Lead reconstruction model training method and electrocardiosignal determination method

Abstract

The invention relates to a lead reconstruction model training method and an electrocardiosignal determination method. The method comprises the following steps: acquiring 12-lead electrocardiogram data samples, and taking the electrocardiogram data samples as labels for model training; extracting a chest electrocardio data sample of precordial leads according to the electrocardio data sample, and determining a difference value of signals between the precordial leads according to the chest electrocardio data sample to obtain a target electrocardio data sample corresponding to the tag; and training a lead reconstruction model to be trained according to the target electrocardiogram data sample and the tag to obtain a trained lead reconstruction model. By means of the method, standard twelve leads of the target object can be reconstructed by measuring a small number of precordial lead signals, and the information amount of electrocardiosignals is increased while the electrocardio patch is fully simplified.

Description

Technical Field

[0001] The present application relates to the technical field of electrocardiogram signal processing, and in particular to a training method for a lead reconstruction model and an electrocardiogram signal determination method. Background Art

[0002] As a commonly used method for diagnosing cardiovascular diseases, the electrocardiogram can record the changes in electrical activity produced by each cardiac cycle of the heart. In clinical practice, a 12-lead electrocardiogram is usually used. The 12 leads include 6 limb leads (I, II, III, aVR, aVL, aVF) obtained through limb electrodes, and 6 precordial leads (V1, V2, V3, V4, V5, V6) obtained through chest electrodes.

[0003] At present, the 12-lead ECG acquisition equipment usually lacks portability and is not suitable for scenarios such as home monitoring and long-term monitoring. In addition, since the recording time is usually short, it is difficult to detect occasional heart problems from the 12-lead ECG. In view of this, small-area single-lead ECG patches have emerged that can meet the needs of portability and long-term monitoring. However, due to the reduction in the number of electrodes, small ECG patches usually reflect limited heart information, which is not conducive to the diagnosis of heart disease.

[0004] Therefore, there is a problem of insufficient information in the current ECG signal acquisition technology. Summary of the invention

[0005] Based on this, it is necessary to provide a training method for a lead reconstruction model, an ECG signal determination method, device, computer equipment, computer-readable storage medium, computer program product and ECG signal acquisition equipment that can increase the amount of information in response to the above-mentioned technical problems.

[0006] In a first aspect, the present application provides a method for training a lead reconstruction model. The method comprises:

[0007] Obtaining twelve-lead ECG data samples, and using the ECG data samples as labels for model training;

[0008] Extracting chest ECG data samples of precordial leads according to the ECG data samples, determining the difference between the signals of the precordial leads according to the chest ECG data samples, and obtaining target ECG data samples corresponding to the labels;

[0009] The lead reconstruction model to be trained is trained according to the target ECG data sample and the label to obtain a trained lead reconstruction model.

[0010] In a second aspect, the present application also provides a method for training a lead reconstruction model. The method comprises:

[0011] Acquire an original lead reconstruction model, and acquire measured twelve-lead signals and measured pre-chest lead signals of the target object; the original lead reconstruction model is trained according to the lead reconstruction model training method described in any one of the first aspects above;

[0012] Using the measured twelve-lead signals as labels for model training, and obtaining target ECG data samples corresponding to the labels according to the differences between the measured pre-chest lead signals;

[0013] The original lead reconstruction model is updated according to the target ECG data sample and the label to obtain an updated lead reconstruction model.

[0014] In a third aspect, the present application provides a method for determining an electrocardiogram signal. The method comprises:

[0015] Acquire chest lead signals of the target object;

[0016] Determining a reference lead signal from the precordial lead signal, and determining the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0017] Determining a lead signal difference between the lead signal to be processed and the reference lead signal;

[0018] The lead signal difference is input into the trained lead reconstruction model as described in the first aspect or the updated lead reconstruction model as described in the second aspect to obtain the twelve-lead ECG signal of the target object.

[0019] In a fourth aspect, the present application also provides an electrocardiogram signal determination device. The device comprises:

[0020] An acquisition module, used for acquiring a chest lead signal of a target object;

[0021] A determination module, used to determine a reference lead signal from the precordial lead signal, and determine the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0022] A difference module, used to determine a lead signal difference between the lead signal to be processed and the reference lead signal;

[0023] A reconstruction module is used to input the lead signal difference into the trained lead reconstruction model as described in the first aspect or the updated lead reconstruction model as described in the second aspect to obtain the twelve-lead ECG signal of the target object.

[0024] In a fifth aspect, the present application further provides an electrocardiogram signal acquisition device. The electrocardiogram signal acquisition device comprises:

[0025] An ECG patch, used to collect the precordial lead signal of the target object and send the precordial lead signal to a signal processor;

[0026] A signal processor is used to determine a reference lead signal from the received precordial lead signal, determine the precordial lead signal other than the reference lead signal as a lead signal to be processed, determine a lead signal difference between the lead signal to be processed and the reference lead signal, input the lead signal difference into the trained lead reconstruction model as described in the first aspect above or the updated lead reconstruction model as described in the second aspect above, to obtain a twelve-lead ECG signal of the target object.

[0027] In a sixth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0028] Acquire chest lead signals of the target object;

[0029] Determining a reference lead signal from the precordial lead signal, and determining the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0030] Determining a lead signal difference between the lead signal to be processed and the reference lead signal;

[0031] The lead signal difference is input into the trained lead reconstruction model as described in the first aspect or the updated lead reconstruction model as described in the second aspect to obtain the twelve-lead ECG signal of the target object.

[0032] In a seventh aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0033] Acquire chest lead signals of the target object;

[0034] Determining a reference lead signal from the precordial lead signal, and determining the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0035] Determining a lead signal difference between the lead signal to be processed and the reference lead signal;

[0036] The lead signal difference is input into the trained lead reconstruction model as described in the first aspect or the updated lead reconstruction model as described in the second aspect to obtain the twelve-lead ECG signal of the target object.

[0037] In an eighth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0038] Acquire chest lead signals of the target object;

[0039] Determining a reference lead signal from the precordial lead signal, and determining the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0040] Determining a lead signal difference between the lead signal to be processed and the reference lead signal;

[0041] The lead signal difference is input into the trained lead reconstruction model as described in the first aspect or the updated lead reconstruction model as described in the second aspect to obtain the twelve-lead ECG signal of the target object.

