Multi-lead electrocardiosignal reconstruction method, device, equipment and medium

Abstract

The invention provides a multi-lead electrocardiosignal reconstruction method and device, equipment and a medium, and the method comprises the steps: carrying out the preprocessing of a single-lead electrocardiosignal, inputting the single-lead electrocardiosignal into a plurality of conditional generative adversarial network generators, and generating a plurality of target lead electrocardiosignals; wherein each conditional generative adversarial network generator corresponds to one target lead, and the target leads comprise at least one of a lead II, a lead III, aVR, aVL, aVF and chest leads V1 to V6; and generating a complete multi-lead electrocardiosignal based on the single-lead electrocardiosignal and the plurality of target lead electrocardiosignals. The 12-lead electrocardiogram containing all chest leads and limb leads is generated end to end only by taking the lead I signal as the input, the dependence of a traditional method on multi-lead input or historical data is broken through, the clinical applicability of single-lead wearable equipment is greatly improved, and the accuracy of multi-lead electrocardiosignal reconstruction is improved.

Description

Technical Field

[0001] This application relates to the field of medical signal processing technology, and in particular to a method, apparatus, device and medium for reconstructing multi-lead electrocardiogram signals. Background Technology

[0002] The standard 12-lead electrocardiogram (ECG) is widely used in clinical diagnosis because it comprehensively reflects the potential changes of the heart in different spatial directions. However, traditional 12-lead ECG acquisition requires attaching electrodes to multiple locations on the patient's limbs and chest wall, which is complex, time-consuming, and usually only performed by professionals in medical institutions, making it difficult to support long-term continuous monitoring. To compensate for the insufficient information of single-lead devices, researchers have attempted to derive more lead signals from a limited number of leads through lead conversion or signal reconstruction methods. Currently, most deep learning lead reconstruction methods require simultaneous measurement data from at least three leads, or require sequential measurement of multiple leads combined with the patient's historical ECG data for reconstruction. This is inconsistent with the reality that single-lead devices such as smartwatches can only measure lead I in real time. Furthermore, schemes that use lead II as input cannot be directly applied to smart devices that primarily measure lead I, limiting the practicality of the technology. Therefore, there is an urgent need for a technical solution that can efficiently and accurately reconstruct a complete 12-lead ECG. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method, device, equipment and medium for reconstructing multi-lead electrocardiogram signals, which generates a 12-lead electrocardiogram including all chest leads and limb leads end-to-end using only lead I signal as input. This breaks through the dependence of traditional methods on multi-lead input or historical data, greatly improves the clinical applicability of single-lead wearable devices and improves the accuracy of multi-lead electrocardiogram signal reconstruction.

[0004] This application provides a method for reconstructing multi-lead electrocardiogram signals, the reconstruction method comprising: The acquired single-lead ECG signal was preprocessed to determine the preprocessed single-lead ECG signal; The preprocessed single-lead ECG signal is input into multiple pre-trained conditional generative adversarial network generators to generate multiple target lead ECG signals; wherein, each conditional generative adversarial network generator corresponds to one target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF and chest leads V1 to V6. Based on the single-lead ECG signal and multiple target lead ECG signals, a complete multi-lead ECG signal is generated.

[0005] In one possible implementation, the conditional generative adversarial network generator is determined through the following steps: Obtain a training sample set containing standard lead electrocardiograms, where each record is labeled with the sample lead I signal and the actual measured signals of multiple leads; For each target lead, the sample lead I signal is used as the input signal, and the real ECG signal of that target lead is used as the supervision signal to construct an input-output paired sample. An initial generator and a conditional discriminator, employing an encoder and decoder structure, constitute an initial generative adversarial network. By alternately optimizing the initial generator and the conditional discriminator, the joint loss function is minimized until the model converges, resulting in a trained conditional generative adversarial network generator. The joint loss function includes generative adversarial network loss and L1 reconstruction loss.

[0006] In one possible implementation, the conditional generative adversarial network loss is: ; in, G For the initial generator, D For the condition discriminator, x For the sample lead I signal, y This represents the actual lead ECG signal. z It is random noise. E For mathematical expectation calculation, Conditional generation of adversarial network loss.

