Electrocardiogram data processing model, device and equipment

Abstract

The invention provides an electrocardiogram data processing model, an electrocardiogram data processing device and electrocardiogram data processing equipment, which are used for designing a corresponding specific model processing architecture on the basis of a deep learning model on the basis of paying attention to time domain features contained in electrocardiogram data and frequency domain features contained in the electrocardiogram data aiming at a right ventricular hypertrophy / enlargement pathological state. In this way, detail features and deep features of the right ventricular hypertrophy / enlargement pathological state in the electrocardiogram data can be more accurately captured, the detection effect with the accuracy obviously improved is achieved, powerful auxiliary data support is provided, and high-quality medical service is promoted.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular to an electrocardiogram data processing model, device and processing equipment. Background Art

[0002] In clinical practice, right ventricular hypertrophy / dilation (RVH / RVD) is a cardiac pathological condition characterized by an increase in the volume and mass of the right ventricle of the heart. This condition is usually caused by chronic pressure overload and may be a secondary manifestation of a variety of lung or heart diseases (such as pulmonary hypertension, chronic obstructive pulmonary disease or congenital heart disease). Therefore, the accurate diagnosis of right ventricular hypertrophy / dilation is important.

[0003] However, there are many challenges in diagnosing right ventricular hypertrophy / enlargement by electrocardiogram (ECG). Specifically, the anterior position of the right ventricle and its smaller muscle mass relative to the left ventricle make it more difficult to detect hypertrophic / enlargement changes on the ECG. In addition, the presence of comorbidities (such as left ventricular hypertrophy or bundle branch block) may also mask the typical ECG manifestations associated with right ventricular hypertrophy / enlargement, complicating the diagnostic process.

[0004] Nevertheless, ECG remains a widely used initial diagnostic tool in the diagnosis of RV hypertrophy / enlargement in practical situations due to its easy availability and noninvasiveness.

[0005] Meanwhile, echocardiography is another important means of detecting right ventricular hypertrophy / dilation, which can provide detailed information on right ventricular size, wall thickness, and function. However, echocardiography is time-consuming, and manual judgment using echocardiography has limited diagnostic efficacy in patients with mild symptoms.

[0006] It can be seen that, for right ventricular hypertrophy / dilation, the existing detection schemes have the major problem of poor detection accuracy. Summary of the invention

[0007] The present application provides an electrocardiogram data processing model, device and processing equipment for the pathological state of right ventricular hypertrophy / enlargement. On the basis of paying attention to the time domain features contained in the electrocardiogram data, it also pays attention to the frequency domain features contained therein, and designs a corresponding specific model processing architecture based on the deep learning model. In this way, the detailed features and deep-level features of the pathological state of right ventricular hypertrophy / enlargement in the electrocardiogram data can be more accurately captured, and a detection effect with significantly improved accuracy can be achieved, thereby providing strong auxiliary data support and promoting high-quality medical services.

[0008] In a first aspect, the present application provides an electrocardiogram data processing method, the method comprising:

[0009] Obtaining the electrocardiogram data to be processed;

[0010] Input the electrocardiogram data to be processed into a preset electrocardiogram data processing model for processing, wherein the electrocardiogram data processing model is specifically a deep learning model, the electrocardiogram data processing model is pre-trained by the labeled sample electrocardiogram data, the electrocardiogram data processing model is used to predict the cardiac pathological state of the patient corresponding to the electrocardiogram data input into the model, the electrocardiogram data processing model includes a first model part, a second model part and a third model part, the first model part is used to extract the time domain features of the electrocardiogram data input into the model, the second model part is used to extract the frequency domain features of the electrocardiogram data input into the model, and the third model part is used to generate the model processing results based on the time domain features and the frequency domain features;

[0011] extracting the cardiac pathological state determination result output by the electrocardiogram data processing model and corresponding to the electrocardiogram data to be processed;

[0012] The results of cardiac pathological status determination are displayed through a visual interface.

[0013] In a second aspect, the present application provides an electrocardiogram data processing device, the device comprising:

[0014] An acquisition unit, used for acquiring electrocardiogram data to be processed;

[0015] A processing unit, used for inputting the to-be-processed electrocardiogram data into a preset electrocardiogram data processing model for processing, wherein the electrocardiogram data processing model is specifically a deep learning model, the electrocardiogram data processing model is pre-trained by the labeled sample electrocardiogram data, the electrocardiogram data processing model is used to predict the cardiac pathological state of the patient corresponding to the electrocardiogram data input into the model, the electrocardiogram data processing model includes a first model part, a second model part and a third model part, the first model part is used to extract the time domain features of the electrocardiogram data input into the model, the second model part is used to extract the frequency domain features of the electrocardiogram data input into the model, and the third model part is used to generate the model processing results based on the time domain features and the frequency domain features;

[0016] An extraction unit, used for extracting the cardiac pathological state determination result output by the electrocardiogram data processing model and corresponding to the electrocardiogram data to be processed;

[0017] The display unit is used to display the results of the determination of the cardiac pathological state through a visual interface.