[0042] The training method, ECG signal determination method, device, computer equipment, storage medium, computer program product and ECG signal acquisition device of the lead reconstruction model are as follows: by acquiring twelve-lead ECG data samples, the ECG data samples are used as labels for model training, chest ECG data samples of precordial leads are extracted according to the ECG data samples, the difference between the signals of the precordial leads is determined according to the chest ECG data samples, and the target ECG data samples corresponding to the labels are obtained; the lead reconstruction model to be trained is trained according to the target ECG data samples and the labels to obtain a trained lead reconstruction model; the standard twelve-lead ECG signal of the target object can be reconstructed based on the trained lead reconstruction model; since the reconstruction is performed using the lead signal difference, the process of calculating the WCT (Wilson's Central Terminal) according to the limb leads is avoided, so that the standard twelve-lead signal can be reconstructed by measuring a small number of precordial lead signals, while fully simplifying the ECG patch, the amount of information of the ECG signal is increased. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of a training method for a lead reconstruction model in one embodiment;

[0044] Figure 2 is a schematic diagram of precordial leads in one embodiment;

[0045] Figure 3 is a flow chart of a training method for a lead reconstruction model in another embodiment;

[0046] Figure 4 is a flow chart of a method for determining an electrocardiogram signal in one embodiment;

[0047] Figure 5 is a flow chart of a training process of a lead reconstruction model in one embodiment;

[0048] Figure 6 is a schematic diagram of a flow chart of lead reconstruction model application in one embodiment;

[0049] Figure 7 It is a flowchart of a method for determining an electrocardiogram signal in another embodiment;

[0050] Figure 8 is a structural block diagram of an electrocardiogram signal determination device in one embodiment;

[0051] Fig. 9 is a structural block diagram of an electrocardiogram signal acquisition device in an embodiment;

[0052] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] In one embodiment, Figure 1 As shown, a training method for a lead reconstruction model is provided. This embodiment uses the method applied to a terminal as an example for illustration. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0055] Step S110, obtaining twelve-lead ECG data samples, and using the ECG data samples as labels for model training.

[0056] The twelve-lead ECG data samples may be multiple groups of twelve-lead ECG signals.

[0057] In a specific implementation, multiple groups of twelve-lead ECG signals can be used as twelve-lead ECG data samples and input into a terminal. A lead reconstruction model to be trained is pre-set at the terminal, and the received twelve-lead ECG data samples are used as labels for training the lead reconstruction model to be trained.

[0058] In practical applications, multiple groups of twelve-lead ECG signals can be collected using ECG acquisition equipment, or multiple groups of twelve-lead ECG signals can be obtained from public data sets as ECG data samples and input into the terminal. The terminal can preprocess the received ECG data samples and use the preprocessed ECG data samples as labels for training the lead reconstruction model to be trained.

[0059] Step S120, extracting chest ECG data samples of precordial leads according to the ECG data samples, determining the difference between the signals of the precordial leads according to the chest ECG data samples, and obtaining target ECG data samples corresponding to the labels.

[0060] The chest ECG data samples of the precordial leads may be a plurality of groups of precordial lead signals, and the target ECG data samples may be the lead signal difference corresponding to each group of precordial lead signals.

[0061] In a specific implementation, the terminal can remove the limb lead data from each group of ECG data samples, use the remaining precordial lead data as the chest ECG data samples, and determine a reference sample therefrom, subtract the chest ECG data samples other than the reference sample from the reference sample to obtain the difference in signals between the precordial leads, and obtain the target ECG data sample corresponding to the group of ECG data samples based on the difference, and the target ECG data sample corresponds to the label.

[0062] Figure 2 A schematic diagram of the precordial leads is provided. Figure 2 , for a group of ECG data samples V1, V2, V3, V4, V5, V6, I, II, III, aVR, aVL, aVF, by removing the limb lead data, we can get chest ECG data samples V1, V2, V3, V4, V5, V6. Taking V1 as the reference sample as an example, V2, V3, V4, V5, V6 are subtracted from the reference sample to get target ECG data samples V2-V1, V3-V1, V4-V1, V5-V1, V6-V1. The target ECG data samples correspond to labels V1, V2, V3, V4, V5, V6, I, II, III, aVR, aVL, aVF.

[0063] It should be noted that the determination of the signal difference between the precordial leads is not limited to the above examples. V2, V3, V4, V5 or V6 can also be selected as reference samples, or adjacent precordial leads can be subtracted to obtain the signal difference between the precordial leads. Table 1 provides several calculation methods for determining the signal difference between the precordial leads, where removing all 0 columns refers to removing V1-V1, V2-V2, V3-V3, V4-V4, V5-V5, and V6-V6.

[0064] Table 1 Calculation method of signal difference between precordial leads

[0065] Calculation method Minuend Subtrahend illustrate Neighbor Subtraction V1 to V5 V2 to V6 V1-V2, V2-V3, ..., V5-V6 Subtract from V1 V1 to V6 V1 Subtract V1 respectively and remove all 0 columns Difference with V2 V1 to V6 V2 Subtract V2 respectively and remove all 0 columns Difference with V3 V1 to V6 V3 Subtract V3 respectively and remove all 0 columns Difference with V4 V1 to V6 V4 Subtract V4 respectively and remove all 0 columns Difference with V5 V1 to V6 V5 Subtract V5 respectively and remove all 0 columns Difference with V6 V1 to V6 V6 Subtract V6 respectively and remove all 0 columns

[0066] Step S130 , training the lead reconstruction model to be trained according to the target ECG data samples and labels to obtain a trained lead reconstruction model.

[0067] Among them, the lead reconstruction model to be trained can be constructed based on, but not limited to, ANN (Artificial Neural Networks), CNN (Convolutional Neural Networks), DNN (Deep Neural Networks) or LSTM (Long Short-term Memory Network).

[0068] In a specific implementation, the terminal can input the target ECG data samples into the lead reconstruction model to be trained. The lead reconstruction model to be trained reconstructs the twelve-lead ECG signal according to the target ECG data samples, compares the reconstructed twelve-lead ECG signal with the label, and trains the lead reconstruction model to be trained according to the error between the two to obtain a trained lead reconstruction model.

[0069] For example, the target ECG data samples V2-V1, V3-V1, V4-V1, V5-V1, and V6-V1 can be input into the lead reconstruction model to be trained. The lead reconstruction model to be trained performs forward propagation according to the target ECG data samples to obtain the reconstructed twelve-lead ECG signals V1', V2', V3', V4', V5', V6', I', II', III', aVR', aVL', and aVF'. The reconstructed twelve-lead ECG signals are input into a pre-constructed loss function, and the loss value of the loss function is calculated. The loss value reflects the error between the reconstructed twelve-lead ECG signals and the labels V1, V2, V3, V4, V5, V6, I, II, III, aVR, aVL, and aVF. Back propagation is performed with the goal of minimizing the loss value, and the parameters of the lead reconstruction model to be trained are updated until a preset number of iterations is reached or the loss value is minimized. At this time, the trained lead reconstruction model is obtained.

[0070] The training method of the lead reconstruction model obtains twelve-lead ECG data samples, uses the ECG data samples as labels for model training, extracts chest ECG data samples of precordial leads according to the ECG data samples, determines the difference between the signals of the precordial leads according to the chest ECG data samples, obtains target ECG data samples corresponding to the labels, and trains the lead reconstruction model to be trained according to the target ECG data samples and the labels to obtain a trained lead reconstruction model; based on the trained lead reconstruction model, the standard twelve-lead ECG signal of the target object can be reconstructed. Since the reconstruction is performed using the lead signal difference, the process of calculating the WCT based on the limb leads is avoided, so that the standard twelve-lead signal can be reconstructed by measuring a small number of precordial lead signals, which increases the amount of information of the ECG signal while fully simplifying the ECG patch.

