Training method of electrocardiogram reconstruction model, electrocardiogram reconstruction method and system

Abstract

This invention provides a training method for an electrocardiogram (ECG) reconstruction model (the ECG reconstruction model adopts a generative flow model), as well as an ECG reconstruction method and system. The training method for the ECG reconstruction model utilizes the reversibility of the flow model and introduces the information entropy difference between an integrated sensing signal source and the ECG signal to train the ECG reconstruction model, thereby supporting the interpretability of the modeling principle of the ECG reconstruction method and improving the quality of the reconstructed ECG.

Description

Technical Field

[0001] This invention relates to the field of machine learning, specifically to the field of deep learning in machine learning, and more specifically to intelligent medical technology applying deep learning, namely, a training method for an electrocardiogram (ECG) reconstruction model, an ECG reconstruction method, and an ECG reconstruction system. Background Technology

[0002] In the field of smart healthcare, electrocardiograms (ECGs), as a physiological indicator, are widely used in daily health monitoring and clinical diagnosis. For example, ECGs can be used for arrhythmia classification and heart failure monitoring. However, the ECG acquisition process requires attaching or fixing multi-lead electrodes to the user's skin. This attachment makes daily ECG monitoring inconvenient, and prolonged contact with the user's skin, such as with adhesive or fixed ECG electrodes, can cause skin allergies.

[0003] To address the aforementioned issues in electrocardiogram (ECG) acquisition, an immersive sensing ECG acquisition method is employed. This method involves using a wearable device on the user to acquire physiological signals (immersive sensing signals or signal sources) with immersive sensing characteristics, and then converting these physiological signals into ECG signals. The method of converting physiological signals into ECG signals is known as the ECG reconstruction method. This ECG reconstruction method combines the advantages of ECG's proven experience in medical auxiliary diagnosis with the ease of acquisition of physiological signals with immersive sensing characteristics. Physiological signals with immersive sensing characteristics include photoplethysmography (PPG) and cardiopulmonary bypass (BCG). PPG and BCG are widely used physiological signals in clinical settings and wearable devices. PPG is obtained by placing a photoelectric sensor on the skin and measuring changes in blood volume beneath the skin using light absorption and reflection. BCG converts minute vibrations on the skin surface caused by heartbeats into electrical signals using a sensor. Both photoplethysmography (PPG) and electrocardiogram (ECG) signals can achieve non-adhesive and user-unobtrusive acquisition characteristics in their acquisition principles, and based on this, immersive sensing ECG acquisition can be realized.

[0004] Based on the signal source type used in ECG reconstruction, existing ECG reconstruction methods are mainly divided into PPG-based ECG reconstruction methods and BCG-based ECG reconstruction methods. These methods are further categorized based on the main technical means employed: ECG reconstruction methods based on non-deep learning models and ECG reconstruction methods based on deep learning models. For example, reference 1 in patent publication number US20210315470 discloses a PPG-to-ECG reconstruction method based on a UNet model structure. This method leverages the powerful nonlinear fitting capabilities of deep learning to learn the mapping relationship from PPG to ECG. During the training phase, the model weights are adjusted, and during the application phase, the trained model is used for PPG-to-ECG reconstruction inference. Reference 2, with patent publication number US20220183606, discloses a method for PPG to ECG reconstruction based on DCT (Discrete Cosine Transform) (a non-deep learning model ECG reconstruction method). This method is based on DCT, a digital signal processing method, and expresses the mapping relationship from PPG to ECG as coefficients in DCT. The coefficients are fitted during the learning phase, and the learned coefficients are used for PPG to ECG reconstruction during the application phase. Reference 3, with patent publication number CN114041801A, discloses a method based on DCGAN (Generative Adversarial Network). This method uses a network composed of a generator and a discriminator. During the training phase, the generator and discriminator undergo adversarial training to iteratively improve the similarity between the reconstructed ECG and the real ECG. In the application phase, only the generator is used, with the BCG signal as input, to output the reconstructed ECG signal. Reference 4 of patent publication number CN114548365A discloses an ECG reconstruction method based on an LSTM neural network model. This method uses the imaging photoplethysmography (IPPG) signal as the input of the LSTM network and the ECG signal as the output of the LSTM neural network model to perform ECG reconstruction.

[0005] In summary, the ECG reconstruction method based on immersive sensing signal sources can effectively alleviate the problems of cumbersome acquisition process and difficult deployment caused by the pasted acquisition method of ECG. In particular, the ECG reconstruction method based on deep learning models can significantly improve the ability to represent the mapping relationship between source signals with immersive sensing acquisition characteristics and ECG signals. However, it still cannot solve the problem of poor ECG fidelity. Summary of the Invention

[0006] Therefore, the purpose of this invention is to overcome the shortcomings of the prior art and provide a training method for an electrocardiogram (ECG) reconstruction model, an ECG reconstruction method, and an ECG reconstruction system.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] According to a first aspect of the present invention, a method for training an electrocardiogram (ECG) reconstruction model is provided, the method comprising:

[0009] S1. Obtain a training dataset consisting of multiple training data pairs. Each training data pair includes ECG signal conversion data and immersive sensing signal data collected by the user at the same time. The ECG signal conversion data is obtained by converting the channel number of the ECG signal data collected by the user according to the channel number of the immersive sensing signal data, so that the channel number of the ECG signal conversion data is the channel number of the immersive sensing signal data plus 1. S2. Obtain an ECG reconstruction model, which is a flow model. S3. Train the ECG reconstruction model using the training dataset. The training includes forward inference and backward inference processes. During forward inference, the ECG reconstruction model generates synthetic immersive sensing signal data and an information entropy difference vector based on the ECG signal conversion data. The information entropy difference vector indicates the information entropy difference between the ECG signal conversion data and the immersive sensing signal data. During backward inference, the ECG reconstruction model generates synthetic ECG signal conversion data based on the immersive sensing signal data and an interpolation vector randomly sampled from a preset probability distribution of the information entropy difference information. The parameters of the ECG reconstruction model are updated using the information entropy difference loss corresponding to the information entropy difference vector.

[0010] In some embodiments of the present invention, the information entropy difference loss is determined based on the similarity between the probability distribution of the information entropy difference vector of each training data and a preset Gaussian probability distribution; or, the information entropy difference loss is equal to the mean of the information entropy difference vectors obtained from each training data pair.

[0011] In some embodiments of the present invention, during training, the parameters of the electrocardiogram reconstruction model are updated based on the total loss determined by the information entropy difference loss, the reconstruction loss between the synthesized ECG signal conversion data and the ECG signal conversion data, and the supervision loss between the synthesized immersive sensing signal data and the immersive sensing signal data.

[0012] In some embodiments of the present invention, the integrated sensing signal data includes cardiac impaction signal data and photoplethysmography (PPG) pulse wave signal data, and the total loss is determined in the following manner:

[0013]

[0014] Among them, L total Let ω1 represent the total loss, ω2 represent the first hyperparameter, and ω3 represent the third hyperparameter. Indicates the reconstruction loss. This represents the synthesized ECG signal conversion data, x ECG This represents ECG signal conversion data. Indicates monitoring losses, This represents the first supervised loss corresponding to the cardiac impaction signal data. This represents the synthesized cardiac impaction signal data, x BCG This represents the cardiac impaction signal data. This represents the second supervisory loss corresponding to the photoplethysmography (PPG) signal data. This represents the synthesized photoplethysmography (PPG) signal data, x PPG This represents the photoplethysmography (PPG) signal data, where T represents the information entropy difference loss.