[0018] In a third aspect, the present application provides a processing device, including a processor and a memory, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the method provided in the first aspect of the present application or any possible implementation method of the first aspect of the present application is executed.

[0019] In a fourth aspect, the present application provides a computer-readable storage medium, which stores multiple instructions, and the instructions are suitable for a processor to load to execute the method provided by the first aspect of the present application or any possible implementation of the first aspect of the present application.

[0020] From the above content, it can be concluded that the present application has the following beneficial effects:

[0021] Under the right ventricular hypertrophy / enlargement detection goal based on deep learning, for the pathological state of right ventricular hypertrophy / enlargement, on the basis of paying attention to the time domain features contained in the electrocardiogram data, we also pay attention to the frequency domain features contained in it, and design a corresponding specific model processing architecture based on the deep learning model. In this way, we can more accurately capture the detailed features and deep-level features of the right ventricular hypertrophy / enlargement pathological state in the electrocardiogram data, and achieve a detection effect with significantly improved accuracy, thereby providing strong auxiliary data support and promoting high-quality medical services. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 A schematic diagram of a flow chart of the electrocardiogram data processing method of the present application;

[0024] Figure 2 A structural schematic diagram of the electrocardiogram data processing model of this application;

[0025] Figure 3 A schematic diagram for comparing abnormal and normal electrocardiogram data of this application;

[0026] Figure 4 A structural schematic diagram of the first model part of this application;

[0027] Figure 5 A structural schematic diagram of the second model part of this application;

[0028] Figure 6 A schematic diagram of the overall structure of the electrocardiogram data processing model of this application;

[0029] Figure 7 A schematic diagram of the structure of the electrocardiogram data processing device of the present application;

[0030] Figure 8 A schematic diagram of the structure of the processing equipment of this application. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0032] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices. The naming or numbering of steps in this application does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The process steps that have been named or numbered can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0033] The division of modules in this application is a logical division. There may be other division methods when it is implemented in actual applications. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between modules can be electrical or other similar forms, which are not limited in this application. In addition, the modules or submodules 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, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present application.

[0034] Before introducing the electrocardiogram data processing method provided by the present application, the background content involved in the present application is first introduced.

[0035] The electrocardiogram data processing method, device and computer-readable storage medium provided in the present application can be applied to processing equipment for the pathological state of right ventricular hypertrophy / enlargement. On the basis of paying attention to the time domain features contained in the electrocardiogram data, attention is also paid to the frequency domain features contained therein, and a corresponding specific model processing architecture is designed based on the deep learning model. In this way, the detailed features and deep-level features of the pathological state of right ventricular hypertrophy / enlargement in the electrocardiogram data can be more accurately captured, and a detection effect with significantly improved accuracy can be achieved, thereby providing strong auxiliary data support and promoting high-quality medical services.

[0036] The electrocardiogram data processing method mentioned in this application can be executed by an electrocardiogram data processing device, or a server, physical host or user equipment (UE) and other different types of processing devices integrated with the electrocardiogram data processing device. Among them, the electrocardiogram data processing device can be implemented in hardware or software, and the UE can be a terminal device such as a smart phone, a tablet computer, a laptop computer, a desktop computer or a personal digital assistant (PDA), and the processing device can be set in a device cluster.

[0037] It can be understood that in specific applications, the focus of the present application is to carry out corresponding ECG data processing and display the results based on a pre-configured ECG data processing model. In this case, the processing device that executes the ECG data processing method of the present application or is equipped with the corresponding application service of the ECG data processing method of the present application, on the basis of being configured with a corresponding display screen (including a touch screen), usually only needs to meet the required data processing capabilities. The specific device types and specific device deployment forms involved are relatively flexible and can be flexibly configured according to actual needs. This application does not make specific limitations.

[0038] If the training of the electrocardiogram data processing model is further involved, further adaptive adjustments can be made based on the actual situation.

[0039] As an example, the processing device may be divided into three parts, that is, the processing device may include a first processing device that performs a training task of the electrocardiogram data processing model and a second processing device that performs an application task of the electrocardiogram data processing model.

[0040] In addition, in terms of result display, the processing device itself can be configured with a display screen for result display, or the results can be displayed through an external display device or in conjunction with other devices with display screens.

[0041] Corresponding to the above example, as another example, the processing device may include a first processing device that performs a training task of the electrocardiogram data processing model, a second processing device that performs an application task of the electrocardiogram data processing model, and a third processing device that displays the processing results of the electrocardiogram data processing model.

[0042] Next, the electrocardiogram data processing method provided by this application is introduced.