[0071] In one embodiment, the above step S120 may specifically include: determining a reference ECG data sample from the chest ECG data samples, and determining the chest ECG data samples other than the reference ECG data samples as the ECG data samples to be processed; and obtaining the target ECG data sample based on the difference between the ECG data sample to be processed and the reference ECG data sample.

[0072] The reference ECG data sample may be a reference sample in the chest ECG data sample, and the ECG data sample to be processed may be a chest ECG data sample other than the reference ECG data sample.

[0073] In a specific implementation, the terminal can select any one of each group of chest ECG data samples as a reference ECG data sample, and use all chest ECG data samples other than the reference ECG data sample as ECG data samples to be processed, and subtract the ECG data sample to be processed from the reference ECG data sample to obtain the target ECG data sample.

[0074] For example, for a group of chest ECG data samples V1, V2, V3, V4, V5, and V6, V1 is selected as the reference ECG data sample, then V2, V3, V4, V5, and V6 are the ECG data samples to be processed, and V2, V3, V4, V5, and V6 are subtracted from V1 to obtain target ECG data samples V2-V1, V3-V1, V4-V1, V5-V1, and V6-V1.

[0075] In this embodiment, a reference ECG data sample is determined from the chest ECG data samples, and the chest ECG data samples other than the reference ECG data samples are determined as the ECG data samples to be processed; according to the difference between the ECG data sample to be processed and the reference ECG data sample, the target ECG data sample is obtained, and the difference between the chest ECG data samples can be used to train the lead reconstruction model, so that the trained lead reconstruction model can reconstruct the twelve leads only based on the precordial leads, without using the limb leads, thereby fully simplifying the ECG patch and increasing the amount of information of the ECG signal.

[0076] In one embodiment, before the above step S110, the method may further include: determining model parameters of the lead reconstruction model to be trained; and generating the lead reconstruction model to be trained according to the model parameters.

[0077] The model parameters may be weights, biases and hyperparameters of the lead reconstruction model, wherein the hyperparameters may be parameters reflecting the architecture of the lead reconstruction model, such as the number of layers and nodes of the lead reconstruction model.

[0078] In a specific implementation, the terminal may predetermine the model parameters of the lead reconstruction model, including the number of layers of the model, the number of nodes in each layer, and the weights and biases of the lead reconstruction model, and generate the lead reconstruction model to be trained according to the determined model parameters.

[0079] For example, taking the construction of a lead reconstruction model based on ANN as an example, a three-layer structure can be set. The first layer is the input layer, which contains 5 neurons, corresponding to the target ECG data sample, the second layer is the hidden layer, which contains n neurons, and the third layer is the output layer, which contains 12 neurons, corresponding to the reconstructed twelve-lead ECG signal. Sigmoid (S-type function) is used as the activation function. Before training the lead reconstruction model, the number n of hidden layer neurons, as well as weights and biases are set using a random initialization method or according to actual conditions to obtain the lead reconstruction model to be trained. During the back propagation process, the lead reconstruction model to be trained is trained using methods such as batch gradient descent to obtain the final number of hidden layer neurons, for example, n=10, as well as weights and biases.

[0080] In this embodiment, by determining the model parameters of the lead reconstruction model to be trained; generating the lead reconstruction model to be trained according to the model parameters, the architecture of the lead reconstruction model can be set according to actual conditions to match it with actual needs, thereby improving the applicability of the lead reconstruction model.

[0081] In one embodiment, the lead reconstruction model to be trained includes an input layer, a hidden layer and an output layer; the number of neurons in the input layer is 5, and the number of neurons in the output layer is 12.

[0082] In a specific implementation, the terminal can set the lead reconstruction model to be trained to a three-layer structure, where the first layer is the input layer, which contains 5 neurons, corresponding to the target ECG data samples, for example, V2-V1, V3-V1, V4-V1, V5-V1, and V6-V1; the second layer is the hidden layer, and the number of neurons is randomly selected or determined according to actual conditions; the third layer is the output layer, and the number of neurons is 12, corresponding to the reconstructed twelve-lead ECG signals, for example, V1', V2', V3', V4', V5', V6', I', II', III', aVR', aVL', and aVF'.

[0083] In this embodiment, by making the lead reconstruction model to be trained include an input layer, a hidden layer and an output layer; the number of neurons in the input layer is 5, and the number of neurons in the output layer is 12, the twelve-lead ECG signal can be reconstructed according to the lead signal difference between the precordial lead signals, thereby fully simplifying the ECG patch and increasing the amount of information of the ECG signal.

[0084] In one embodiment, the above step S110 may specifically include: resampling the original twelve-lead signal corresponding to the ECG data sample to obtain a resampled twelve-lead signal; filtering the resampled twelve-lead signal to obtain a filtered twelve-lead signal; and normalizing the filtered twelve-lead signal to obtain a twelve-lead ECG data sample.

[0085] The original twelve-lead signals may be multiple groups of twelve-lead electrocardiogram signals originally acquired.

[0086] In a specific implementation, the original twelve-lead signal can be input into the terminal, the terminal resamples the original twelve-lead signal to ensure that the twelve-lead signals from different sources have the same frequency, and obtains a resampled twelve-lead signal. The resampled twelve-lead signal is filtered to remove noise interference to obtain a filtered twelve-lead signal, and the filtered twelve-lead signal is normalized to obtain a twelve-lead ECG data sample for training a lead reconstruction model.

[0087] In practical applications, multiple groups of twelve-lead signals collected by ECG acquisition equipment or multiple groups of twelve-lead signals obtained from public data sets can be resampled to ensure that the twelve-lead signals from different sources have the same frequency. For example, the resampling frequency can be set to 200 Hz, and then filtered to remove noise interference in the twelve-lead signals. Specifically, low-pass filtering or band-pass filtering can be used. The filter can be, but is not limited to, a Butterworth filter or a Chebyshev filter. Finally, normalization is performed. The specific formula is:

[0088]

[0089] Among them, max and min represent the maximum and minimum values ​​of the filtered twelve-lead signal respectively, and x is the filtered twelve-lead signal. It is the twelve-lead ECG data sample obtained after normalization.

[0090] In this embodiment, the original twelve-lead signal corresponding to the ECG data sample is resampled to obtain a resampled twelve-lead signal; the resampled twelve-lead signal is filtered to obtain a filtered twelve-lead signal; the filtered twelve-lead signal is normalized to obtain a twelve-lead ECG data sample, and the original twelve-lead signal can be preprocessed to make the obtained ECG data sample suitable for model training, thereby improving the applicability of the ECG data sample.

[0091] In one embodiment, Figure 3 As shown, another training method for the lead reconstruction model is provided. Taking the application in a terminal as an example, the method includes the following steps:

[0092] Step S210, obtaining the original lead reconstruction model, and obtaining the measured twelve-lead signals and the measured pre-chest lead signals of the target object;

[0093] Step S220, using the measured twelve-lead signal as a label for model training, and obtaining a target ECG data sample corresponding to the label according to the difference between the measured precordial lead signals;

[0094] Step S230: updating the original lead reconstruction model according to the target ECG data sample and label to obtain an updated lead reconstruction model.