[0015] In some embodiments of the present invention, the channel number conversion is performed in the following manner:

[0016] The ECG signal data is subjected to multiple wavelet transforms, each wavelet transform yielding an approximate component and a detail component. Zero-value interpolation is then performed on the approximate component and detail component obtained from the last wavelet transform, as well as the detail component from the wavelet transform preceding the last wavelet transform, to obtain the converted ECG signal data.

[0017] In some embodiments of the present invention, the electrocardiogram reconstruction model includes at least one recurrent neural network for extracting time-domain features and at least one fully connected neural network for extracting spatial-domain features.

[0018] According to a second aspect of the present invention, an electrocardiogram (ECG) reconstruction method based on an ECG reconstruction model is provided. The method includes the following steps: C1, acquiring historical data of a user, the historical data including ECG signal data and immersive sensing signal data of the user at multiple time points; C2, preprocessing the physiological signals obtained in step C1 according to a preset processing method to obtain training starting data composed of multiple data pairs, wherein each data pair includes ECG signal data and immersive sensing signal data of the user at least one heartbeat cycle; C3, adding 1 to the number of channels in the immersive sensing signal data in step C2 to obtain the number of conversion channels; C4, performing channel number conversion on the ECG signal data in the data pair obtained in step C2 to obtain ECG signal conversion data, which consists of ECG signal conversion data and immersive sensing signal data in a data pair. A training data pair is formed, and multiple training data pairs form a training dataset; C5. The ECG reconstruction model is trained multiple times using the training method described in the first aspect of the present invention to obtain a trained ECG reconstruction model; C6. The user's current immersive sensing signal is acquired, and the user's current immersive sensing signal is preprocessed according to a preset processing method to obtain the user's current immersive sensing signal data, and an insertion vector is generated by sampling from the preset information entropy difference information probability distribution between the immersive sensing signal and the ECG signal; C7. The user's current immersive sensing signal data and the sampled insertion vector are input into the trained ECG reconstruction model to obtain the user's current reconstructed ECG signal conversion data; C8. The user's current reconstructed ECG signal conversion data is subjected to an inverse channel number transformation to obtain the reconstructed ECG signal data.

[0019] In some embodiments of the present invention, the preset processing method includes one or more of the following methods: filtering, alignment, segmentation, and normalization.

[0020] According to a third aspect of the present invention, an electrocardiogram reconstruction system is provided, the system comprising:

[0021] An information acquisition terminal is configured to acquire a user's physiological signals, perform noise reduction processing on the acquired physiological signals, and output them; wherein the physiological signals include ECG signals and immersive sensing signals; a data management module is configured to receive the data output by the information acquisition terminal and process it according to a preset data structure; a database is configured to store the data processed by the data management module; and a reconstruction module is configured to execute the method described in the second aspect of the present invention to obtain reconstructed ECG signal data.

[0022] In some embodiments of the present invention, the reconfiguration module includes:

[0023] A data reading unit is configured to: acquire and output training physiological signals of the detection target, wherein the training physiological signals are ECG signals and immersive sensing signals of the detection target in multiple consecutive heartbeat cycles; a data preprocessing unit is configured to: receive the physiological signals output by the data reading unit, preprocess them according to a preset processing method to obtain and output training starting data composed of data pairs of the detection target in each heartbeat cycle, wherein each data pair is ECG signal data and immersive sensing signal data of the detection target in the same heartbeat cycle; a signal type judgment unit is configured to: use the number of channels of the immersive sensing signal data plus 1 as the conversion channel number, and convert the ECG signal data into ECG signal conversion data, wherein the number of channels of the ECG signal conversion data is equal to the conversion channel number; a model construction unit is configured to: store a pre-initialized electrocardiogram reconstruction model; and a training judgment unit is configured to: The system determines whether the ECG reconstruction model stored in the model building unit needs training and outputs a training judgment result. A model training unit is configured to: when the training judgment result output by the training unit indicates that training is required, execute the training method described in the first aspect of the invention to update the parameters of the model in the model building unit. An insertion vector sampling module is configured to: sample and generate an insertion vector from a preset information entropy difference probability distribution between the immersive sensing signal and the ECG signal. A reconstruction inference unit is configured to: when the training judgment unit outputs that training is not required, use the ECG reconstruction model stored in the model building unit as the trained ECG reconstruction model, acquire the immersive sensing signal in the physiological signal of the detection target and process it according to a preset processing method to obtain immersive sensing signal data, acquire the insertion vector obtained by the insertion vector sampling module, and use the trained ECG reconstruction model to perform ECG reconstruction.

[0024] Compared with the prior art, the advantages of this invention are as follows: In this invention, the channel number conversion is to ensure that the proposed ECG reconstruction method can adapt to different reconstruction source channel numbers. Specifically, the number of channels in the ECG signal conversion data is the number of channels in the integrated sensing signal data plus 1, that is, the number of channels in the ECG signal conversion data is one more than the number of channels in the integrated sensing signal data. This allows the ECG reconstruction model to process the introduced information entropy difference information vector randomly sampled from the information entropy difference information probability distribution function. Furthermore, by utilizing the invertibility of the flow model and introducing the information entropy difference between the integrated sensing signal source and the ECG signal, the ECG reconstruction model is trained, thereby supporting the interpretability of the ECG reconstruction method in terms of modeling principles and improving the quality of reconstructed ECGs. Attached Figure Description

[0025] The embodiments of the present invention will be further described below with reference to the accompanying drawings, wherein:

[0026] Figure 1 This is a schematic diagram of an electrocardiogram reconstruction system provided according to an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram illustrating the structure and data flow of an electrocardiogram reconstruction model employing a Flow model architecture, according to an embodiment of the present invention.

[0028] Figure 3 This is a schematic diagram of a transform subnetwork provided according to an embodiment of the present invention;

[0029] Figure 4 This is a flowchart illustrating a training method for an electrocardiogram reconstruction model according to an embodiment of the present invention.

[0030] Figure 5 This is a schematic flowchart of an electrocardiogram reconstruction method according to an embodiment of the present invention;

[0031] Figure 6 This is a flowchart illustrating a specific implementation of an electrocardiogram reconstruction method according to an embodiment of the present invention;

[0032] Figure 7 This is a schematic diagram illustrating an electrocardiogram reconstruction system according to an embodiment of the present invention, combined with a specific application scenario.

[0033] Figure 8 This is a schematic diagram comparing the original ECG signal and the reconstructed ECG signal provided according to an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention is further described in detail below through specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0035] As mentioned in the background section, ECG reconstruction methods based on immersive sensing signal sources can effectively alleviate the problems of cumbersome acquisition process and difficult deployment caused by the pasted acquisition method of ECG. In particular, the ECG reconstruction method based on deep learning models can significantly improve the ability to represent the mapping relationship between source signals and ECG signals with immersive sensing acquisition characteristics. However, it still cannot solve the problem of poor ECG fidelity.