[0043] First, see Figure 1 , Figure 1 A schematic flow chart of the electrocardiogram data processing method of the present application is shown. The electrocardiogram data processing method provided by the present application may specifically include the following steps S101 to S104:

[0044] Step S101, obtaining electrocardiogram data to be processed;

[0045] It can be understood that in actual applications, corresponding to the detection target of detecting right ventricular hypertrophy / enlargement based on electrocardiogram, it is necessary to obtain the electrocardiogram data of the current patient, which is recorded as the electrocardiogram data to be processed for the convenience of explanation.

[0046] It is worth adding that, in addition to being applicable to the detection of right ventricular hypertrophy / enlargement, the present application scheme can also be applied to the detection of other types of cardiac pathological conditions. For the sake of convenience of explanation, the following description will focus on right ventricular hypertrophy / enlargement as an example / target to explain the present application scheme.

[0047] Among them, for the acquisition and processing of the electrocardiogram data to be processed, both manual entry and automatic acquisition can be adopted. The device can extract the data from the local storage space or the storage space of other devices, or the device can receive data sent by other devices. It can also collect data in real time through its own or external electrocardiograph. All of these are possible, corresponding to the flexible and changeable application needs in actual situations.

[0048] In addition, in actual applications, the present application solution is usually processed in the form of work tasks, that is, electrocardiogram data processing tasks, and the task can be initiated by manual entry or automatic initiation. The device can initiate it autonomously according to the preset autonomous initiation strategy / rules, or the device can receive tasks sent by other devices. This is similar to the above and meets the diverse application needs of the solution.

[0049] Furthermore, the electrocardiogram data involved in the present application may specifically be 12-lead electrocardiogram data collected by a 12-lead electrocardiograph. Of course, in actual applications, the number of leads involved may be adjusted as needed. For example, 18-lead electrocardiogram data collected by an 18-lead electrocardiograph or 8-lead electrocardiograph collected by an 8-lead electrocardiograph may be used. In addition, lead data identified as meaningless in subsequent data processing may be further streamlined and deleted. For example, 18-lead electrocardiogram data may be streamlined into 12-lead electrocardiogram data. For example, individual lead electrocardiogram data may be deleted from the 12-lead electrocardiogram data. For the specific lead electrocardiogram data retained, the lead electrocardiogram data that can make an effective contribution is determined in the model training process for processing. This also corresponds to a situation where the deletion and neglect of irrelevant lead electrocardiogram data may also be handled autonomously by the model.

[0050] Step S102, inputting the to-be-processed electrocardiogram data into a preset electrocardiogram data processing model for processing, wherein the electrocardiogram data processing model is specifically a deep learning model, the electrocardiogram data processing model is pre-trained by the labeled sample electrocardiogram data, the electrocardiogram data processing model is used to predict the cardiac pathological state of the patient corresponding to the electrocardiogram data input into the model, the electrocardiogram data processing model includes a first model part, a second model part and a third model part, the first model part is used to extract the time domain features of the electrocardiogram data input into the model, the second model part is used to extract the frequency domain features of the electrocardiogram data input into the model, and the third model part is used to generate the model processing results based on the time domain features and the frequency domain features;

[0051] It can be seen that based on the introduction of artificial intelligence (AI) technology and its powerful computing power to process ECG data, this application focuses on the deep learning model in the machine learning model, and combines the ECG data processing logic / solution involved in this application to build a corresponding specific model architecture and obtain an ECG data processing model.

[0052] For the ECG data processing model, refer to Figure 2 A structural schematic diagram of the electrocardiogram data processing model of the present application is shown, and it can be learned that it specifically includes three major model parts, namely:

[0053] 1) The first model part is responsible for time domain feature extraction processing, specifically for extracting the time domain features of the electrocardiogram data input into the model;

[0054] 2) The second model part is responsible for frequency domain feature extraction processing, specifically for extracting frequency domain features of the electrocardiogram data input into the model;

[0055] 3) The third model part is specifically used to generate model processing results based on time domain characteristics and frequency domain characteristics.

[0056] Regarding the above model processing architecture / structure, specifically, the inventors of the present application found that the problem with the existing solutions is that when detecting right ventricular hypertrophy / enlargement, they habitually rely on the time series features contained in the periodic electrocardiogram data (corresponding to the above time domain features), but have never paid attention to or ignored the valuable frequency domain features.

[0057] Taking the Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) adopted by existing solutions as examples, these models can capture the time series characteristics of electrocardiogram data and use them to detect right ventricular hypertrophy / enlargement. However, these models do not fully capture the abnormalities of time series signals, and modeling based only on time series characteristics cannot capture the distribution of frequencies, resulting in low accuracy and instability in detection performance. This limitation affects the development of related clinical work.