[0095] The original lead reconstruction model may be a lead reconstruction model obtained through initial training.

[0096] The target object may be a user who is to undergo a heart examination.

[0097] The measured twelve-lead signal may be a standard twelve-lead signal obtained by actual measurement. The measured pre-chest lead signal may be a pre-chest potential obtained by actual measurement.

[0098] In a specific implementation, as described in the aforementioned embodiment, the terminal can obtain twelve-lead ECG data samples, use the ECG data samples as labels for model training, extract chest ECG data samples of precordial leads based on the ECG data samples, determine the difference in signals between the precordial leads based on the chest ECG data samples, obtain target ECG data samples corresponding to the labels, and train the lead reconstruction model to be trained based on the target ECG data samples and the labels to obtain the original lead reconstruction model. To further improve the accuracy of lead reconstruction, the measured twelve-lead signal of the target object can be obtained through a twelve-lead ECG acquisition device, and the measured precordial lead signal of the target object can be obtained through a small ECG patch. The measured twelve-lead signal is used as a label, and the difference between the measured precordial lead signals is used as a target ECG data sample. The target ECG data sample is input into the original lead reconstruction model. The original lead reconstruction model reconstructs the twelve-lead ECG signal according to the target ECG data sample, and the reconstructed twelve-lead ECG signal is compared with the label. The original lead reconstruction model is updated according to the error between the two to obtain an updated lead reconstruction model.

[0099] For example, an ECG acquisition device can be used to acquire multiple groups of twelve-lead ECG signals, or multiple groups of twelve-lead ECG signals can be acquired from a public data set. ECG data samples are obtained after preprocessing, and the ECG data samples are used as labels for model training. Chest ECG data samples are extracted from the ECG data samples, and target ECG data samples are determined based on the chest ECG data samples. The model is trained using the target ECG data samples and labels to obtain the original lead reconstruction model. Since there are individual differences between different users, and the original lead reconstruction model is a universal model, the twelve-lead ECG signals obtained after reconstruction may be inaccurate for a specific user. In order to increase the accuracy of lead reconstruction, a personalized model can be constructed for different users. Specifically, the target object can be made to wear a twelve-lead ECG acquisition device and a small ECG patch at the same time. The target object's standard twelve-lead signal, i.e., the measured twelve-lead signal, is acquired through the twelve-lead ECG acquisition device, and the measured twelve-lead signal is used as a label. At the same time, the target object's six pre-chest potentials UV1, UV2, UV3, UV4, UV5, and UV6, i.e., the measured pre-chest lead potentials, are acquired through the small ECG patch. signals, and calculate the difference between the prethoracic potentials, and use the obtained UV2-UV1, UV3-UV1, UV4-UV1, UV5-UV1, and UV6-UV1 as target ECG data samples, and input the target ECG data samples into the original lead reconstruction model. The original lead reconstruction model reconstructs the twelve-lead ECG signal according to UV2-UV1, UV3-UV1, UV4-UV1, UV5-UV1, and UV6-UV1, and compares the reconstructed twelve-lead ECG signal with the measured twelve-lead signal. According to the error between the two, the model parameters of the original lead reconstruction model are adjusted until the preset conditions are met, and the updated lead reconstruction model is obtained.

[0100] It should be noted that the determination of the difference between the precordial potentials is not limited to the above examples, and UV2, UV3, UV4, UV5 or UV6 can also be selected as the subtrahend, or adjacent precordial potentials can be subtracted. The specific details are similar to the above embodiments and will not be repeated here.

[0101] The training method of the lead reconstruction model mentioned above obtains the original lead reconstruction model, as well as the measured twelve-lead signal and the measured precordial lead signal of the target object, uses the measured twelve-lead signal as the label for model training, obtains the target ECG data sample corresponding to the label according to the difference between the measured precordial lead signals, updates the original lead reconstruction model according to the target ECG data sample and the label, and obtains the updated lead reconstruction model; based on the updated lead reconstruction model, the standard twelve-lead ECG signal of the target object can be reconstructed, and since the lead reconstruction model is personalized for different target objects, the accuracy of lead reconstruction is increased.

[0102] In one embodiment, Figure 4 As shown, a method for determining an electrocardiogram signal is provided. This embodiment uses the method applied to a terminal as an example for illustration. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server.

[0103] In this embodiment, the method includes the following steps:

[0104] Step S310, obtaining a precordial lead signal of the target object;

[0105] Step S320, determining a reference lead signal from the precordial lead signal, and determining the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0106] Step S330, determining a lead signal difference between the lead signal to be processed and the reference lead signal;

[0107] Step S340: input the lead signal difference into the trained lead reconstruction model or the updated lead reconstruction model to obtain the twelve-lead ECG signal of the target object.

[0108] The chest lead signal may be an electrocardiogram signal collected by electrodes placed on the chest, for example, an electric potential.

[0109] The reference lead signal may be a reference signal in the precordial lead signal, for example, a reference potential. The lead signal to be processed may be a precordial lead signal other than the reference signal. The lead signal difference may be the difference between the lead signal to be processed and the reference lead signal.

[0110] In a specific implementation, electrodes can be placed on the chest of the target object, and the pre-chest lead signals of the target object can be collected through the electrodes and input into the terminal. After the terminal obtains the pre-chest lead signals of the target object, a reference lead signal can be determined from the pre-chest lead signals, and all pre-chest lead signals other than the reference lead signal can be determined as lead signals to be processed. The lead signals to be processed are subtracted from the reference lead signals respectively to obtain lead signal differences, and the lead signal differences are input into a trained lead reconstruction model or an updated lead reconstruction model. The trained lead reconstruction model or the updated lead reconstruction model identifies the lead signal differences to obtain the twelve-lead ECG signals of the target object.

[0111] For example, six chest electrodes can be placed on the chest of the target object to measure the potentials at six different positions on the chest of the target object, and six chest potentials UV1', UV2', UV3', UV4', UV5', and UV6' are obtained respectively. The six chest potentials are input to the terminal, and the terminal uses the received six chest potentials as the chest lead signals of the target object. According to the potential difference between the chest potential and WCT, the chest lead can be obtained. Since WCT needs to be introduced, and WCT needs to be obtained by installing electrodes on the limbs, but in order to achieve the miniaturization of ECG acquisition equipment, it is necessary to reduce the number of electrodes. Therefore, how to achieve the acquisition of chest leads without introducing limb lead electrodes is a problem that needs to be solved. Given that the chest lead is mainly used to represent the activity information of the heart on the horizontal plane, and the difference between different chest potentials is equivalent to the amount of information carried by the chest lead, it can be used to solve the problem of introducing WCT. For example, the difference between the six chest potentials can be calculated. Specifically, one can be selected from the six precordial potentials as the reference lead signal, and the remaining five precordial potentials can be used as lead signals to be processed. The lead signal to be processed is subtracted from the reference lead signal to obtain the lead signal difference. For example, UV1' is used as the reference lead signal, and UV2', UV3', UV4', UV5', and UV6' are used as lead signals to be processed, then the lead signal differences are UV2'-UV1', UV3'-UV1', UV4'-UV1', UV5'-UV1', and UV6'-UV1'. A lead reconstruction model is constructed based on ANN, CNN or DNN. For patients with ventricular tachycardia, congestive heart failure, ECG ST-segment elevation myocardial infarction and other diseases, a lead reconstruction model can also be constructed based on LSTM. UV2'-UV1', UV3'-UV1', UV4'-UV1', UV5'-UV1', and UV6'-UV1' are input into the trained lead reconstruction model. The trained lead reconstruction model can output standard 12 leads including 6 precordial leads and 6 limb leads.