[0036] To address the issue of poor fidelity in reconstructed ECGs, the inventors of this application have conducted further analysis of existing technologies. The analysis process will be explained below using the technologies provided in the four references (Reference 1, Reference 2, Reference 3, and Reference 4) mentioned in the background section as examples.

[0037] ECG reconstruction methods based on non-deep learning models (such as reference 2 mentioned above) mainly fit the mapping relationship between ECG and PPG signals using a small number of adjustable parameters, and then reconstruct the ECG signal based on a predefined calculation formula. This method has relatively weak nonlinear representation capabilities, and the reconstructed ECG signal may deviate significantly from the true ECG signal. In contrast, ECG reconstruction methods based on deep learning models (such as references 1, 2, and 3 mentioned above) can effectively address this issue of significant deviation. While deep learning-based ECG reconstruction methods significantly improve the representation capability of the mapping relationship between source signals and ECG signals with immersive sensing acquisition characteristics (thus solving the problem of significant deviation), they neglect the difference in information entropy between the ECG signal (which more faithfully reflects cardiac activity) and the source signal, forcibly fitting them (i.e., mapping the source signal with immersive sensing acquisition characteristics to the ECG signal). This makes the modeling principle of the ECG reconstruction method lack interpretability, thus failing to solve the problem of poor ECG fidelity. In other words, existing ECG reconstruction methods cannot interpolate the information entropy difference between the ECG signal and the immersive sensing source signal. From the perspective of the sensing principle of the reconstructed source signal and target signal, the modeling principle of existing ECG reconstruction methods lacks interpretability and cannot solve the problem of poor ECG fidelity. Therefore, to solve the above problems, this invention provides a training method for an ECG reconstruction model (the ECG reconstruction model adopts a generative flow model), as well as an ECG reconstruction method and system. The ECG reconstruction model training method utilizes the reversibility of the flow model and introduces the information entropy difference between the immersive sensing signal source and the ECG signal to train the ECG reconstruction model, realizing the interpolation of the information entropy difference between the reconstructed source signal (immersive sensing signal source) and the target signal (ECG signal), thereby supporting the interpretability of the ECG reconstruction method in terms of modeling principle, improving the learning ability of the mapping relationship between the reconstructed source signal and the target signal, and thus improving the quality of the reconstructed ECG. First, let's explain the reversibility of the flow model. The reversibility of the flow model includes forward inference and backward inference. In forward inference, the ECG signal is used as input, and cascaded basic building blocks are used for forward computation. In backward inference uses an immersive sensing signal source as input and performs backward computation through cascaded basic building blocks. The following uses a basic building block to illustrate the reversibility of the flow model.

[0038] An example of a forward reasoning calculation process can be represented as:

[0039] ECG1, ECG2 = split(ECG)

[0040] (BCG / PPG)′1=ECG1+F(ECG2)

[0041] (BCG / PPG)′2=ECG2·exp(H((BCG / PPG)′1))+G((BCG / PPG)′1)

[0042] (BCG / PPG)′=Concat((BCG / PPG)′1, (BCG / PPG)′2)

[0043] Where split(*) represents dividing the signal data into two equal parts according to the time dimension, Concat(*) represents merging the signal data according to the time dimension, exp(*) represents the exponential function, ECG1 and ECG2 represent the first and second equal parts of the ECG signal data after being divided into two parts; (BCG / PPG)′1 represents the first equal part of the data after forward inference calculation, (BCG / PPG)′2 represents the second equal part of the data after forward inference calculation, and (BCG / PPG)′ represents the final output of forward inference as the synthesized immersive sensing signal data. F(*) represents the first transform subnetwork, H(*) represents the second transform subnetwork, and G(*) represents the third transform subnetwork.

[0044] An example of the reverse reasoning calculation process can be represented as follows:

[0045] (BCG / PPG)1, (BCG / PPG)2=split((BCG / PPG))

[0046] ECG′2=((BCG / PPG)2-G((BCG / PPG)1))·exp(-H((BCG / PPG)1)),

[0047] ECG′1=(BCG / PPG)1-F(ECG′2)

[0048] ECG′=Concat(ECG′1, ECG′2)

[0049] In this context, (BCG / PPG)1 and (BCG / PPG)2 represent the first and second equal parts of the immersive sensing signal data (BCG / PPG) after being divided into two equal parts. ECG′1 represents the first equal part after inverse inference calculation, ECG′2 represents the second equal part after inverse inference calculation, and ECG′ represents the reconstructed ECG signal data as the final output of inverse inference. It should be noted that the sub-networks in the basic building block can consist of a first transform sub-network and a second transform sub-network, or they can consist of a first transform sub-network, a second transform sub-network, and a third transform sub-network. Here, for ease of description, the first, second, and third transform sub-networks are chosen to form the basic building block for explaining reversibility. When illustrating reversibility using the basic building blocks composed of the first and second transformation subnetworks, the corresponding terms of G(*) in the above process illustration can be omitted. For example, the forward inference calculation process example, after omitting the corresponding terms of G(*), is represented as: (BCG / PPG)′2=ECG2·exp(H((BCG / PPG)′1)).

[0050] It should be noted that in this embodiment of the invention, the reversibility of the flow model is introduced when reconstructing the ECG signal. Simultaneously, during the reverse inference process of the flow model, the distribution of the actual immersive sensing signal and the randomly sampled difference signal is mapped to the ECG signal. In other words, the probability distribution of the information entropy difference between the reconstructed source signal and the ECG signal is preset and randomly sampled. The sampled vector is then combined with the immersive sensing signal and input into the ECG reconstruction model. This achieves information entropy difference interpolation between the reconstructed source signal and the target signal (ECG signal), improving the fidelity of the reconstructed ECG while ensuring interpretability during the modeling process. In training the ECG reconstruction model, the reversibility of the flow model is utilized to inversely map the immersive sensing signal source and the information entropy difference signal distribution vector to the ECG signal; the immersive sensing signal corresponding to the ECG signal is output in the forward direction, realizing a bidirectional mapping between the ECG signal and the immersive sensing signal. This ensures that the information entropy difference between the immersive sensing signal source and the ECG signal is interpolated during the modeling process, so that the modeling process of the ECG reconstruction method satisfies the physiological relationship between the reconstructed source signal and the target signal (that is, it satisfies the interpretability of the ECG reconstruction method in terms of modeling principle), significantly improving the quality of the reconstructed ECG. Continuing with the example of the above inverse reasoning calculation process, before the inverse reasoning begins, the preset information entropy difference probability distribution between the reconstructed source signal (immersive sensing signal) and the ECG signal is randomly sampled. The sampled interpolation vector is combined with the immersive sensing signal data (BCG / PPG) and input into the flow model to realize the interpolation of the information entropy difference between the reconstructed source signal and the ECG signal, improving the fidelity of the reconstructed ECG, while also ensuring interpretability at the modeling process level. This invention designs an adaptive channel number reconstruction method to address the issue of variable channel number in the immersive sensing signal source in the proposed ECG signal reconstruction method. It utilizes channel number conversion (e.g., wavelet transform levels) to adjust the number of input signal channels, supporting variable-channel immersive sensing signal source input. The ECG reconstruction method and system provided in this invention use immersive sensing signals and sampled differential signal distributions to perform ECG reconstruction using an ECG reconstruction model. This provides an ECG acquisition method utilizing immersive sensing signals when ECG acquisition equipment is unavailable.