[0007] In one possible implementation, the initial generator employing an encoder and decoder structure and the conditional discriminator constitute an initial generative adversarial network. By alternately optimizing the initial generator and the conditional discriminator to minimize the joint loss function until the model converges, a trained conditional generative adversarial network generator is obtained, comprising: The initial generator performs multi-level downsampling on the input signal through an encoder to extract multi-level feature representations, and then uses a decoder combined with a skip connection structure to gradually upsample and reconstruct the signal, generating the corresponding predictive lead ECG signal. The predicted lead ECG signal, the monitoring signal, and the input signal are input into the condition discriminator to determine the probability value of whether the pairing between the predicted lead ECG signal and the input signal is a true physiological pairing. Based on the feedback results of the conditional discriminator and the joint loss function, the network parameters of the initial generator and the conditional discriminator are updated synchronously, so that the initial generator gradually learns the nonlinear physiological mapping relationship between a single lead and the target lead, until the model converges to generate the conditional generative adversarial network generator.

[0008] In one possible implementation, after obtaining the trained conditional generative adversarial network generator, the reconstruction method further includes: Train and deploy a conditional generative adversarial network generator that generates target limb lead signals from a single lead I; wherein the target limb lead is either lead II or lead III. For the target limb, which is lead III, the Einthoven triangle law is used to generate the electrocardiogram signals corresponding to leads II, aVR, aVL and aVF based on the sample lead I signal and the target limb lead signal. For the chest leads V1 to V6, independent conditional generative adversarial network generators are trained respectively.

[0009] In one possible implementation, the step of generating the lead ECG signals corresponding to leads II, aVR, aVL, and aVF based on the single-lead ECG signal and the target limb lead signal using Einthoven's triangle law includes: Based on the sum of the target limb lead signal and the sample lead I signal, the ECG signal of lead II is determined. The sum of the ECG signals in lead I and lead II is determined, and the sum is averaged to determine the aVR ECG signal. The first quotient is determined by dividing the ECG signal in lead II by the target value. The aVL ECG signal is determined by subtracting the first quotient from the sample lead I signal. The target value is 2. The second quotient is determined by dividing the sample lead I signal by the target value. The aVF electrocardiogram signal is determined by subtracting the second quotient from the lead II electrocardiogram signal.

[0010] In one possible implementation, after generating the complete multi-lead ECG signal, the reconstruction method further includes: The multi-lead ECG signal is input into a trained deep neural network classification model. Spatiotemporal features are extracted through multi-layer residual blocks, and the detection and analysis results are determined through a fully connected layer and a Softmax function. The deep neural network classification model is a one-dimensional convolutional neural network based on a residual structure.

[0011] This application embodiment also provides a reconstruction device for multi-lead electrocardiogram signals, the reconstruction device comprising: The preprocessing module is used to preprocess the acquired single-lead ECG signal and determine the preprocessed single-lead ECG signal. The lead generation module is used to input the preprocessed single-lead ECG signal into multiple pre-trained conditional generative adversarial network generators to generate multiple target lead ECG signals; wherein, each conditional generative adversarial network generator corresponds to a target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF and chest leads V1 to V6. The module is used to generate a complete multi-lead ECG signal based on the single-lead ECG signal and multiple target lead ECG signals.

[0012] This application also provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the multi-lead electrocardiogram signal reconstruction method described above are performed.

[0013] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the multi-lead electrocardiogram signal reconstruction method described above.

[0014] This application provides a method, apparatus, device, and medium for reconstructing multi-lead electrocardiogram (ECG) signals. The reconstruction method includes: preprocessing the acquired single-lead ECG signal to determine the preprocessed single-lead ECG signal; inputting the preprocessed single-lead ECG signal into multiple pre-trained conditional generative adversarial network (GAN) generators to generate multiple target lead ECG signals; wherein each GAN generator corresponds to one target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF, and chest leads V1 to V6; and generating a complete multi-lead ECG signal based on the single-lead ECG signal and the multiple target lead ECG signals. This method generates a 12-lead ECG containing all chest and limb leads end-to-end using only the lead I signal as input, overcoming the dependence of traditional methods on multi-lead input or historical data, greatly improving the clinical applicability of single-lead wearable devices, and enhancing the accuracy of multi-lead ECG signal reconstruction.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a method for reconstructing multi-lead electrocardiogram signals provided in an embodiment of this application; Figure 2 One of the structural schematic diagrams of a multi-lead electrocardiogram signal reconstruction device provided in the embodiments of this application; Figure 3 A second schematic diagram of a multi-lead electrocardiogram signal reconstruction device provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0019] First, the applicable scenarios for this application will be introduced. This application can be applied to the field of medical signal processing technology.