[0112] The above-mentioned ECG signal determination method obtains the precordial lead signal of the target object, determines the reference lead signal from the precordial lead signal, determines the precordial lead signal other than the reference lead signal as the lead signal to be processed, determines the lead signal difference between the lead signal to be processed and the reference lead signal, inputs the lead signal difference into a trained lead reconstruction model or an updated lead reconstruction model, and obtains the twelve-lead ECG signal of the target object; since the lead reconstruction is performed using the lead signal difference, the process of calculating the WCT based on the limb leads is avoided, and the standard twelve-lead of the target object can be reconstructed by measuring a small number of precordial lead signals, so that the amount of information of the ECG signal is increased while the ECG patch is fully simplified.

[0113] In one embodiment, the above step S310 may specifically include: acquiring six pre-chest potentials obtained by collecting data from six pre-chest electrodes on the target object; and determining the six pre-chest potentials as pre-chest lead signals of the target object.

[0114] The chest electrode may be an electrode placed on the chest, and the chest potential may be a potential on the chest collected by the chest electrode.

[0115] In a specific implementation, six chest electrodes may be placed on the chest of the target object to measure the potentials at six different positions on the chest of the target object, thereby obtaining six chest potentials, which are used as chest lead signals of the target object.

[0116] In actual application, the following positions can be preset: Position 1, 4th intercostal space on the right side of the sternum; Position 2, 4th intercostal space on the left side of the sternum; Position 3, midpoint between Position 2 and Position 4; Position 4, intersection of the left midclavicular line and the 5th intercostal space; Position 5, intersection of the horizontal position 4 and the anterior axillary line; Position 6, intersection of the horizontal position 4 and the mid-axillary line. The 6 electrodes are placed at the above preset positions respectively, and the 6 potentials measured by the 6 electrodes are the precordial lead signals.

[0117] In this embodiment, six pre-chest potentials are obtained by acquiring six pre-chest electrodes for the target object; the six pre-chest potentials are determined as the pre-chest lead signals of the target object, and there is no need to install electrodes on the limbs, thereby ensuring the portability of the ECG signal acquisition device.

[0118] In order to facilitate those skilled in the art to have a deeper understanding of the embodiments of the present application, a specific example will be described below.

[0119] Small-area single-lead ECG patches are more portable, have a significant improvement in wearing comfort compared to traditional ECG acquisition equipment, and have standby capabilities, which are of great significance for observing occasional heart problems such as irregular heartbeats. At the same time, small ECG patches also have the defect of uncontrollable wearing position, and the amount of information collected by ECG signals is less than that of traditional 12-lead ECG patches. On the premise of clarifying the wearing position, only the precordial leads are recorded, which can further simplify the ECG patch. However, the precordial leads need to be determined by the Wilson center point, and the determination of the Wilson center point requires wearing limb electrodes, resulting in the ECG patch not being further simplified. Based on this, the present application proposes a method for converting a standard 12-lead ECG based on a small ECG patch. By wearing an ECG patch on the chest, a 12-lead ECG is obtained, so that the amount of ECG signal information is increased while simplifying the ECG acquisition equipment.

[0120] Figure 5 A flowchart of the training process of the lead reconstruction model is provided. Figure 5, the training process of the lead reconstruction model may include the following steps:

[0121] Step 1: collect the original twelve-lead ECG signals and divide them into a training set, a validation set and a test set. The training set is used to determine the weights and biases of the lead reconstruction model, the validation set is used to determine the hyperparameters of the lead reconstruction model, such as the number of model layers and the number of neurons, and the test set is used to evaluate the trained lead reconstruction model.

[0122] Step 2, preprocess the original twelve-lead ECG signal, first resample the original twelve-lead ECG signal to ensure that the original twelve-lead ECG signals from different sources have the same frequency. This application sets the resampling frequency to 200Hz; then filter the resampled twelve-lead ECG signal to remove noise interference. The filter cutoff frequency band can be a low-pass filter or a band-pass filter. The filter can be a Butterworth filter, a Chebyshev filter, or other filters with equivalent effects; finally, normalize according to the following formula

[0123]

[0124] Among them, max and min are the maximum and minimum values ​​in the training set respectively, and x is the filtered 12-lead ECG signal. It is a twelve-lead ECG data sample obtained after preprocessing.

[0125] Step 3: Use the twelve-lead ECG data samples obtained after preprocessing as labels for training the lead reconstruction model.

[0126] Step 4, remove the 6 limb lead data from the 12-lead ECG data sample, and use the remaining 6 chest lead data as chest ECG data samples, determine the difference between the chest lead signals based on the chest ECG data samples, and then obtain the target ECG data samples, which are input into the lead reconstruction model to be trained. Among them, since the chest lead cannot be obtained by the chest electrode alone, the introduction of WCT is also required, and WCT needs to be calculated using three limbs. In view of this, in the process of training the lead reconstruction model, the introduction of WCT can be avoided by subtracting the chest lead. For example, the chest ECG data sample V1 is used as the subtrahend, and the remaining 5 chest ECG data samples V2, V3, V4, V5, and V6 are used as the minuends. The target ECG data samples V2-V1, V3- V1, V4-V1, V5-V1, and V6-V1 are used to train the lead reconstruction model, among which V2-V1, V3-V1, V4-V1, V5-V1, and V6-V1 are equivalent to taking the five precordial potentials UV2, UV3, UV4, UV5, and UV6 as minuends and the precordial potential UV1 as subtrahend, and the resulting five potential differences are UV2-UV1, UV3-UV1, UV4-UV1, UV5-UV1, and UV6-UV1.

[0127] Step 5, determine the structure of the lead reconstruction model. The prediction of the lead reconstruction model is a type of regression prediction, including two implementation methods: linear and nonlinear. The linear method is mainly (multivariate) linear regression, and the nonlinear method is mainly neural network. Since the nonlinear method is not limited by the XOR problem, the nonlinear lead conversion method is usually better than the linear method. Taking the ANN method to construct the lead reconstruction model as an example, the target ECG data samples V2-V1, V3-V1, V4-V1, V5-V1, and V6-V1 are used as model inputs, and the standard twelve leads are used as model outputs. Therefore, the ANN is a three-layer structure: the first layer is the input layer, which contains 5 neurons, the second layer is the hidden layer, and after testing, it can be determined that 10 neurons are optimal, and the third layer is the output layer, which contains 12 neurons. Among them, the sigmoid function is used as the activation function. Before ANN training, the weights and biases of the hidden layer neurons are set using the random initialization method. During the back propagation process, the batch gradient descent method is used to update the model parameters.