[0051] According to an embodiment of the present invention, an electrocardiogram (ECG) reconstruction method based on an ECG reconstruction model is provided. The method includes the following steps: C1, acquiring historical data of the user, the historical data including ECG signal data and immersive sensing signal data of the user at multiple time points; C2, preprocessing the physiological signals obtained in step C1 according to a preset processing method to obtain training starting data composed of multiple data pairs, wherein each data pair includes ECG signal data and immersive sensing signal data of the user at least one heartbeat cycle; C3, adding 1 to the number of channels in the immersive sensing signal data in step C2 as the number of conversion channels; C4, converting the number of channels in the ECG signal data in the data pair obtained in step C2 to obtain ECG signal conversion data, wherein the ECG signal conversion data and immersive sensing signal data in a data pair constitute a training starting data. The training data pairs consist of multiple training data pairs forming a training dataset; C5. The ECG reconstruction model is trained multiple times using the training method of the ECG reconstruction model provided in this invention to obtain a trained ECG reconstruction model; C6. The user's current immersive sensing signal is acquired and preprocessed according to a preset processing method to obtain the user's current immersive sensing signal data, and an insertion vector is generated by sampling from the preset information entropy difference probability distribution between the reconstruction source signal (immersive sensing signal) and the ECG signal; C7. The user's current immersive sensing signal data and the sampled insertion vector are input into the trained ECG reconstruction model to obtain the user's current reconstructed ECG signal conversion data; C8. The user's current reconstructed ECG signal conversion data is subjected to an inverse channel number transformation to obtain the reconstructed ECG signal data.

[0052] According to one embodiment of the present invention, such as Figure 1As shown, an electrocardiogram (ECG) reconstruction system is provided. The system includes a signal acquisition terminal 400, a data management module 406, a database 407, and an ECG reconstruction module 408. The signal acquisition terminal 400 is configured to acquire the user's physiological signals, perform noise reduction processing on the acquired physiological signals, and output them; wherein the physiological signals include ECG signals and immersive sensing signals; the data management module 406 is configured to receive the data output by the signal acquisition terminal and process it according to a preset data structure; the database 407 is configured to store the data processed by the data management module; the reconstruction module (i.e., the ECG reconstruction module 408) is configured to execute an ECG reconstruction method based on an ECG reconstruction model to obtain reconstructed ECG signal data.The ECG reconstruction module 408 includes a data reading unit 4081, a data preprocessing unit 4082, a signal type determination unit 4083, a model building unit 4084, a training determination unit 4085, a model training unit 4086, an ECG reconstruction inference unit 4087, and an insertion vector sampling module 4088. The data reading unit 4081 is configured to acquire and output training physiological signals of the detection target, wherein the training physiological signals are ECG signals and immersive sensing signals of the detection target over multiple consecutive heartbeat cycles. The data preprocessing unit 4082 is configured to acquire and output training physiological signals of the detection target over multiple consecutive heartbeat cycles. The system is configured to: receive physiological signals output by the data reading unit, preprocess them according to a preset processing method to obtain and output training starting data consisting of data pairs of the target in each heartbeat cycle, wherein each data pair consists of ECG signal data and immersive sensing signal data of the target in the same heartbeat cycle; the signal type judgment unit 4083 is configured to: use the number of channels of the immersive sensing signal data plus 1 as the conversion channel number, and convert the ECG signal data into ECG signal conversion data, wherein the number of channels of the ECG signal conversion data is equal to the conversion channel number; the model structure... The model building unit 4084 is configured to: store a pre-initialized electrocardiogram (ECG) reconstruction model; the training judgment unit 4085 is configured to: determine whether the ECG reconstruction model needs training and output a training judgment result; the model training unit 4086 is configured to: when the training judgment result output by the training judgment unit indicates that training is required, execute the training method based on the ECG reconstruction model provided by this invention to obtain a trained ECG reconstruction model and update the parameters of the model in the model building unit; and the insertion vector sampling module 4088 is configured to: extract data from the reconstruction source signal (immersive sensing signal). An insertion vector is generated by sampling from the preset information entropy difference probability distribution between the ECG signal and the ECG signal; the ECG reconstruction inference unit 4087 is configured to: when the output result of the training judgment unit is that no training is required, use the ECG reconstruction model stored in the model building unit as the trained ECG reconstruction model, and obtain the immersive sensing signal in the physiological signal of the detection target and process it according to the preset processing method to obtain the immersive sensing signal data and obtain the insertion vector obtained by the insertion vector sampling module 4088, and use the trained ECG reconstruction model to perform ECG reconstruction.

[0053] According to an embodiment of the present invention, a training method for an electrocardiogram (ECG) reconstruction model is provided. The method includes: S1, acquiring a training dataset consisting of multiple training data pairs, wherein each training data pair includes ECG signal conversion data and immersive sensing signal data collected by the user at the same time, wherein the ECG signal conversion data is obtained by converting the channel number of the ECG signal data collected by the user according to the channel number of the immersive sensing signal data, so that the channel number of the ECG signal conversion data is the channel number of the immersive sensing signal data plus 1; S2, acquiring an ECG reconstruction model, which is a flow model; S3, training the model using the training dataset. The electrocardiogram (ECG) reconstruction model is described, wherein training includes forward inference and backward inference processes. During forward inference, the ECG reconstruction model generates synthesized immersive sensing signal data and an information entropy difference vector based on ECG signal conversion data. The information entropy difference vector indicates the information entropy difference between the ECG signal conversion data and the immersive sensing signal data. During backward inference, the ECG reconstruction model generates synthesized ECG signal conversion data based on the immersive sensing signal data and an interpolation vector randomly sampled from a preset probability distribution of information entropy difference information. The parameters of the ECG reconstruction model are then updated using the information entropy difference loss corresponding to the information entropy difference vector.

[0054] To better explain this invention, the following mainly focuses on the perspective of electrocardiogram (ECG) reconstruction, and in conjunction with the accompanying drawings, explains the database construction, ECG reconstruction model, reconstruction data processing, ECG reconstruction model training, and ECG reconstruction process.

[0055] I. Database Construction

[0056] The ECG reconstruction method provided in this invention can realize ECG reconstruction in cases where the number of immersive sensing signal sources is not fixed. In other words, in this invention, the ECG reconstruction method can accommodate different numbers of input signal channels, meaning that there can be multiple immersive sensing signal sources. According to one embodiment of the invention, the immersive sensing signal includes one or more of the following: impactogram signal and photoplethysmography (PPG) signal. To facilitate the explanation of the database construction process in this invention, the historical data consisting of impactogram signal and PPG signal as immersive sensing signals and ECG signals required for training the ECG reconstruction model is used as an example. The following explains the database construction process in conjunction with the signal acquisition terminal 400 and the data management module 406.