[0020] Research has shown that traditional 12-lead electrocardiogram (ECG) acquisition requires attaching electrodes to multiple locations on the patient's limbs and chest wall, a complex and time-consuming process typically performed only by professionals in medical institutions, making it difficult to support long-term continuous monitoring. To compensate for the limited information from single-lead devices, researchers have attempted to derive signals from more leads from a limited number of leads through lead conversion or signal reconstruction methods. Currently, most deep learning lead reconstruction methods require simultaneous measurement data from at least three leads, or sequential measurement of multiple leads combined with the patient's historical ECG data for reconstruction. This is inconsistent with the reality that single-lead devices such as smartwatches can only measure lead I in real time. Furthermore, schemes using lead II as input cannot be directly applied to smart devices that primarily measure lead I, limiting the technology's practicality. Therefore, a technical solution capable of efficiently and accurately reconstructing a complete 12-lead ECG is urgently needed.

[0021] Based on this, the embodiments of this application provide a method for reconstructing multi-lead electrocardiogram signals, which generates a 12-lead electrocardiogram containing all chest and limb leads end-to-end using only lead I signal as input. This breaks through the dependence of traditional methods on multi-lead input or historical data, greatly improves the clinical applicability of single-lead wearable devices, and improves the accuracy of multi-lead electrocardiogram signal reconstruction.

[0022] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for reconstructing multi-lead electrocardiogram signals provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the reconstruction method includes: S101: Preprocess the acquired single-lead ECG signal to determine the preprocessed single-lead ECG signal. This step involves acquiring and preprocessing lead I ECG signals from a single-lead wearable device. Specifically, lead I signals are acquired using a single-lead device such as an ECG patch at a sampling rate of 250Hz. The acquired continuous ECG signal is segmented into 2.5-second segments. Amplitude normalization is performed on the segmented signals, mapping the signal values ​​to the [-1, 1] interval to form one-dimensional time series data. The preprocessing does not perform traditional signal processing operations such as filtering or baseline correction, preserving all characteristic information of the original signal.

[0023] It should be noted that the single-lead ECG signal is the lead I ECG signal.

[0024] The preprocessing includes baseline drift correction for all lead signals using a zero-phase second-order high-pass filter (cutoff frequency 0.5Hz); removal of power line interference using a notch filter (50 / 60Hz); elimination of electromyographic noise using wavelet denoising or adaptive filtering methods; and signal normalization, such as Z-score normalization or maximum value normalization by channel, to improve model training stability. Time segmentation involves dividing the continuous ECG signal into non-overlapping or sliding segments according to fixed duration windows (e.g., 5 seconds, 10 seconds) to obtain multiple time-aligned ECG segments. Each segment contains a sequence of synchronous signal values ​​from lead I and other leads within the same time period.

[0025] S102: The preprocessed single-lead ECG signal is input into multiple pre-trained conditional generative adversarial network generators to generate multiple target lead ECG signals; wherein, each conditional generative adversarial network generator corresponds to a target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF and chest leads V1 to V6.

[0026] In this step, the preprocessed single-lead ECG signal is input into multiple pre-trained conditional generative adversarial network generators to generate multiple target lead ECG signals.

[0027] It should be noted that during the generation of the target lead ECG signal, Einthoven's triangle law is used to generate the ECG signals corresponding to leads II, aVR, aVL, and aVF based on the lead I signal and the target lead III signal. Alternatively, the ECG signals corresponding to leads III, aVR, aVL, and aVF are generated based on the lead I signal and the target lead II signal. The precordial leads V1 to V6 are generated using a corresponding conditional generative adversarial network generator.

[0028] In one possible implementation, the conditional generative adversarial network generator is determined through the following steps: A: Obtain a training sample set containing standard lead electrocardiograms, where each record is labeled with the sample lead I signal and the actual measurement signals of multiple leads.

[0029] Here, a raw electrocardiogram (ECG) signal dataset containing 12-lead simultaneous recordings is acquired, collected under different physiological states (including normal rhythm, arrhythmia, ischemic changes, etc.) to ensure data diversity and representativeness. Each record includes time-aligned voltage signals from lead I and all other target leads, with a sampling frequency of no less than 250Hz and an analog-to-digital conversion accuracy of no less than 12 bits.

[0030] B: For each target lead, the sample lead I signal is used as the input signal, and the actual ECG signal of the target lead is used as the supervision signal to construct an input-output paired sample.

[0031] Here, independent paired sample sets are constructed for different target leads. For example, all samples with lead I as input and lead II as output form a subset; similarly, corresponding subsets are constructed for leads III, aVF, V1, etc. These subsets can be used to jointly train multi-task models or to train single-target dedicated models separately.

[0032] C: An initial generator and the conditional discriminator, using an encoder and decoder structure, constitute an initial generative adversarial network. By alternately optimizing the initial generator and the conditional discriminator, the joint loss function is minimized until the model converges, resulting in a trained conditional generative adversarial network generator. The joint loss function includes generative adversarial network loss and L1 reconstruction loss.