[0128] Generally, the ANN model can achieve lead reconstruction well, but in some cases, the ANN model has limitations, for example, when the target subject suffers from the following diseases: ventricular tachycardia, congestive heart failure, ST-segment elevation myocardial infarction. In this case, the effect of lead reconstruction can be enhanced by building an LSTM model. The structure of the LSTM model is similar to that of the ANN, except that the second hidden layer is set to the LSTM structure. After testing, it can be determined that the optimal number of neurons is 32.

[0129] Step 6, initialization of the lead reconstruction model parameters. The initialization of the lead reconstruction model mainly initializes the weights and biases in each neuron. There are three common initialization methods: random initialization, fixed value initialization, and pre-training initialization. For the lead reconstruction model to be trained, random initialization or fixed value initialization is usually selected; for the trained lead reconstruction model, if you need to continue training, you can choose pre-training initialization.

[0130] Step 7, forward propagation. Forward propagation refers to the process of data being transmitted from the input layer, passing through the hidden layer, and finally reaching the output layer. The result of forward propagation is determined by the parameters of the lead reconstruction model and the input data. One forward propagation is the result of one fitting of the target ECG data sample by the lead reconstruction model.

[0131] Step 8: Lead reconstruction model training. Model training mainly includes constructing loss function, minimizing loss value and back propagation. The main process includes: 1. Calculating the state and activation value of each layer of the network until the last layer; 2. Calculating the error of each layer, and the error calculation process is carried forward from the last layer; 3. Update parameters. The details are as follows:

[0132] Constructing loss function: In the lead reconstruction model, the reconstruction result is the value of the twelve leads, which is a regression task. In the regression task, the calculation of the loss function is mainly the calculation of the error, including the mean square error loss function, the absolute value error loss function, and the mean absolute value error loss function.

[0133] Minimize the loss value: The error obtained by batch calculation is called the loss value. The commonly used method to reduce the loss value is the gradient descent algorithm. The gradient descent algorithm can be selected from the stochastic gradient descent algorithm, the small batch gradient descent method, the RMSprop (Root Mean Square prop) optimization algorithm, the adaptive moment estimation algorithm, etc.

[0134] Back propagation: The error back propagation algorithm is referred to as the back propagation algorithm. Back propagation is mainly used to update the parameters of the lead reconstruction model, and the parameters of the lead reconstruction model are updated layer by layer through the chain rule.

[0135] Update parameters: The parameters updated are mainly weights and biases. In the process of updating weights and biases, the parameters of the gradient descent algorithm are also updated.

[0136] Step 9: Lead reconstruction model training is completed. After the lead reconstruction model meets the set number of iterations, or the calculated loss value is minimized, the model training is completed and a trained lead reconstruction model is obtained.

[0137] Figure 6 A flowchart of the lead reconstruction model application is provided. Figure 6The training method of the lead reconstruction model can be applied to construct a general model or a personalized model, wherein a plurality of groups of twelve-lead ECG signals collected by an ECG acquisition device or a plurality of groups of twelve-lead ECG signals obtained from a public data set can be used to construct a general lead reconstruction model, which is an offline model and the model parameters cannot be updated; and the general lead reconstruction model can be personalized updated according to user data, and the new data collected by the ECG patch worn by the user can be accepted, and the model parameters can be updated according to the collected new data, so as to accurately convert the twelve-lead ECG signal of the user.

[0138] Specifically, a general model can be constructed according to path 1, that is, the previously trained lead reconstruction model. The offline dataset is used as input data to construct a lead reconstruction model with excellent generalization performance and strong robustness. The correlation coefficient (CC) and root mean square error (RMSE) can be used for model evaluation.

[0139] A personalized model can also be constructed according to Path 2, that is, the updated lead reconstruction model mentioned above. For the personalized model, the user's own situation needs to be considered. There are obvious differences in ECG signals between users with different physical conditions. Therefore, it is necessary to collect twelve-lead ECG data through the user wearing a twelve-lead ECG acquisition device to preliminarily construct a lead reconstruction model. Subsequently, six chest lead ECG data are collected through a small ECG patch worn by the user, and the parameters of the lead reconstruction model are updated based on this data, so that the updated lead reconstruction model can better learn the characteristics of the user's ECG data, and thus accurately convert the user's twelve-lead ECG data based on the user's six chest leads.

[0140] ECG signal is a weak signal with strong nonlinearity, non-stationarity and randomness. During its collection process, it is easily affected by the internal and external environment. In addition, users who need ECG monitoring usually suffer from heart-related diseases, so there are obvious individual differences in ECG signals of different patients. To solve this problem, a lead reconstruction model for the user himself is constructed to achieve differentiated detection and recognition of ECG signals of different patients. The characteristics of the personalized model mainly include:

[0141] Nonlinear methods based on neural networks: ANN and LSTM models based on neural networks are proposed. Such models can effectively reconstruct standard 12-lead ECG using precordial leads. Compared with traditional linear methods or machine learning methods, neural network-based methods can avoid the process of manually extracting ECG features. At the same time, by modifying the parameters to train the model, the model can be prevented from falling into the local optimal situation and the model's ability to fit ECG reconstruction can be improved. For the general model, multiple public databases such as the PTB database and the WCT-ECG database are used as input, and the results of model reconstruction are better than those of other methods.

[0142] Build a model based on the user's own data: Most of the current ECG lead reconstruction models do not take into account the individual differences between different users. In fact, the differences are so significant that they cannot be generalized. There are obvious limitations in using a universal model to reconstruct the ECG data of different users. To address this problem, this application proposes a method for building a personalized model based on the user's own data. For users with different heart diseases, the personalized model shows better lead reconstruction effects than the universal model.

[0143] In addition, the advantages of personalized models mainly include:

[0144] Effectively applicable to users with different heart diseases: ECG signals vary greatly between users with different heart diseases. The personalized model can learn the characteristics of the user's ECG signal by collecting the user's own data, thereby effectively reconstructing the ECG leads.

[0145] Can be used to monitor the recovery of illness: For users with heart disease, their electrocardiogram can reflect the severity of the disease to a certain extent. Through continuous monitoring of the user, the lead signal reconstructed by the personalized model can show the user's recovery status.

[0146] Protect user privacy and avoid data leakage: The training data of the personalized model comes from the user himself, and the collected data can be stored locally, thus avoiding the transfer and flow of data between different entities, and effectively preventing user data leakage.

[0147] Personalized locations can be set based on the user: Different users have different wearing habits and locations when wearing ECG patches, which can also cause signal differences. Learning the user's own ECG characteristics through a personalized model can effectively avoid such problems. In addition, for users who are unable to place the electrodes in the regular wearing location due to their own reasons, a personalized location can be set for them.