[0057] like Figure 1As shown, the signal acquisition terminal 400 includes a data acquisition unit 4001, a data processing unit 4002, and a data transmission unit 4003. The data acquisition unit 4001 in the signal acquisition terminal 400 simultaneously acquires information from the subject (also known as the user) using a physiological signal sensor, thereby obtaining a raw dataset of ECG signals and an immersive sensing signal composed of cardiac impaction signals and photoplethysmography (PPG) pulse wave signals at the same time. This raw dataset is then sent to the data processing unit 4002 for further processing. The data processing unit 4002 in the signal acquisition terminal 400 can configure different filter cutoff frequencies according to the frequency bandwidth range of the raw signals in the raw dataset, and filter baseline drift and high-frequency noise. The filtered signal data is then sent to the data transmission unit 4003. The data transmission unit 4003 can transmit the digitized signal to the data management module 406 via Wi-Fi, Bluetooth, or wired connections.

[0058] like Figure 1 As shown, according to an embodiment of the present invention, the data management module 406 includes a data receiving unit 4061 and a data management unit 4062. The data receiving unit 4061 receives data (data streams of the basic filtered ECG signal and the immersive sensing signal) transmitted by the data transmission unit 4003 and performs data aggregation. The data receiving unit 4061 sends the aggregated data to the data management unit 4062 for standardized data storage and management. The data management unit 4062 can write the received data streams into the database 407 according to different data channels and a certain data format, thereby forming data for ECG reconstruction model training, as well as signal data required for ECG reconstruction during specific implementation (immersive sensing signal data required for ECG reconstruction). The data for ECG reconstruction model training is also the original signal dataset in the database 407 (the original signal dataset is also the user's historical data, which includes ECG signal data and immersive sensing signal data of the user at multiple times). The database 407 also stores a pre-set probability distribution of the information entropy difference between the reconstruction source signal (immersive sensing signal) and the ECG signal. Specifically, the pre-set information entropy difference probability distribution can be understood as setting a deviation information distribution function between the immersive sensing signal and the ECG signal. According to one embodiment of the present invention, the deviation information distribution function is a Gaussian distribution. It should be noted that the specific data format here can be understood as a format adapted to the storage structure of database 407. This format is a technique known in the art based on the type of database, and will not be elaborated upon here.

[0059] II. Electrocardiogram Reconstruction Model

[0060] Because the ECG reconstruction model in this embodiment of the invention employs a generative flow model, it can improve the ability to learn the mapping relationship between integrated sensing signals and ECG signals. According to one embodiment of the invention, the ECG reconstruction model adopts a flow model. For example... Figure 2 As shown, the flow model includes multiple cascaded flow model basic building blocks (TimeFlew basic blocks), and adjacent flow model basic building blocks are connected by 1×1 convolutional layers (1×1 Conv). The flow model basic building blocks are configured to generate output data based on their input data. Taking the l-th flow model basic building block composed of three transform sub-networks as an example, and combining... Figure 2 The explanation of the flow model involves the forward and backward reasoning processes of the electrocardiogram reconstruction model. Figure 2 In this context, Length represents the length of the input data of the l-th basic building block of the flow model, Length / 2 represents the length of the input data of the l-th basic building block of the flow model after being split, F, H, and G represent the structures of the first, second, and third transformation sub-networks, respectively; ☉ represents the operational relationship, split represents the bipartition function, and Concat represents data merging.

[0061] The data flow representation in the transformed subnetwork of the l-th flow model building block in forward inference is as follows:

[0062]

[0063]

[0064] The data flow representation in the sub-transform network of the l-th flow model building block in reverse reasoning is as follows:

[0065]

[0066]

[0067] in, This represents the first segment of data input to the l-th flow model building block in forward inference. This represents the second segment of input data for the l-th flow model building block in forward reasoning. This represents the first segment of data output by the l-th flow model building block in forward inference. This represents the second segment of data output by the l-th flow model building block in forward inference. This represents the first segment of data input to the l-th flow model building block in reverse inference. This represents the second segment of data input to the l-th flow model building block in reverse inference. This represents the first segment of data output by the l-th basic building block of the flow model in reverse inference. F(*) represents the second segment of data output by the l-th flow model building block in reverse inference, H(*) represents the first transformation subnetwork, G(*) represents the second transformation subnetwork, and G(*) represents the third transformation subnetwork.

[0068] According to one embodiment of the present invention, the electrocardiogram reconstruction model includes at least one recurrent neural network for extracting time-domain features and at least one fully connected neural network for extracting spatial-domain features. The structure of the aforementioned transform sub-networks can be an existing structure or an improved structure. Preferably, the transform sub-networks F, H, and G have the same model structure, using recurrent neural networks and fully connected neural networks as basic operators. The data stream is first passed through the recurrent neural network operator to extract the time-domain relationships in the data stream. Subsequently, the output of the recurrent neural network operator is used as the input of the fully connected neural network operator to extract the spatial-domain mapping relationship between the reconstructed source signal and the target ECG signal, which can further significantly improve the quality of the reconstructed ECG. Figure 3 The diagram shows the network structure of the transform subnetwork. For ease of representation, LSTM is used to denote a computational unit within the recurrent neural network. X0, X1, ..., Xn represent the input data to the recurrent neural network, and h0, h1, ..., hn represent the output data. Since the specific structures and functions of recurrent neural networks and fully connected neural networks are well-known to those skilled in the art, their structures and functions will not be elaborated upon here.

[0069] III. Data Reconstruction Processing

[0070] During ECG reconstruction, the data from the database obtained in the first part needs further processing to adapt to the ECG reconstruction model and ECG reconstruction. This invention embodiment combines... Figure 1 We will continue with the explanation of data processing.

[0071] The data reading unit 4081 in the ECG reconstruction module 408 is configured to acquire and output the training physiological signals of the detection target. These training physiological signals are the ECG signals and immersive sensing signals of the detection target over multiple consecutive heartbeat cycles. The unit can convert the signal data (training physiological signals) from binary format into a software-processable data structure (a preset data structure), and then send it to the data preprocessing unit 108 for subsequent data preprocessing. The software-processable data structure here should be understood as conventional computer processing, which is known to those skilled in the art and will not be described in detail here.