[0033] Here, a deep learning architecture based on Conditional Generative Adversarial Network (ConditionalGAN) is adopted to generate signals for the remaining 11 target leads from a single-lead I ECG signal. This architecture consists of an initial generator and a conditional discriminator forming the initial GAN, which is optimized through an adversarial joint training mechanism to finally obtain the trained Conditional GAN ​​generator.

[0034] In one possible implementation, the conditional generative adversarial network loss is: ; in, G For the initial generator, D For the condition discriminator, x For the sample lead I signal, y This represents the actual lead ECG signal. z It is random noise. E For mathematical expectation calculation, Conditional generation of adversarial network loss.

[0035] Here, the loss function contains two terms: the first term The discriminator is measured by its pairing with the real signal. The discriminative ability of the conditional discriminator is maximized by maximizing this term, that is, outputting a probability value close to 1 for true pairings, so that... Close to 0; second term The ability of a conditional discriminator to distinguish between pairs of generated signals is measured, where This indicates that the generator generates the target lead based on the lead I signal x and the noise z. The discriminator maximizes this term, that is, the probability value of generating a paired output close to 0, so that... It is close to 0. From the generator's perspective, its optimization objective is to minimize this loss function, that is, to minimize the second term. This is equivalent to maximizing This causes the discriminator to output a probability value close to 1 for generated pairs, thus "tricking" the discriminator into believing that the generated signal pairs are real. From the discriminator's perspective, its optimization objective is to maximize this loss function, that is, to simultaneously improve the recognition rate of real pairs and the rejection rate of generated pairs.

[0036] The L1 reconstruction loss function is:

[0037] in, Let represent the L1 norm, which is calculated as the sum of the absolute values ​​of the differences between corresponding sampling points of two signals, and y represent the actual target lead signal. This represents the generated predicted lead ECG signal. For a signal containing N sampling points, the L1 loss is calculated as follows:

[0038] This loss function ensures that the generated signal is not only realistic in its overall distribution but also accurate in its amplitude at each time point. The final optimization objective is to combine the two loss functions with weights, balancing the requirements of adversarial training and accurate reconstruction through weight coefficients. The choice of weight coefficients requires a trade-off: too high an adversarial loss weight may result in an overly diverse but inaccurate generated signal, while too high a reconstruction loss weight may result in an overly conservative and unrealistic generated signal.

[0039] In this application, a dual constraint mechanism combining Conditional Generative Adversarial Network (GAN) and L1 pixel-level reconstruction loss is employed to ensure that the generated signal is not only realistic in overall distribution and waveform morphology, but also highly accurate in key diagnostic feature points (such as QRS complex width and amplitude, ST segment shift trend, T wave direction, etc.). Specifically, the conditional discriminator enforces physiological consistency across leads by judging whether "lead I and the target lead are from the same real measurement at the same time"; while the L1 loss effectively suppresses blurring and preserves signal details.

[0040] In one possible implementation, the initial generator employing an encoder and decoder structure and the conditional discriminator constitute an initial generative adversarial network. By alternately optimizing the initial generator and the conditional discriminator to minimize the joint loss function until the model converges, a trained conditional generative adversarial network generator is obtained, comprising: a: The initial generator performs multi-level downsampling on the input signal through the encoder, extracts multi-level feature representations, and then uses the decoder combined with the skip connection structure to gradually upsample and reconstruct the signal, generating the corresponding predictive lead ECG signal.

[0041] Here, the initial generator adopts a U-Net-type encoder-decoder structure to learn the nonlinear mapping relationship between the input lead I and the target lead. The encoder part contains 7 convolutional layers, each using a 4×1 convolutional kernel (one-dimensional convolution along the time axis) with a stride of 2, to progressively downsample the input signal and extract multi-scale feature representations. Except for the first layer, each convolutional layer is followed by a batch normalization layer and a ReLU activation function (slope of 0.2) to enhance model stability and nonlinear expressive power. As the network depth increases, the number of feature map channels doubles layer by layer, forming a semantic feature pyramid from coarse to fine. The decoder part correspondingly contains 7 deconvolutional layers (or transposed convolutional layers), also using a 4×1 convolutional kernel with a stride of 2, to progressively upsample the low-dimensional latent space features to restore them to the original temporal length; each layer is followed by a batch normalization layer and a ReLU activation function. Crucially, the decoder incorporates U-Net's unique skip connection mechanism, concatenating the feature map output from the encoder's i-th layer with the input from the decoder's (8-i)-th layer along the channel dimension. This preserves high-resolution spatial detail information, which is beneficial for reconstructing key waveform structures such as the QRS complex, ST segment, and T wave. Ultimately, the decoder generates the predicted lead ECG signal.