[0148] This application explains the effectiveness of the lead reconstruction model based on the general model. Using two public databases (PTB database and WCT-ECG database), the five chest leads except the V1 lead are subtracted from the V1 lead, that is, the V2 lead minus the V1 lead, the V3 lead minus the V1 lead, the V4 lead minus the V1 lead, the V5 lead minus the V1 lead, and the V6 lead minus the V1 lead. The ECG signals of the five chest lead differences are obtained and used as inputs in turn to construct the ANN model. The correlation coefficient and the root mean square error are used as evaluation indicators.

[0149] Table 2 provides the evaluation indicators obtained by building the ANN model using the PTB database, and Table 3 provides the evaluation indicators obtained by building the ANN model using the WCT-ECG database. The table shows the maximum value (max), minimum value (min), average value (mean) and median value (median) of the evaluation indicators. From the average and median values, it can be found that the lead reconstruction model shows better lead reconstruction results. Since both datasets contain subjects with multiple heart diseases and the signal fluctuations between different subjects are large, the minimum value of the result shows certain limitations. This limitation can be solved by building a personalized model.

[0150] Table 2 Reconstruction results of adjacent lead subtraction by ANN model in PTB database

[0151]

[0152] Table 3 Reconstruction results of adjacent lead subtraction by ANN model in WCT-ECG database

[0153]

[0154] The above method of converting standard twelve leads based on a small ECG patch uses a subset of twelve leads (precordial leads) to convert twelve-lead ECG signals. Compared with the existing technology, it has the following advantages: the precordial lead pasting position is small; the precordial lead has a clear pasting position and no additional requirements are made on the user; the precordial lead pasting position covers from the right chest to the left mid-axillary line, with more spatial three-dimensional information; the precordial lead has 6 continuous position information, which can better restore the twelve-lead information; the use of precordial lead difference can avoid the introduction of the Wilson center point.

[0155] In one embodiment, Figure 7 As shown, a method for determining an electrocardiogram signal is provided, and the method is applied to a terminal as an example for explanation, and includes the following steps:

[0156] Step S410, obtaining twelve-lead ECG data samples, and using the ECG data samples as labels for model training;

[0157] Step S411, extracting chest ECG data samples of precordial leads according to the ECG data samples, determining the difference between the signals of the precordial leads according to the chest ECG data samples, and obtaining target ECG data samples corresponding to the labels;

[0158] Step S412, training the lead reconstruction model to be trained according to the target ECG data sample and label to obtain a trained lead reconstruction model;

[0159] Step S420, obtaining the original lead reconstruction model, and obtaining the measured twelve-lead signals and the measured pre-chest lead signals of the target object;

[0160] Step S421, using the measured twelve-lead signal as a label for model training, and obtaining a target ECG data sample corresponding to the label according to the difference between the measured precordial lead signals;

[0161] Step S422, updating the original lead reconstruction model according to the target ECG data sample and label to obtain an updated lead reconstruction model;

[0162] Step S430, determining a reference lead signal from the precordial lead signal, and determining the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0163] Step S431, determining a lead signal difference between the lead signal to be processed and the reference lead signal;

[0164] Step S432: input the lead signal difference into the trained lead reconstruction model or the updated lead reconstruction model to obtain the twelve-lead ECG signal of the target object.

[0165] In a specific implementation, the terminal can use the acquired twelve-lead ECG data samples as labels, remove the limb leads from the ECG data samples, obtain chest ECG data samples of precordial leads, determine the difference between the signals of the precordial leads corresponding to the chest ECG data samples, use them as target ECG data samples, train the lead reconstruction model according to the target ECG data samples and the labels, and obtain the trained lead reconstruction model, which is a general model; the trained lead reconstruction model can also be used as the original lead reconstruction model, and the measured twelve-lead signal and the measured precordial lead signal of the target object can be obtained respectively through a twelve-lead ECG acquisition device and a small ECG patch, and the measured twelve-lead signal can be used as a new label, and a new target ECG data sample can be obtained according to the difference corresponding to the measured precordial lead signal, and the model parameters of the original lead reconstruction model can be updated according to the new target ECG data samples and the new labels to obtain an updated lead reconstruction model, which is a personalized model. When lead reconstruction is required, 6 pre-chest electrodes are used to collect 6 pre-chest potentials of the target object to obtain the pre-chest lead signal of the target object, a reference lead signal is determined from the pre-chest lead signal, the pre-chest lead signal other than the reference lead signal is subtracted from the reference lead signal to obtain the lead signal difference, and the lead signal difference is input into the trained lead reconstruction model or the updated lead reconstruction model to obtain the twelve-lead ECG signal of the target object reconstructed by the trained lead reconstruction model or the updated lead reconstruction model.

[0166] The above-mentioned ECG signal determination method determines a universal or personalized lead reconstruction model, reconstructs the lead signal difference according to the model, and obtains the twelve-lead ECG signal of the target object; since the lead reconstruction is performed using the lead signal difference, the process of calculating the WCT based on the limb leads is avoided, and the standard twelve-lead of the target object can be reconstructed by measuring a small number of chest lead signals, so that the amount of information of the ECG signal is increased while the ECG patch is fully simplified.

[0167] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0168] Based on the same inventive concept, the embodiment of the present application also provides a lead reconstruction model training device, an ECG signal determination device, and an ECG signal acquisition device for implementing the lead reconstruction model training method and ECG signal determination method involved above. The implementation scheme for solving the problem provided by the device and the equipment is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more device and equipment embodiments provided below can refer to the limitations of the lead reconstruction model training method and ECG signal determination method above, and will not be repeated here.

[0169] In one embodiment, Figure 8 As shown, an electrocardiogram signal determination device is provided, comprising: an acquisition module 510, a determination module 520, a difference module 530 and a reconstruction module 540, wherein:

[0170] An acquisition module 510 is used to acquire a precordial lead signal of a target object;

[0171] A determination module 520, configured to determine a reference lead signal from the precordial lead signal, and determine the precordial lead signal other than the reference lead signal as a lead signal to be processed;

[0172] A difference module 530, configured to determine a lead signal difference between the lead signal to be processed and the reference lead signal;

[0173] The reconstruction module 540 is used to input the lead signal difference into a trained lead reconstruction model or an updated lead reconstruction model to obtain a twelve-lead ECG signal of the target object.

[0174] In one embodiment, the acquisition module 510 is further used to acquire six pre-chest potentials obtained by collecting data from six pre-chest electrodes on the target object; and determine the six pre-chest potentials as the pre-chest lead signals of the target object.

[0175] In one embodiment, a training device for a lead reconstruction model is provided, comprising: a sample label module, a data sample module and a model training module, wherein:

[0176] A sample labeling module is used to obtain twelve-lead ECG data samples and use the ECG data samples as labels for model training;

[0177] A data sample module, used to extract a chest ECG data sample of the precordial leads according to the ECG data sample, determine the difference between the signals of the precordial leads according to the chest ECG data sample, and obtain a target ECG data sample corresponding to the label;

[0178] The model training module is used to train the lead reconstruction model to be trained according to the target ECG data sample and the label to obtain a trained lead reconstruction model.