[0072] The data preprocessing unit 4082 in the ECG reconstruction module 408 is configured to receive the physiological signals output by the data reading unit, preprocess them according to a preset processing method, and output training starting data composed of data pairs of the detection target in each heartbeat cycle. Each data pair consists of ECG signal data and immersive sensing signal data of the detection target in the same heartbeat cycle. Specifically, the data preprocessing unit 4082 is mainly used to process the data used for ECG reconstruction model training (processing the original ECG signal data, PPG signal data, and BCG signal data) into the format required for the ECG reconstruction model input. First, referring to the correspondence between the waveforms of the original ECG signal, PPG signal, and BCG signal and cardiac physiological indicators, the peak point of the R wave of the ECG signal, the trough point of the PPG signal, and the peak point of the J wave of the BCG are aligned in the time domain. Then, the ECG, PPG, and BCG signals are divided into heartbeat-level (heartbeat cycle) data according to the above alignment points and paired. Each heartbeat-level data segment is normalized to obtain the data pairs in the training starting data. It should be noted that when performing ECG reconstruction, the PPG signal and BCG signal of the source signal (immersive sensing signal) are reconstructed. The trough of the PPG signal and the J peak of the BCG are aligned in the time domain. The aligned data is then divided into heartbeat level (heartbeat cycle) data. Each heartbeat level data segment is normalized to obtain the data corresponding to the reconstructed source signal during ECG reconstruction.

[0073] IV. Training of the ECG Reconstruction Model

[0074] The following combination Figure 4 This section explains the training process of the ECG reconstruction model. Preferably, the data initialized by the data preprocessing unit 4082 is converted to a higher channel count to obtain the training dataset for training. This section primarily focuses on the detailed training process of the ECG reconstruction model. The configuration functions of other modules in the ECG reconstruction module 408 (model building unit 4084, training and judgment unit 4085, model training unit 4086, ECG reconstruction inference unit 4087, and insertion vector sampling module 4088) will be discussed in detail during the ECG reconstruction process below.

[0075] Specifically, the training method for the ECG reconstruction model includes: acquiring a training dataset consisting of multiple training data pairs, wherein each training data pair includes ECG signal conversion data and immersive sensing signal data collected by the user at the same time. The ECG signal conversion data is obtained by converting the channel number of the ECG signal data collected by the user according to the channel number of the immersive sensing signal data, so that the channel number of the ECG signal conversion data is the channel number of the immersive sensing signal data plus 1; acquiring an ECG reconstruction model, which is a flow model; and training the ECG reconstruction model using the training dataset. The training includes forward inference and backward inference processes. During forward inference, the ECG reconstruction model generates synthetic immersive sensing signal data and an information entropy difference vector based on the ECG signal conversion data. The information entropy difference vector indicates the information entropy difference between the ECG signal conversion data and the immersive sensing signal data. During backward inference, the ECG reconstruction model generates synthetic ECG signal conversion data based on the immersive sensing signal data and an interpolation vector randomly sampled from a preset probability distribution of the information entropy difference information. The parameters of the ECG reconstruction model are updated using the information entropy difference loss corresponding to the information entropy difference vector.

[0076] like Figure 4 As shown, it should be noted that the real ECG signal data (ECG signal conversion data) in the dataset (i.e., the training dataset) is input into the Flow reconstruction model (corresponding to the ECG reconstruction model). The Flow reconstruction model performs forward inference and outputs the synthesized integrated sensing signal and the information entropy difference information vector. The information entropy difference information vector output by forward inference is one of the data dimensions calculated by the Flow reconstruction model. The information entropy difference information vector output by forward inference will be used to calculate the loss function and to train the model, so that the information entropy difference information is mapped to a Gaussian distribution. The multidimensional data obtained by combining the integrated sensing signal in the dataset and the information entropy difference information vector randomly sampled from the information entropy difference information probability distribution function is input into the Flow reconstruction model. The Flow reconstruction model performs reverse inference and outputs the reconstructed ECG. The reconstructed ECG, the synthesized integrated sensing signal, and the information entropy difference information vector are substituted into the loss function to calculate the loss, and the optimizer is used to update the model weights.

[0077] Each training data pair includes ECG signal conversion data and immersive sensing signal data collected by the user at the same time. The ECG signal conversion data is obtained by converting the channel number of the ECG signal data collected by the user according to the channel number of the immersive sensing signal data, so that the channel number of the ECG signal conversion data is the channel number of the immersive sensing signal data plus 1. According to one embodiment of the present invention, the channel number conversion is performed by an adaptive method for reconstructing the source channel number. First, the channel number of the immersive sensing signal data is determined, and then wavelet transform is used to satisfy the different signal channel number conditions during the reconstruction of the ECG signal. Specifically, when the number of input channels of the immersive sensing signal source is small, a shallow wavelet transform is used for the preprocessing of the original ECG signal; when the number of input channels of the immersive sensing signal source increases, a deeper wavelet transform is used for the preprocessing of the original ECG signal, ensuring that the number of signal channels of the ECG signal after wavelet transform corresponds to the number of channels of the immersive sensing signal source, thereby satisfying the different input signal channel number conditions. Specifically, the channel number conversion is to ensure that the ECG reconstruction method proposed in this invention can adapt to different channel numbers of reconstruction sources. The number of channels in the ECG signal conversion data is one more than the number of channels in the integrated sensing signal data, meaning the ECG signal conversion data has one more channel than the integrated sensing signal data. This allows the ECG reconstruction model to handle the introduced information entropy difference information. In forward reasoning, the channel-number-converted data is concatenated before being input into the ECG reconstruction model for processing. In reverse reasoning, the interpolation vector randomly sampled from a preset probability distribution of information entropy difference information and the integrated sensing signal data are concatenated before being input into the ECG reconstruction model for processing (e.g., ...). Figure 2 (C in the middle represents the concatenated data).

[0078] According to one embodiment of the present invention, channel number conversion is performed as follows: multiple wavelet transforms are performed on the ECG signal data, each wavelet transform yielding an approximate component and a detail component; zero-value interpolation is then performed on the approximate component and detail component obtained from the last wavelet transform, as well as the detail component from the wavelet transform preceding the last wavelet transform, to obtain the converted ECG signal data. It should be noted that when using immersive sensing signal data for ECG reconstruction, performing a single wavelet transform on the ECG signal data yields an approximate component and a detail component, thus obtaining the converted ECG signal data.

[0079] The following example uses an immersive sensing signal composed of an ECG signal and a photoplethysmography (PPG) signal to explain the channel number transformation process. In this embodiment, the number of immersive sensing signals used is two. The adaptive method for reconstructing the source channel number performs a two-stage wavelet transform on the ECG channels. ECG where φ a,b and ψ a,b Let represent the approximation function and detail function of the Bb5 wavelet basis function with scale 'a' and shift 'b', respectively. N represents the length of the ECG signal. In this embodiment, the wavelet transform shift is 0 (i.e., b = 0), and the transform scales are 1 and 2. First, using the original ECG signal as input, a transform with scale 1 (i.e., a = 1) is performed, outputting approximation components and first-level detail components. Then, using the approximation components as input, a transform with scale 2 (i.e., a = 2) is performed, outputting second-level approximation components and second-level detail components. Zero-value interpolation is performed on the second-level approximation components, first-level detail components, and second-level detail components to expand them into ECG data of equal length. Wavelet reconstruction is then performed, ultimately converting the data into ECG signal data composed of three channels: second-level approximation components, first-level detail components, and second-level detail components. The output data structure conforms to the interface format requirements of the reconstruction model. It should be noted that the reconstructed ECG signal data obtained in the inverse reconstruction process is obtained by using the inverse wavelet transform to obtain the reconstructed ECG signal data with merged channels. It should be noted that the second-level approximation component corresponds to the PPG signal component data in the ECG signal data, the first-level detail component corresponds to the BCG signal component data in the ECG signal data, and the second-level detail component corresponds to the information entropy difference information component data in the BCG signal component data in the ECG signal data. It should also be noted that the adaptive method for the number of reconstructed source channels can be configured in the signal type determination unit 4083, which is configured to: use the number of channels in the immersive sensing signal data plus 1 as the conversion channel number, and convert the ECG signal data into ECG signal conversion data, wherein the number of channels in the ECG signal conversion data is equal to the conversion channel number; the adaptive method for the number of reconstructed source channels can be configured to run in the data preprocessing unit 4082. In a preferred embodiment of the present invention, the data preprocessing unit 4082 is configured to run the above-mentioned adaptive method for the number of reconstructed source channels.