[0042] b: Input the predicted lead ECG signal, the monitoring signal, and the input signal into the condition discriminator to determine the probability value of whether the pairing between the predicted lead ECG signal and the input signal is a true physiological pairing.

[0043] Here, the input to the conditional discriminator is a pair of ECG signals: one is the original acquired lead I signal x as the conditional input; the other is the predicted lead ECG signal y' to be discriminated, which is the synthesized signal output by the generator G(x) in the sample generation path. The discriminator's task is to determine whether the signal pair (x, y') is a real synchronous measurement data pair from the same patient at the same time, that is, to assess its physiological consistency and pairing authenticity. In specific implementation, the discriminator first concatenates the lead I signal x and the predicted lead ECG signal y' in the channel dimension to form a dual-channel one-dimensional time series input. Then, it is processed through multiple convolutional layers, each layer using a 4×1 convolutional kernel with a stride of 2, progressively extracting joint feature representations; except for the first layer, subsequent convolutional layers are configured with batch normalization layers and LeakyReLU activation functions. The network does not need skip connections because its task is global true / false discrimination rather than fine reconstruction. Finally, the high-dimensional features are compressed into a fixed-length vector through a global average pooling layer, and then passed through a fully connected layer and a sigmoid activation function to output a probability value D(x,y')∈[0,1], which represents the probability that the discriminator judges the signal pair as a real pair. A probability value close to 1 indicates that the discriminator considers it a real measured pair signal, while a probability value close to 0 indicates that the discriminator considers it a pair signal forged by the generator.

[0044] c: Based on the feedback results of the conditional discriminator and the joint loss function, the network parameters of the initial generator and the conditional discriminator are updated synchronously, so that the initial generator gradually learns the nonlinear physiological mapping relationship between a single lead and the target lead, until the model converges to generate the conditional generative adversarial network generator.

[0045] Here, the feedback results of the conditional discriminator and the joint loss function are used to synchronously update the network parameters of the initial generator and the conditional discriminator, so that the initial generator gradually learns the nonlinear physiological mapping relationship between a single lead and the target lead, until the model converges to generate the conditional generative adversarial network generator.

[0046] In each iteration, the initial generator parameters are first fixed, and the discriminator parameters are updated using real and generated samples to improve the discriminator's ability to distinguish between different targets. Then, the discriminator parameters are fixed, and the generator parameters are updated to improve the generation quality. This alternating optimization ensures that the generator and discriminator maintain a dynamic balance. The training process continues until the model converges, that is, the generator loss and discriminator loss tend to stabilize, and the quality of the generated signal no longer improves significantly.

[0047] In one possible implementation, after obtaining the trained conditional generative adversarial network generator, the reconstruction method further includes: (1): Train and deploy a conditional generative adversarial network generator that generates target limb lead signals from a single lead I; wherein the target limb lead is either lead II or lead III.

[0048] It should be noted that the training process of the conditional generative adversarial network generator for the target limb lead signal is consistent with the training process of the conditional generative adversarial network generator described above, and will not be repeated here.

[0049] (2): For the target limb, which is lead III, the ECG signals corresponding to leads II, aVR, aVL, and aVF are generated based on the sample lead I signal and the target limb lead signal using Einthoven's triangle law. For the chest leads V1 to V6, independent conditional generative adversarial network generators are trained respectively.

[0050] It should be noted that, for the target limb, which is lead II, the ECG signals corresponding to leads III, aVR, aVL, and aVF are generated based on the sample lead I signal and the target limb lead signal using Einthoven's triangle law.

[0051] In one possible implementation, the step of generating the lead ECG signals corresponding to leads II, aVR, aVL, and aVF based on the single-lead ECG signal and the target limb lead signal using Einthoven's triangle law includes: The ECG signal of lead II is determined by adding the target limb lead signal and the sample lead I signal; the sum of the sample lead I signal and the ECG signal of lead II is determined, and the sum is averaged to determine the aVR ECG signal; the ECG signal of lead II is determined by dividing the target value by the target value, and the ECG signal of aVL is determined by subtracting the first quotient from the sample lead I signal; wherein the target value is 2; the ECG signal of aVF is determined by dividing the sample lead I signal by the target value.

[0052] This section introduces an optimized training strategy based on Einthoven's triangle law from electrophysiology. This strategy is grounded in the physiological principle that a definite algebraic relationship exists between standard limb leads, improving computational efficiency by reducing the number of generative models that need to be trained. Only the generated lead II or lead III is used to calculate the remaining limb leads using the following formula, thus reducing computational complexity.