[0058] The inventor of the present application believes that for patients with right ventricular hypertrophy / enlargement, the distribution of abnormal ECG data is often more extreme than that of normal ECG data, and there are often more fluctuations, and the waveform periodicity is not obvious. The time series information can be decomposed into the frequency domain, and the abnormality in the time domain can be spread to the frequency domain. For details, please refer to Figure 3 A comparative schematic diagram of abnormal and normal electrocardiogram data of the present application is shown. In view of these valuable observations, the inventors of the present application constructed a specific model architecture involved in the electrocardiogram data processing model of the present application, which can also be understood using a high-performance collaborative framework, which can simultaneously model time domain features and frequency domain features, and use the detailed and deep features of the time and frequency domains to advance deeper and more accurate detection and processing of right ventricular hypertrophy / enlargement, and further improve the detection performance of the model.

[0059] As for the three major model parts in the electrocardiogram data processing model of this application, this application also provides corresponding implementation supporting solutions from a more specific implementation aspect.

[0060] As an exemplary embodiment, refer to Figure 4A schematic structural diagram of the first model part of the present application is shown. The first model part in the electrocardiogram data processing model of the present application may specifically include a Mamba module. After the initial temporal features aggregated by the one-dimensional temporal convolutional layer from the input electrocardiogram data of the model, the Mamba module continues to perform compression and extraction of the temporal features.

[0061] In the Mamba module, the input electrocardiogram data of the model can be denoted as Imput ECG Time Series, and let V = [v 1 , v 2 ,..., v T . T is the sequence length. Corresponding to the case where the electrocardiogram data itself is a kind of time series data, the one-dimensional temporal convolutional layer can be denoted as Temporal Conv1D, which is a kind of CNN layer used to achieve the preliminary aggregation of temporal features and provide input data for the powerful, selective state space-based Mamba module. Through the state space, the historical signal information can be better compressed, and the Mamba module can learn the temporal dependence and important features and better perform the secondary extraction of the temporal features.

[0062] Among them, for the Mamba module, it is a new type of deep learning architecture that can effectively compress and memorize the historical electrocardiogram signals, learn effective sequence representations, and at the same time solve the problem of large computational overhead of the Transformer model when processing long sequence data. Its design inspiration comes from the classical state space model, combines the characteristics of the recurrent neural network (RNN) and the convolutional neural network, and can achieve near-linear computational scalability while maintaining a modeling ability comparable to that of the Transformer. Therefore, it is more suitable for being the first model part in the electrocardiogram data processing model of the present application and can well complete the task of extracting time domain features.

[0063] Specifically, the Mamba module of the present application is constructed based on the structured state space sequence model (S4), which is a new type of sequence model in deep learning, and is widely related to RNNs, CNNs and classical state space models. In S4, the update mode of the hidden state is time-invariant, which is not enough to learn the dynamic changes of the ECG time series. At the same time, traditional RNNs and CNNs may "forget" the important features and periodic patterns in long sequences, and lack the ability to learn abnormal details. In particular, the Mamba module adopts a selective state space model, adds time-varying factors to the state selection process, and updates the hidden state by using a time-varying semi-separable lower triangular matrix, which can better compress the historical signal details in the state space, thereby giving the learned ECG representation. These advantages make the Mamba module a powerful tool for learning time series representations with stronger expressive power and performing well in diagnostic tasks.

[0064] On the other hand, as an exemplary embodiment, referring to Figure 4 A structural schematic diagram of the second model part of the present application is shown. The second model part in the electrocardiogram data processing model of the present application specifically includes a GRU module. After the initial frequency domain features of the electrocardiogram data input by the model are obtained by fast Fourier transform, the GRU module continues to extract frequency domain features.

[0065] The frequency domain / frequency features initially extracted from the ECG Time Series data input by the model through Fast Fourier Transform (FFT) are recorded as FrequencyDistribution and expressed as d = [d 1 ,d 2 ,...,d F ] indicates that the overall preliminary frequency domain feature is composed of multiple sub-frequency domain features. Each sub-frequency domain feature is input into a GRU unit in the GRU module, namely, a GRU Cell. In this way, each GRU unit will perform further processing based on the output features of the previous GRU unit and the sub-frequency domain features, and the final output is obtained by iterative processing.

[0066] It is understandable that since the electrocardiogram data of right ventricular hypertrophy / enlargement are continuous and special in the frequency domain, they also have regular sequences that can automatically identify specific patterns, such as presenting a high power spectrum in certain low-frequency ranges. Therefore, this application specifically uses the GRU module to learn and identify abnormal distributions in the frequency domain.

