically includes the following leads: Leads I, II, V1, V2, V3, V4, V5, and V6.

4. The method according to claim 1, characterized in that, The QRS group types specifically include normal type, LBBB type, RBBB type, AVPR type, WPW-A type, and WPW-B type; The output of the QRS group anomaly classification model is the probability value of the ECG signal input to the model belonging to one of the six QRS group types.

5. The method according to claim 1, characterized in that, The sample ECG signal and the ECG signal input to the model have a signal length of 10 seconds and a sampling frequency of 500Hz.

6. The method according to claim 1, characterized in that, After training the QRS group anomaly classification model based on the labeled sample ECG signals, the method further includes: Acquire the target ECG signal; The target ECG signal is input into the QRS group anomaly classification model; Extract the QRS group anomaly classification results corresponding to the target ECG signal from the output of the QRS group anomaly classification model.

7. A processing device for a QRS group anomaly classification model, characterized in that, The device includes: The acquisition unit is used to acquire sample ECG signals of different types of QRS complexes, wherein the sample ECG signals include ECG signals of normal QRS complexes and ECG signals of abnormal QRS complexes, and the sample ECG signals are specifically ECG signals of a preset lead format, which is specifically a 4-lead format or an 8-lead format. The annotation unit is used to configure corresponding annotations for the sample ECG signal to annotate the corresponding QRS group type; The training unit is used to train a QRS cluster anomaly classification model based on the labeled sample ECG signals. The QRS cluster anomaly classification model is a deep learning model architecture. The QRS cluster anomaly classification model is used to perform corresponding QRS cluster anomaly classification processing based on the ECG signals input to the model in the preset lead format. The QRS wavegroup anomaly classification model is based on the ResNet model architecture and includes 37 layers with 15 residual blocks in the middle. Each residual block is composed of two Conv2D convolutional layers connected by shortcuts. Each of the convolutional layers is preceded by a BN layer and a ReLU layer. The BN layer normalizes the input, and the ReLU layer introduces non-linearity. The feature map processed by the previous layer is compressed by the max pooling layer; The global average pooling layer generates the final features; The last layer is the softmax layer, which provides the posterior probability for each class through the softmax function.

8. A processing device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the method as described in any one of claims 1 to 6 when it invokes the computer program in the memory.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the method of any one of claims 1 to 6.

Images