[0053]

[0054]

[0055]

[0056]

[0057] Among them, LeadII is the ECG signal of lead II, LeadI is the ECG signal of sample lead I, LeadIII is the ECG signal of lead III, aVR is the ECG signal of lead II, aVL is the ECG signal of lead I, and aVF is the ECG signal of lead III.

[0058] Unlike traditional methods that require independently training 11 generators to reconstruct each target lead, this application utilizes Einthoven's triangle law. By training a generator for a key limb lead (such as lead II) and using linear algebraic relationships to derive the remaining limb leads (such as lead III, aVR, aVL, and aVF), the number of generative models required for training is reduced from 11 to 7 (i.e., only lead II and V1–V6 need to be modeled independently). This optimization strategy significantly reduces model training costs, storage overhead, and inference latency without sacrificing diagnostic performance, and significantly improves the feasibility of deploying the system on edge devices (such as mobile phones and wearable terminals), which is conducive to promoting the transformation of technology into real products.

[0059] S103: Based on the single-lead ECG signal and multiple target lead ECG signals, generate a complete multi-lead ECG signal.

[0060] In this step, the single-lead ECG signal is combined with multiple target lead ECG signals to generate a complete multi-lead ECG signal.

[0061] In one possible implementation, after generating the complete multi-lead ECG signal, the reconstruction method further includes: The multi-lead ECG signal is input into a trained deep neural network classification model. Spatiotemporal features are extracted through multi-layer residual blocks, and the detection and analysis results are determined through a fully connected layer and a Softmax function. The deep neural network classification model is a one-dimensional convolutional neural network based on a residual structure.

[0062] Here, the deep neural network classification model is trained using standard supervised learning methods. The cross-entropy loss function is used to measure the difference between the predicted results and the true labels, and the network parameters are optimized through backpropagation and gradient descent. The dataset is divided into training, validation, and test sets according to a standard ratio. The training set is used for parameter learning, the validation set for hyperparameter tuning and early stopping strategies, and the test set for final performance evaluation. Data augmentation techniques (such as random pruning and amplitude perturbation) are used during training to improve the model's generalization ability and prevent overfitting.

[0063] This application provides a method for reconstructing multi-lead electrocardiogram (ECG) signals. The method includes: preprocessing the acquired single-lead ECG signal to determine the preprocessed single-lead ECG signal; inputting the preprocessed single-lead ECG signal into multiple pre-trained conditional generative adversarial network (GAN) generators to generate multiple target lead ECG signals; wherein each GAN generator corresponds to one target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF, and chest leads V1 to V6; and generating a complete multi-lead ECG signal based on the single-lead ECG signal and the multiple target lead ECG signals. This method generates a 12-lead ECG containing all chest and limb leads end-to-end using only the lead I signal as input, overcoming the dependence of traditional methods on multi-lead input or historical data, greatly improving the clinical applicability of single-lead wearable devices, and enhancing the accuracy of multi-lead ECG signal reconstruction.

[0064] Please see Figure 2 , Figure 3 , Figure 2 One of the structural schematic diagrams of a multi-lead electrocardiogram signal reconstruction device provided in the embodiments of this application; Figure 3 This is a second schematic diagram of a multi-lead electrocardiogram signal reconstruction device provided in an embodiment of this application. Figure 2 As shown, the reconstruction device 200 includes: The preprocessing module 210 is used to preprocess the acquired single-lead ECG signal and determine the preprocessed single-lead ECG signal. The lead generation module 220 is used to input the preprocessed single-lead ECG signal into multiple pre-trained conditional generative adversarial network generators to generate multiple target lead ECG signals; wherein, each conditional generative adversarial network generator corresponds to a target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF and chest leads V1 to V6. The assembly module 230 is used to generate a complete multi-lead electrocardiogram signal based on the single-lead electrocardiogram signal and multiple target lead electrocardiogram signals.

[0065] Furthermore, such as Figure 3 As shown, the reconstruction device 200 also includes a training module 240, which determines the conditional generative adversarial network generator through the following steps: Obtain a training sample set containing standard lead electrocardiograms, where each record is labeled with the sample lead I signal and the actual measured signals of multiple leads; For each target lead, the sample lead I signal is used as the input signal, and the real ECG signal of that target lead is used as the supervision signal to construct an input-output paired sample. An initial generator and a conditional discriminator, employing an encoder and decoder structure, constitute an initial generative adversarial network. By alternately optimizing the initial generator and the conditional discriminator, the joint loss function is minimized until the model converges, resulting in a trained conditional generative adversarial network generator. The joint loss function includes generative adversarial network loss and L1 reconstruction loss.