[0067] As for the GRU module, it is a variant of RNN, designed to process continuous data with specific features. It is simplified on the basis of LSTM, introducing fewer parameters and structural complexity. It effectively solves the gradient vanishing and gradient exploding problems of traditional RNN by using a gating mechanism. It is particularly suitable for processing long-term dependent data. Its design purpose is to maintain computational efficiency while having high performance. It is suitable for continuous data representation learning. Therefore, it is suitable as the second model part in the electrocardiogram data processing model of this application, and can well complete the task of extracting frequency domain features.

[0068] Next, continue to refer to Figure 6 A structural schematic diagram of the overall electrocardiogram data processing model of the present application is shown. It can be seen that for the third model part in the electrocardiogram data processing model, it is not necessarily necessary to directly input the features output by the previous two model parts, that is, the time domain features and the frequency domain features. Further data processing may be involved in between so that the third model part can better perform data processing.

[0069] In this regard, as an exemplary embodiment, in the electrocardiogram data processing model, both the time domain features and the frequency domain features are average pooled and then spliced, and the spliced ​​result is used as the input of the third model part.

[0070] Among them, average pooling can be recorded as Mean Pooling, the time domain features after average pooling can be recorded as Time-Domain Output, and the frequency domain features after average pooling can be recorded as Frequency-Domain Output. The two are then concatenated (which can be recorded as Concatenation) to obtain a feature tensor with more comprehensive information.

[0071] It should be noted that the average pooling process, such as Figure 5 What is shown is the tail of the first and second model parts, namely the Mamba module and the GRU module.

[0072] In addition, from Figure 6It can also be seen that as an exemplary embodiment, the third model part in the electrocardiogram data processing model of the present application can be specifically a two-layer MLP module. Among them, MLP is Multilayer Perceptron. Specifically, MLP is composed of multiple layers composed of several neurons, including an input layer, a hidden layer and an output layer, wherein the input layer receives input data and passes the data to the hidden layer, and the hidden layer converts the input value into an output value through an activation function, and continues to pass it to the output layer, and the output layer gives the final prediction result. Each neuron has weights and biases in the hidden layer and the output layer, which can be regarded as a nonlinear function, which accepts input from the neurons in the previous layer, and performs a corresponding series of calculations based on the weights and biases, and finally generates an output. The neurons in the hidden layer and the output layer transmit information through connection information, and the connection has a weight to express the strength of the connection.

[0073] The basic idea of ​​the multilayer perceptron can be understood as optimizing the weights and biases of neurons through training so that the model can learn the mapping relationship between input and output. Its main advantage is that it can handle complex nonlinear relationships. It can also construct deep neural networks and improve the expression ability by increasing the depth of the hidden layer. Therefore, it is suitable as the third model part in the electrocardiogram data processing model of this application, and can well complete the detection task of right ventricular hypertrophy / enlargement based on the feature content in both time domain and frequency domain.

[0074] In addition, it should be noted that the two-layer MLP module does not mean that there are two MLP modules, but refers to a two-layer perceptron.

[0075] It can be understood that the above three exemplary embodiments all start from the specific model architecture level to provide more specific explanations of the practical implementation solutions that can be adopted in actual applications of the electrocardiogram data processing model of the present application, thus forming a complex specific model architecture with powerful detection performance of the present application, which has better practical significance and practical value.

[0076] In terms of the specific model processing results, that is, the cardiac pathological state processed by the model, as an exemplary embodiment, the cardiac pathological state specifically corresponds to at least one of right ventricular hypertrophy and right ventricular enlargement, and is divided into two states: diseased and non-diseased. That is, the model performs a binary classification detection task, in which the diseased state can be identified as 1 and the non-diseased state can be identified as 0.

[0077] Among them, for different types of cardiac pathological conditions, the specific labeling method can be flexibly adjusted according to actual conditions.

[0078] In addition, considering that the electrocardiogram data processing model of the present application has significantly improved detection performance, it is highly targeted and sensitive to the pathological state of right ventricular hypertrophy / enlargement. Therefore, when the presence or condition of right ventricular hypertrophy / enlargement is detected, the specific degree of right ventricular hypertrophy / enlargement can be analyzed and quantified within a preset degree range. For example, the value can be taken in the range of [1,2,3,...,10]. The larger the value or the higher the level, the more serious the degree of the disease. For example, it can be selected within the range of low risk, medium risk and high risk. The specific degree of disease quantification scheme / strategy can obviously be flexibly configured according to actual needs, so as to further provide more delicate auxiliary data support, help the detection and treatment of right ventricular hypertrophy / enlargement, and provide more hierarchical and valuable decision support.

[0079] At the same time, corresponding to the application of the model, the present application may also involve the early model training link. In this regard, as an exemplary embodiment, the present application method may also include:

[0080] Get sample ECG data;

[0081] Configure annotations for sample ECG data;

[0082] Based on the labeled sample ECG data, the ECG data processing model is trained.