[0179] In one embodiment, the above-mentioned data sample module is also used to determine a reference ECG data sample from the chest ECG data sample, and determine the chest ECG data sample other than the reference ECG data sample as the ECG data sample to be processed; and obtain the target ECG data sample based on the difference between the ECG data sample to be processed and the reference ECG data sample.

[0180] In one embodiment, the training device for the lead reconstruction model further includes:

[0181] A model parameter module, used to determine the model parameters of the lead reconstruction model to be trained;

[0182] The initial model module is used to generate the lead reconstruction model to be trained according to the model parameters.

[0183] In one embodiment, the lead reconstruction model to be trained includes an input layer, a hidden layer and an output layer; the number of neurons in the input layer is 5, and the number of neurons in the output layer is 12.

[0184] In one embodiment, the sample label module is further used to resample the original twelve-lead signal corresponding to the ECG data sample to obtain a resampled twelve-lead signal; filter the resampled twelve-lead signal to obtain a filtered twelve-lead signal; and normalize the filtered twelve-lead signal to obtain the twelve-lead ECG data sample.

[0185] In one embodiment, another lead reconstruction model training device is provided, comprising: a sample label module, a data sample module and a model training module, wherein:

[0186] An original model module, used to obtain an original lead reconstruction model, and to obtain a measured twelve-lead signal and a measured pre-chest lead signal of a target object; the original lead reconstruction model is obtained by training according to the lead reconstruction model training method described in any one of claims 1 to 5;

[0187] A sample updating module, used to use the measured twelve-lead signal as a label for model training, and obtain a target ECG data sample corresponding to the label according to the difference between the measured precordial lead signals;

[0188] The model updating module is used to update the original lead reconstruction model according to the target ECG data sample and the label to obtain an updated lead reconstruction model.

[0189] In one embodiment, Fig. 9 As shown, an electrocardiogram signal acquisition device is provided, including: an electrocardiogram patch 610 and a signal processor 620, wherein:

[0190] The ECG patch 610 is used to collect the precordial lead signal of the target object and send the precordial lead signal to the signal processor 620;

[0191] The signal processor 620 is used to determine a reference lead signal from the received precordial lead signal, determine the precordial lead signal other than the reference lead signal as a lead signal to be processed, determine a lead signal difference between the lead signal to be processed and the reference lead signal, input the lead signal difference into a trained lead reconstruction model or an updated lead reconstruction model, and obtain a twelve-lead ECG signal of the target object.

[0192] Each module in the lead reconstruction model training device, ECG signal determination device and ECG signal acquisition device can be implemented in whole or in part by software, hardware and their combination. Each module can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0193] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.10As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a training method for a lead reconstruction model and an electrocardiogram signal determination method are implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0194] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0195] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0196] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0197] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0198] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0199] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0200] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0201] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A training method for a lead reconstruction model, characterized in that: The method comprises: Obtaining twelve-lead ECG data samples, and using the ECG data samples as labels for model training; Extracting chest ECG data samples of precordial leads according to the ECG data samples, determining the difference between the signals of the precordial leads according to the chest ECG data samples, and obtaining target ECG data samples corresponding to the labels; The lead reconstruction model to be trained is trained according to the target ECG data sample and the label to obtain a trained lead reconstruction model.

2. The method according to claim 1, characterized in that The step of determining the difference between the signals of the pre-chest leads according to the chest ECG data sample to obtain the target ECG data sample corresponding to the label includes: Determine a reference ECG data sample from the chest ECG data samples, and determine the chest ECG data samples other than the reference ECG data sample as the ECG data samples to be processed; The target ECG data sample is obtained according to the difference between the ECG data sample to be processed and the reference ECG data sample.

3. The method according to claim 1, characterized in that Before obtaining a twelve-lead ECG data sample and using the ECG data sample as a label for model training, the method further includes: Determining model parameters of the lead reconstruction model to be trained; The lead reconstruction model to be trained is generated according to the model parameters.

4. The method according to claim 1, characterized in that: The lead reconstruction model to be trained includes an input layer, a hidden layer and an output layer; the number of neurons in the input layer is 5, and the number of neurons in the output layer is 12.

5. The method according to claim 1, characterized in that The step of obtaining twelve-lead ECG data samples includes: Resampling the original twelve-lead signal corresponding to the electrocardiogram data sample to obtain a resampled twelve-lead signal; filtering the resampled twelve-lead signal to obtain a filtered twelve-lead signal; The filtered twelve-lead signals are normalized to obtain the twelve-lead ECG data samples.

6. A method for training a lead reconstruction model, characterized in that: The method comprises: Acquire an original lead reconstruction model, and acquire measured twelve-lead signals and measured pre-chest lead signals of a target object; the original lead reconstruction model is obtained by training according to the lead reconstruction model training method described in any one of claims 1 to 5; Using the measured twelve-lead signals as labels for model training, and obtaining target ECG data samples corresponding to the labels according to the differences between the measured pre-chest lead signals; The original lead reconstruction model is updated according to the target ECG data sample and the label to obtain an updated lead reconstruction model.

7. A method for determining an electrocardiogram signal, characterized in that: The method comprises: Acquire chest lead signals of the target object; Determining a reference lead signal from the precordial lead signal, and determining the precordial lead signal other than the reference lead signal as a lead signal to be processed; determining a lead signal difference between the lead signal to be processed and the reference lead signal; The lead signal difference is input into the trained lead reconstruction model according to any one of claims 1 to 5 or the updated lead reconstruction model according to claim 6 to obtain the twelve-lead electrocardiogram signal of the target object.

8. The method according to claim 7, characterized in that The step of obtaining the chest lead signal of the target object includes: Acquiring six pre-chest potentials obtained by collecting data from six pre-chest electrodes on the target object; The six precordial potentials are determined as the precordial lead signals of the target object.

9. An electrocardiogram signal determination device, characterized in that: The device comprises: An acquisition module, used for acquiring a chest lead signal of a target object; A determination module, used to determine a reference lead signal from the precordial lead signal, and determine the precordial lead signal other than the reference lead signal as a lead signal to be processed; A difference module, used to determine a lead signal difference between the lead signal to be processed and the reference lead signal; A reconstruction module is used to input the lead signal difference into the trained lead reconstruction model according to any one of claims 1 to 5 or the updated lead reconstruction model according to claim 6 to obtain the twelve-lead ECG signal of the target object.

10. An electrocardiogram signal acquisition device, characterized in that: The electrocardiogram signal acquisition device comprises: An ECG patch, used to collect the precordial lead signal of the target object and send the precordial lead signal to a signal processor; A signal processor, for determining a reference lead signal from the received precordial lead signal, determining the precordial lead signal other than the reference lead signal as a lead signal to be processed, determining a lead signal difference between the lead signal to be processed and the reference lead signal, inputting the lead signal difference into the trained lead reconstruction model as described in any one of claims 1 to 5 or the updated lead reconstruction model as described in claim 6, to obtain a twelve-lead electrocardiogram signal of the target object.

Images