[0066] Furthermore, the training module 240 is used to construct an initial generative adversarial network (GAN) using the initial generator with the encoder and decoder structure and the conditional discriminator. By alternately optimizing the initial generator and the conditional discriminator, the joint loss function is minimized until the model converges, thus obtaining the trained conditional GAN ​​generator. The initial generator performs multi-level downsampling on the input signal through an encoder to extract multi-level feature representations, and then uses a decoder combined with a skip connection structure to gradually upsample and reconstruct the signal, generating the corresponding predictive lead ECG signal. The predicted lead ECG signal, the monitoring signal, and the input signal are input into the condition discriminator to determine the probability value of whether the pairing between the predicted lead ECG signal and the input signal is a true physiological pairing. Based on the feedback results of the conditional discriminator and the joint loss function, the network parameters of the initial generator and the conditional discriminator are updated synchronously, so that the initial generator gradually learns the nonlinear physiological mapping relationship between a single lead and the target lead, until the model converges to generate the conditional generative adversarial network generator.

[0067] Furthermore, the training module 240 is also used for: Train and deploy a conditional generative adversarial network generator that generates target limb lead signals from a single lead I; wherein the target limb lead is either lead II or lead III. For the target limb, which is lead III, the Einthoven triangle law is used to generate the electrocardiogram signals corresponding to leads II, aVR, aVL and aVF based on the sample lead I signal and the target limb lead signal. For the chest leads V1 to V6, independent conditional generative adversarial network generators are trained respectively.

[0068] Furthermore, the training module 240 is used to generate the lead ECG signals corresponding to leads II, aVR, aVL, and aVF based on the single-lead ECG signal and the target limb lead signal, using Einthoven's triangle law. Based on the sum of the target limb lead signal and the sample lead I signal, the ECG signal of lead II is determined. The sum of the ECG signals in lead I and lead II is determined, and the sum is averaged to determine the aVR ECG signal. The first quotient is determined by dividing the ECG signal in lead II by the target value. The aVL ECG signal is determined by subtracting the first quotient from the sample lead I signal. The target value is 2. The second quotient is determined by dividing the sample lead I signal by the target value. The aVF electrocardiogram signal is determined by subtracting the second quotient from the lead II electrocardiogram signal.

[0069] Furthermore, such as Figure 3 As shown, the reconstruction device 200 also includes a classification module 250, which is used for: The multi-lead ECG signal is input into a trained deep neural network classification model. Spatiotemporal features are extracted through multi-layer residual blocks, and the detection and analysis results are determined through a fully connected layer and a Softmax function. The deep neural network classification model is a one-dimensional convolutional neural network based on a residual structure.

[0070] This application provides a multi-lead ECG signal reconstruction device, comprising: a preprocessing module for preprocessing acquired single-lead ECG signals to determine preprocessed single-lead ECG signals; a lead generation module for inputting the preprocessed single-lead ECG signals into multiple pre-trained conditional generative adversarial network (GAN) generators to generate multiple target lead ECG signals; wherein each GAN generator corresponds to one target lead, and the target leads include at least one of lead II, lead III, aVR, aVL, aVF, and chest leads V1 to V6; and a component module for generating a complete multi-lead ECG signal based on the single-lead ECG signals and the multiple target lead ECG signals. This device generates a 12-lead ECG containing all chest and limb leads end-to-end using only lead I as input, overcoming the dependence of traditional methods on multi-lead input or historical data, greatly improving the clinical applicability of single-lead wearable devices, and enhancing the accuracy of multi-lead ECG signal reconstruction.

[0071] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

[0072] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, they can perform the operations described above. Figure 1 The steps of the multi-lead ECG signal reconstruction method in the illustrated method embodiment can be found in the method embodiment for specific implementation, and will not be repeated here.

[0073] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of the multi-lead ECG signal reconstruction method in the illustrated method embodiment can be found in the method embodiment for specific implementation, and will not be repeated here.

[0074] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0076] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0077] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0078] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0079] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method of reconstruction of a multi-lead electrocardiogram signal, characterized in that, The reconstruction method includes: The acquired single-lead ECG signal was preprocessed to determine the preprocessed single-lead ECG signal; The preprocessed single-lead ECG signal is input into multiple pre-trained conditional generative adversarial network generators to generate multiple target lead ECG signals; wherein, each conditional generative adversarial network generator corresponds to one target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF and chest leads V1 to V6. Based on the single-lead ECG signal and multiple target lead ECG signals, a complete multi-lead ECG signal is generated.