[0083] Among them, for the description of the method for obtaining the sample electrocardiogram data, reference can be made to the description of the method for obtaining the electrocardiogram data to be processed, and this application will not make too many repetitive descriptions here.

[0084] The purpose of labeling is to provide theoretical model output results to assist in the training of the model's sensitivity and pertinence to the detection target during the model training process. The labeling process involved can be done manually or by corresponding automatic labeling tools. Among them, the automatic labeling tool needs to be pre-configured with the corresponding automatic labeling logic.

[0085] In this way, after the labeling is completed, the initial model can be trained based on the sample ECG data to obtain a trained ECG data processing model that can be put into practical use.

[0086] For a specific model cycle, it usually includes:

[0087] In each round of model training, a sample ECG data is input into the model so that the model predicts the corresponding cardiac pathological state and realizes forward propagation. Then, based on the cardiac pathological state determination result output by the model and the labeling result, the loss function is calculated. Based on the loss function calculation result, the model parameters are optimized to realize reverse propagation. In this way, when the preset model training requirements such as training time, number of trainings, and detection accuracy are met, the model training can be completed. At this time, the model is an ECG data processing model that can be put into practical use.

[0088] It should be noted that for the model training scheme and loss function involved in the model training phase, you can either use the existing scheme, or make further optimization and improvement on the basis of the existing scheme, or even use a novel self-developed scheme. These are all possible in actual situations, and you can make adaptive configurations according to specific needs.

[0089] As an example, for sample ECG data, the data set specifically configured in this application contains 17,665 records (excluding or excluding relevant content involving patient privacy), of which 11,126 are normal ECG data and 6,539 are abnormal ECG data (i.e., ECG data of patients with right ventricular hypertrophy / enlargement). After removing the noise points in the original ECG data, taking 12 leads as an example, in the 12-lead ECG data, the original data of each lead contains a digital vector with a length of 5000. Considering that the ECG data has a strong periodicity, in order to improve the detection efficiency of deep learning, the first 1,250 values ​​of each lead can be specifically intercepted, that is, 1,250 can be specifically selected as the specific sequence length of the ECG data, the normal ECG data is marked as 0, and the abnormal ECG data is marked as 1, and finally a third-order 17,665×12×1,250 time series tensor data set is obtained.

[0090] During the training and evaluation of the model, the original data was shuffled with a batch size of 128, and the Adam optimizer was used in the training process with a learning rate set to 0.001. The original dataset was divided into training, validation and test sets, all of which were completely shuffled and reorganized from the constructed dataset, with proportions of 80%, 5% and 15% respectively. In addition, the binary cross entropy (BCE) loss was used as the loss function. For the stopping criterion, the early stopping method was used, that is, the training was stopped when the validation loss failed to continue to decrease within 10 cycles. In terms of evaluation, recall, F1 score, precision, accuracy, specificity and BCE loss were used as evaluation indicators.

[0091] The evaluation shows that compared with other baseline models, the model of this application has achieved significant improvement in detection accuracy while maintaining a high detection efficiency. In this way, in practical applications, it can provide robust and reliable auxiliary functions for related clinical work.

[0092] Step S103, extracting the cardiac pathological state determination result corresponding to the electrocardiogram data to be processed output by the electrocardiogram data processing model;

[0093] After the electrocardiogram data processing model processes the current electrocardiogram data to be processed and obtains the corresponding cardiac pathological state determination result (such as the right ventricular hypertrophy / enlargement determination result mainly involved in this application), the model will output it.

[0094] Correspondingly, the pathological state determination result output by the electrocardiogram data processing model can be extracted.

[0095] Step S104, displaying the cardiac pathological status determination result through a visual interface.

[0096] It can be understood that after obtaining the results of the heart pathological state determination, it can be displayed. The visualization interface or visualization page involved can provide specific result display services through its own display screen, external display device or other device with a display screen.

[0097] In this process, the application of web services may also be involved to achieve the effect of remote display.

[0098] In addition, it can be understood that when the results of the heart pathological status determination are obtained, in addition to the result display, local storage, remote storage, result forwarding, output of completed processing prompts, or further data processing and analysis (such as analysis of treatment plans, analysis of causes, etc.) can be performed. Obviously, the specific data application content involved is relatively flexible and can be adjusted according to the pre- and real-time configured data application strategies / rules, and this application does not make specific limitations.

[0099] Finally, for the above solutions, in general, under the right ventricular hypertrophy / enlargement detection goal based on deep learning, for the pathological state of right ventricular hypertrophy / enlargement, on the basis of paying attention to the time domain features contained in the electrocardiogram data, we also pay attention to the frequency domain features contained in it, and design the corresponding specific model processing architecture based on the deep learning model. In this way, we can more accurately capture the detailed features and deep-level features of the right ventricular hypertrophy / enlargement pathological state in the electrocardiogram data, and achieve a detection effect with significantly improved accuracy, thereby providing strong auxiliary data support and promoting high-quality medical services.