[0080] According to one embodiment of the present invention, during training, the parameters of the electrocardiogram reconstruction model are updated based on the total loss determined by the information entropy difference loss, the reconstruction loss between the synthesized ECG signal conversion data and the ECG signal conversion data, and the supervision loss between the synthesized immersive sensing signal data and the immersive sensing signal data.

[0081] According to one embodiment of the present invention, the immersive sensing signal data, including cardiac impaction signal data and photoplethysmography (PPG) signal data, determines the total loss in the following manner:

[0082]

[0083] Among them, L totalLet ω1 represent the total loss, ω2 represent the first hyperparameter, and ω3 represent the third hyperparameter. Indicates the reconstruction loss. This represents the synthesized ECG signal conversion data, x ECG This represents ECG signal conversion data. Indicates monitoring losses, This represents the first supervised loss corresponding to the cardiac impaction signal data. This represents the synthesized cardiac impaction signal data, x BCG This represents the cardiac impaction signal data. This represents the second supervisory loss corresponding to the photoplethysmography (PPG) signal data. This represents the synthesized photoplethysmography (PPG) signal data, x PPG This represents the photoplethysmography (PPG) signal data, where T represents the information entropy difference loss.

[0084] According to one embodiment of the present invention, the information entropy difference loss is determined based on the similarity between the probability distribution of the information entropy difference vector of each training data and a preset Gaussian probability distribution, and the information entropy difference loss is calculated according to the following rules:

[0085]

[0086] Where CrossEntropy(*) represents the cross-entropy function, and p(z) is the probability distribution of the information entropy difference vector z. This represents the probability distribution of the input ECG signal data. This indicates that the value of z is The probability at time T is transformed for the first time into a variational form of cross-entropy. Using the forward distribution, the expression for T is transformed for the second time as follows: q(ECG) represents the probability distribution of the input ECG signal data, and the information entropy difference loss T represents... The similarity between the predicted distribution of z and the true distribution of z, E q(ECG) Let Z represent the expected similarity between the probability of the stream model inference output Z and the actual probability of Z, given all ECGs observed as inputs when the probability distribution q(ECG) of the unknown ECG is unknown.

[0087] According to one embodiment of the present invention, the information entropy difference loss is equal to the mean of the information entropy difference vectors obtained from each training data pair, and the information entropy difference loss is calculated according to the following rule:

[0088]

[0089] Where M represents the number of training data pairs, k represents the k-th training data pair in the training dataset, and N represents the length of the BCG, PPG, and ECG data in a data pair. This represents the interpolation vector output by the forward inference of the flow model.

[0090] Taking the probability distribution of the information entropy difference vector z as a one-dimensional Gaussian distribution (i.e., the standard normal distribution) as an example, the formula is explained as follows:

[0091]

[0092] Substituting μ = 0 and σ = 1 into the above equation, we get:

[0093]

[0094] p(z) is the probability distribution of the information entropy difference vector z, corresponding to f(x) in formula (2). Then, substituting into the information entropy difference loss formula (1):

[0095]

[0096] In the process of calculating the loss function As a constant, z can be ignored, and in formula (3), z is replaced by... This leads to the derivation of the subsequent loss calculation formula:

[0097]

[0098] when When represented in discrete form and the number of training data pairs is M,

[0099] When constructing a loss for corresponding data based on a single data point, according to an embodiment of the present invention, the total loss is determined as follows:

[0100] Where N represents the length of the BCG, PPG, and ECG data in a data pair. This represents the interpolation vector output by the forward inference of the flow model.

[0101] V. Electrocardiogram Reconstruction Process

[0102] The following combination Figure 5 This section provides a comprehensive overview of the electrocardiogram (ECG) reconstruction process from a methodological and procedural perspective. For example... Figure 5As shown, the specific process of the electrocardiogram reconstruction method is as follows: Step 201 is the starting step, where the signal acquisition terminal acquires and filters data, and transmits the data to the data management module through the data transmission module. In step 202, the data management module receives the data and stores it in the dataset. In step 203, it is determined whether reconstruction model training is needed. The indicators used for this determination are the relative root mean square error (RMS) and the Pearson correlation coefficient (PRC). The criterion is whether the indicators have reached the test results on the test set under the model's convergence state. If the criterion is not met (the training determination result indicates that training is needed), then model training is required, and step 204 is executed. If the model is to be used, step 211 is executed. In step 204, the system determines whether the data collected by the data management module is sufficient for ECG reconstruction model training. If it is insufficient, it returns to step 201 to continue collecting data. If it is sufficient (e.g., 5 hours of data), it proceeds to step 205 to read the collected dataset and perform preprocessing processes such as data filtering, alignment, segmentation, and normalization. Wavelet transform decomposition is also performed on the ECG signal, and the final output meets the input and output format requirements of the Flow-based reconstruction model. Step 206 mainly divides the preprocessed ECG signal and immersive sensing signal data segments into training and test sets for subsequent reconstruction model training. In step 207, the Flow-based ECG reconstruction model is constructed and initialized, and basic components such as the optimizer are built for model training. Then, in step 208, the system uses the dataset and optimizer to adjust the weights of the reconstruction model and train it. In step 209, the system determines whether the reconstruction model has converged, i.e., whether the model can reconstruct high-quality ECG signals on the test set. If the model has not converged, it returns to step 208 to continue training the reconstruction model. If the model has converged, it proceeds to step 210 to store the model structure and model weight data. If the system selects a model usage mode, it proceeds to step 211. In step 211, the system reads the immersive sensing signal and performs data preprocessing to construct a dataset that meets the input format requirements of the reconstruction model, then proceeds to step 212 for model inference. In step 212, the system reads the model structure and model weight data stored in step 210, updates the model using the model weight data, and reconstructs the model to obtain a trained ECG reconstruction model. The reconstruction model only needs to obtain the randomly sampled difference vector (also called the interpolation vector when it is discrete data) and the immersive sensing signal source obtained in step 211 as inputs, and can output a high-quality reconstructed ECG using the trained ECG reconstruction model.