2. The reconstruction method of claim 1, wherein, The conditional generative adversarial network generator is determined through the following steps: Obtain a training sample set containing standard lead electrocardiograms, where each record is labeled with the sample lead I signal and the actual measured signals of multiple leads; For each target lead, the sample lead I signal is used as the input signal, and the real ECG signal of that target lead is used as the supervision signal to construct an input-output paired sample. An initial generator and a conditional discriminator, employing an encoder and decoder structure, constitute an initial generative adversarial network. By alternately optimizing the initial generator and the conditional discriminator, the joint loss function is minimized until the model converges, resulting in a trained conditional generative adversarial network generator. The joint loss function includes generative adversarial network loss and L1 reconstruction loss.

3. The reconstruction method according to claim 2, characterized in that, The loss of the conditional generative adversarial network is: ； in, G For the initial generator, D For the condition discriminator, x For the sample lead I signal, y This represents the actual lead ECG signal. z It is random noise. E For mathematical expectation calculation, Conditional generation of adversarial network loss.

4. The reconstruction method according to claim 2, characterized in that, The initial generator employing an encoder and decoder structure, along with the conditional discriminator, constitutes an initial generative adversarial network. By alternately optimizing the initial generator and the conditional discriminator, the joint loss function is minimized until the model converges, resulting in a trained conditional generative adversarial network generator, including: The initial generator performs multi-level downsampling on the input signal through an encoder to extract multi-level feature representations, and then uses a decoder combined with a skip connection structure to gradually upsample and reconstruct the signal, generating the corresponding predictive lead ECG signal. The predicted lead ECG signal, the monitoring signal, and the input signal are input into the condition discriminator to determine the probability value of whether the pairing between the predicted lead ECG signal and the input signal is a true physiological pairing. Based on the feedback results of the conditional discriminator and the joint loss function, the network parameters of the initial generator and the conditional discriminator are updated synchronously, so that the initial generator gradually learns the nonlinear physiological mapping relationship between a single lead and the target lead, until the model converges to generate the conditional generative adversarial network generator.

5. The reconstruction method according to claim 2, characterized in that, After obtaining the trained conditional generative adversarial network generator, the reconstruction method further includes: Train and deploy a conditional generative adversarial network generator that generates target limb lead signals from a single lead I; wherein the target limb lead is either lead II or lead III. For the target limb, which is lead III, the Einthoven triangle law is used to generate the electrocardiogram signals corresponding to leads II, aVR, aVL and aVF based on the sample lead I signal and the target limb lead signal. For the chest leads V1 to V6, independent conditional generative adversarial network generators are trained respectively.

6. The reconstruction method according to claim 5, characterized in that, The method utilizes Einthoven's triangle law to generate ECG signals corresponding to leads II, aVR, aVL, and aVF based on single-lead ECG signals and target limb lead signals, including: Based on the sum of the target limb lead signal and the sample lead I signal, the ECG signal of lead II is determined. The sum of the ECG signals in lead I and lead II is determined, and the sum is averaged to determine the aVR ECG signal. The first quotient is determined by dividing the ECG signal in lead II by the target value. The aVL ECG signal is determined by subtracting the first quotient from the sample lead I signal. The target value is 2. The second quotient is determined by dividing the sample lead I signal by the target value. The aVF ECG signal is then determined by subtracting the second quotient from the lead II ECG signal.

7. The reconstruction method according to claim 1, characterized in that, After generating the complete multi-lead ECG signal, the reconstruction method further includes: The multi-lead ECG signal is input into a trained deep neural network classification model. Spatiotemporal features are extracted through multi-layer residual blocks, and the detection and analysis results are determined through a fully connected layer and a Softmax function. The deep neural network classification model is a one-dimensional convolutional neural network based on a residual structure.

8. A device for reconstructing multi-lead electrocardiogram signals, characterized in that, The reconstruction device includes: The preprocessing module is used to preprocess the acquired single-lead ECG signal and determine the preprocessed single-lead ECG signal. The lead generation module is used to input the preprocessed single-lead ECG signal into multiple pre-trained conditional generative adversarial network generators to generate multiple target lead ECG signals; wherein, each conditional generative adversarial network generator corresponds to a target lead, and the target lead includes at least one of lead II, lead III, aVR, aVL, aVF and chest leads V1 to V6. The module is used to generate a complete multi-lead ECG signal based on the single-lead ECG signal and multiple target lead ECG signals.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the multi-lead electrocardiogram signal reconstruction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method for reconstructing multi-lead electrocardiogram signals as described in any one of claims 1 to 7.

Images