[0100] The above is an introduction to the electrocardiogram data processing method provided by the present application. In order to facilitate better implementation of the electrocardiogram data processing method provided by the present application, the present application also provides an electrocardiogram data processing device from the perspective of functional modules.

[0101] See also Figure 7 , Figure 7 This is a schematic diagram of the structure of the electrocardiogram data processing device of the present application. In the present application, the electrocardiogram data processing device 700 may specifically include the following structure:

[0102] An acquisition unit 701 is used to acquire electrocardiogram data to be processed;

[0103] The processing unit 702 is used to input the to-be-processed electrocardiogram data into a preset electrocardiogram data processing model for processing, wherein the electrocardiogram data processing model is specifically a deep learning model, the electrocardiogram data processing model is pre-trained by the labeled sample electrocardiogram data, the electrocardiogram data processing model is used to predict the cardiac pathological state of the patient corresponding to the electrocardiogram data input into the model, the electrocardiogram data processing model includes a first model part, a second model part and a third model part, the first model part is used to extract the time domain features of the electrocardiogram data input into the model, the second model part is used to extract the frequency domain features of the electrocardiogram data input into the model, and the third model part is used to generate the model processing results based on the time domain features and the frequency domain features;

[0104] An extraction unit 703 is used to extract the cardiac pathological state determination result output by the electrocardiogram data processing model and corresponding to the electrocardiogram data to be processed;

[0105] The display unit 704 is used to display the cardiac pathological state determination result through a visual interface.

[0106] In an exemplary embodiment, the first model part specifically includes a Mamba module. After the initial time series features are obtained by aggregating the electrocardiogram data input to the model through a one-dimensional time series convolution layer, the Mamba module continues to extract time series features.

[0107] In another exemplary embodiment, the second model part specifically includes a GRU module, and after the initial frequency domain features are obtained by fast Fourier transform of the electrocardiogram data input to the model, the GRU module continues to extract frequency domain features.

[0108] In another exemplary embodiment, both the time domain features and the frequency domain features are average pooled and then concatenated, and the concatenated result is used as the input of the third model part.

[0109] In yet another exemplary embodiment, the third model part is specifically a double-layer MLP module.

[0110] In yet another exemplary embodiment, the cardiac pathological state specifically corresponds to at least one of a right ventricular hypertrophy disorder and a right ventricular enlargement disorder, and is divided into two states: a diseased state and a non-diseased state.

[0111] In yet another exemplary embodiment, the apparatus further includes a training unit 705, configured to:

[0112] Get sample ECG data;

[0113] Configure annotations for sample ECG data;

[0114] Based on the labeled sample ECG data, the ECG data processing model is trained.

[0115] This application also provides a processing device from the perspective of hardware structure, see Figure 8 , Figure 8 801, a memory 802, and an input / output device 803. The processor 801 is used to execute the computer program stored in the memory 802 to implement the following Figure 1 The steps of the electrocardiogram data processing method in the corresponding embodiment; or, the processor 801 is used to execute the computer program stored in the memory 802 to implement the following Figure 7 Corresponding to the functions of each unit in the embodiment, the memory 802 is used to store the processor 801 executing the above Figure 1 The computer program required by the electrocardiogram data processing method in the corresponding embodiment.

[0116] Exemplarily, the computer program may be divided into one or more modules / units, one or more modules / units are stored in the memory 802, and executed by the processor 801 to complete the present application. One or more modules / units may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.

[0117] The processing device may include, but is not limited to, a processor 801, a memory 802, and an input / output device 803. Those skilled in the art will appreciate that the illustration is merely an example of a processing device and does not constitute a limitation on the processing device, and may include more or fewer components than shown in the illustration, or a combination of certain components, or different components. For example, the processing device may also include a network access device, a bus, etc., and the processor 801, the memory 802, the input / output device 803, etc. are connected via a bus.

[0118] The processor 801 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the processing device, and uses various interfaces and lines to connect various parts of the entire device.

[0119] The memory 802 can be used to store computer programs and / or modules. The processor 801 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 802 and calling the data stored in the memory 802. The memory 802 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the processing device, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0120] When the processor 801 is used to execute the computer program stored in the memory 802, the following functions can be implemented:

[0121] Obtaining the electrocardiogram data to be processed;

[0122] Input the electrocardiogram data to be processed into a preset electrocardiogram data processing model for processing, wherein the electrocardiogram data processing model is specifically a deep learning model, the electrocardiogram data processing model is pre-trained by the labeled sample electrocardiogram data, the electrocardiogram data processing model is used to predict the cardiac pathological state of the patient corresponding to the electrocardiogram data input into the model, the electrocardiogram data processing model includes a first model part, a second model part and a third model part, the first model part is used to extract the time domain features of the electrocardiogram data input into the model, the second model part is used to extract the frequency domain features of the electrocardiogram data input into the model, and the third model part is used to generate the model processing results based on the time domain features and the frequency domain features;

[0123] extracting the cardiac pathological state determination result output by the electrocardiogram data processing model and corresponding to the electrocardiogram data to be processed;

[0124] The results of cardiac pathological status determination are displayed through a visual interface.

[0125] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the electrocardiogram data processing device, processing equipment and corresponding units described above can refer to the following. Figure 1 The description of the electrocardiogram data processing method in the corresponding embodiment will not be repeated here.

[0126] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0127] To this end, the present application provides a computer-readable storage medium, in which a plurality of instructions are stored, and the instructions can be loaded by a processor to execute the present application as follows: Figure 1 The steps of the electrocardiogram data processing method in the corresponding embodiment, the specific operation can refer to the following Figure 1 The description of the electrocardiogram data processing method in the corresponding embodiment will not be repeated here.

[0128] The computer-readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0129] Due to the instructions stored in the computer-readable storage medium, the present application can be executed. Figure 1 The steps of the electrocardiogram data processing method in the corresponding embodiment, therefore, the present application can be implemented as follows Figure 1 The beneficial effects that can be achieved by the electrocardiogram data processing method in the corresponding embodiment are detailed in the previous description and will not be repeated here.

[0130] The above is a detailed introduction to the electrocardiogram data processing method, device, processing equipment and computer-readable storage medium provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the core idea of ​​the present application; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for processing electrocardiogram data, characterized in that: The method comprises: Obtaining the electrocardiogram data to be processed; Input the to-be-processed electrocardiogram data into a preset electrocardiogram data processing model for processing, wherein the electrocardiogram data processing model is specifically a deep learning model, the electrocardiogram data processing model is pre-trained by labeled sample electrocardiogram data, the electrocardiogram data processing model is used to predict the cardiac pathological state of the patient corresponding to the electrocardiogram data input into the model, the electrocardiogram data processing model comprises a first model part, a second model part and a third model part, the first model part is used to extract the time domain features of the electrocardiogram data input into the model, the second model part is used to extract the frequency domain features of the electrocardiogram data input into the model, and the third model part is used to generate a model processing result based on the time domain features and the frequency domain features; extracting the cardiac pathological state determination result output by the electrocardiogram data processing model and corresponding to the electrocardiogram data to be processed; The cardiac pathological state determination result is displayed through a visual interface.

2. The method according to claim 1, characterized in that The first model part specifically includes a Mamba module. After the initial time series features are obtained by aggregating the electrocardiogram data input by the model through a one-dimensional time series convolution layer, the Mamba module continues to extract time series features.

3. The method according to claim 1, characterized in that The second model part specifically includes a GRU module. After the initial frequency domain features are obtained by fast Fourier transform of the electrocardiogram data input to the model, the GRU module continues to extract frequency domain features.

4. The method according to claim 1, characterized in that The time domain features and the frequency domain features are respectively average-pooled and then concatenated, and the concatenated result is used as the input of the third model part.

5. The method according to claim 1, characterized in that The third model part is specifically a double-layer MLP module.

6. The method according to claim 1, characterized in that The cardiac pathological state specifically corresponds to at least one of right ventricular hypertrophy and right ventricular enlargement, and is divided into two states: diseased and non-disease.

7. The method according to claim 1, characterized in that The method further comprises: Acquiring the sample electrocardiogram data; Configuring annotations for the sample electrocardiogram data; The electrocardiogram data processing model is trained based on the labeled sample electrocardiogram data.

8. An electrocardiogram data processing device, characterized in that: The device comprises: An acquisition unit, used for acquiring electrocardiogram data to be processed; A processing unit, used for inputting the to-be-processed electrocardiogram data into a preset electrocardiogram data processing model for processing, wherein the electrocardiogram data processing model is specifically a deep learning model, the electrocardiogram data processing model is pre-trained by labeled sample electrocardiogram data, the electrocardiogram data processing model is used to predict the cardiac pathological state of the patient corresponding to the electrocardiogram data input into the model, the electrocardiogram data processing model comprises a first model part, a second model part and a third model part, the first model part is used to extract the time domain features of the electrocardiogram data input into the model, the second model part is used to extract the frequency domain features of the electrocardiogram data input into the model, and the third model part is used to generate a model processing result based on the time domain features and the frequency domain features; an extraction unit, configured to extract a cardiac pathological state determination result output by the electrocardiogram data processing model and corresponding to the electrocardiogram data to be processed; A display unit is used to display the cardiac pathological state determination result through a visual interface.

9. A processing device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the method according to any one of claims 1 to 7 when calling the computer program in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method according to any one of claims 1 to 7.

Images