[0103] The following combination Figure 1This section provides a comprehensive overview of the electrocardiogram (ECG) reconstruction process from a systems perspective. For example... Figure 1As shown, the reconstruction process of the ECG reconstruction system is as follows: The data acquisition unit 4001 of the signal acquisition terminal 400 collects subject information through physiological signal sensors to obtain raw data of ECG and immersive sensing signals. The raw data is sent to 4002 for further processing. The data processing unit 4002 in the signal acquisition terminal 400 can configure different filter cutoff frequencies according to the frequency bandwidth range of the raw signal to filter out baseline drift and high-frequency noise. The filtered signal is sent to the data transmission unit 4003. The data transmission unit 4003 can transmit the digitized signal to the data receiving unit 4061 for data aggregation via Wi-Fi, Bluetooth, or wired connection. The data receiving unit 4061 receives the data stream of ECG signal and immersive sensing signal after basic filtering and sends it to the data management unit 4062 for standardized data storage and management. The data management unit 4062 can write the received data stream into the database 407 according to a certain data format through different data channels to form a dataset for ECG reconstruction model training. When the amount of data in database 407 is sufficient (in step 204), for example, 5 hours of data, preferably, the data is sent to data reading unit 4081 for data reading and subsequent model training. Data reading unit 4081 can read data from the dataset storing the original signals and output it to data preprocessing unit 4082 according to a specific data structure for subsequent data preprocessing. Data preprocessing unit 4082 mainly performs data filtering based on the bandwidth information of the ECG signal and the immersive sensing signal, aligns signals by referring to the correspondence between different signals and cardiac activity, divides the signal into heartbeat-level data segments, and performs signal normalization transformation. After preprocessing, signal type judgment unit 4083 is called to run the adaptive method for the number of reconstruction source channels (i.e., channel number conversion operation). Based on the number of channels input by the immersive sensing signal, wavelet transform is used to convert the ECG signal into multiple channels of approximate and detail components, where the approximate components correspond to the immersive sensing signal, i.e., low frequency. Information, the detail components correspond to the information entropy difference signal, i.e., the high-frequency signal. Based on the correspondence between the wavelet components and the immersive sensing signal and the information entropy difference signal, these wavelet channels are combined in the data dimension. When the number of input channels of the immersive sensing signal source is small, the preprocessing of the original ECG signal adopts a shallow wavelet transform. When the number of input channels of the immersive sensing signal source increases, the preprocessing of the original ECG signal adopts a deeper wavelet transform, generating a data structure that meets the interface format requirements of the flow model reconstruction method, supporting the input of immersive sensing signal sources with variable channels, and improving the accuracy and efficiency of ECG reconstruction.The processed data is fed into the model building unit 4084, which primarily initializes the basic operator layers of the Flow-based ECG reconstruction model and connects these layers. It also initializes the model training optimizer, defines the loss function, and encapsulates the data according to the format requirements for the reconstruction model. The completed model, optimizer, and data package are then passed to the training decision unit 4085. The training decision unit 4085 determines whether reconstruction model training is necessary. If training is required, the encapsulated, completed model, optimizer, and data package are passed to the training unit 4086 for subsequent model training. The model training unit 4086 primarily uses formatted ECG signals and immersive sensing signal data to train and adjust the weights of the reconstruction model, gradually enabling it to output high-quality reconstructed ECG signals. After model convergence, the trained reconstruction model is passed to the ECG reconstruction inference unit 4087 for use in ECG reconstruction scenarios. The ECG reconstruction inference unit 4087 mainly reads the immersive sensing signal and the insertion vector sampling module 4088 samples the preset information entropy difference information probability distribution between the reconstruction source signal (immersive sensing signal) and the ECG signal to generate an insertion vector, and outputs a high-fidelity reconstructed ECG for health monitoring.

[0104] The following conclusion Figure 7 and Figure 8 This paper explains the ECG reconstruction process from an application perspective. In this embodiment, we take the joint acquisition and input of BCG and PPG into the ECG reconstruction model, and then output the reconstructed ECG as an example to demonstrate the Flow-based ECG reconstruction method.

[0105] like Figure 7 As shown, the ECG acquisition terminal 401 deploys a signal acquisition terminal 400 for acquiring ECG signals and transmitting them; the PPG smartwatch 402 deploys a signal acquisition terminal 400 for acquiring PPG signals and transmitting them; the BCG scale 403 deploys a signal acquisition terminal 400 for acquiring BCG signals and transmitting them; the laptop 404 and smartphone 405 are used to deploy the data management module 406, store the dataset 407 (i.e., the database), and run the ECG reconstruction module 408. In general scenarios, especially those involving daily health monitoring, exercise recording, and physiological indicator monitoring during work, the ECG acquisition terminal 401 does not possess the characteristics of convenient, non-intrusive, and effective application, because in these scenarios, users need to perform other activities while wearing health monitoring devices, such as exercising, running, or working. To obtain high-quality ECG signals in these scenarios, we first need to build a complete ECG reconstruction system, and then, in the application, simply wearing the smartwatch 402 and using the BCG scale 403 will allow us to obtain high-quality ECG signals.

[0106] The data acquisition unit in the ECG acquisition terminal 401 includes ECG acquisition electrodes, an operational amplifier, and an ADC analog-to-digital converter. It acquires the wearer's ECG data, and the raw ECG data is sent to the data processing unit in the ECG acquisition terminal 401 for further processing. Based on the ECG signal spectrum distribution, the data processing unit in the ECG acquisition terminal 401 uses an FIR filter with a cutoff frequency of 0.3Hz to 40Hz to filter out baseline drift caused by conductor disturbances and high-frequency noise caused by the surrounding environment. The filtered ECG signal is then sent to the data transmission unit in the ECG acquisition terminal 401. The data transmission unit in the ECG acquisition terminal 401 uses Bluetooth as the wireless transmission method to transmit the digitized ECG signal to a laptop computer (404) for data aggregation and subsequent use.

[0107] The PPG data acquisition unit in the smartwatch uses a PPG sensor to detect the wearer's PPG data. The raw PPG data is then sent to the smartwatch's data processing unit for further processing. Based on the PPG signal spectrum distribution, the data processing unit employs a Butterworth filter with a cutoff frequency of 1Hz to 8Hz to filter out baseline drift caused by wire disturbances and high-frequency noise from the surrounding environment. The filtered PPG signal is then sent to the smartwatch's data transmission unit. The data transmission unit preferably uses Bluetooth as the wireless transmission method to transmit the digitized PPG signal to a 404 laptop for data aggregation and subsequent use.

[0108] The data acquisition unit in the BCG collection scale uses minute changes in the load sensor to sense the current subject's BCG data. The raw BCG data is sent to the data preprocessing unit in the BCG collection scale. Based on the signal spectrum distribution of BCG, the current filter uses an FIR filter with a cutoff frequency of 0.5Hz to 20Hz to filter out noise in the BCG signal. The filtered BCG signal is sent to the data transmission unit in the BCG collection scale, and Bluetooth is preferred as the wireless transmission method to transmit the digitized BCG signal to a 404 laptop for data aggregation and subsequent use.

[0109] In this embodiment, laptops 404 and 405 handle three main functions: deployment of the data management module 406, storage of the dataset 407, and operation of the ECG reconstruction module 408. The data management module 406 uses a Bluetooth receiver to collect ECG, PPG, and BCG data from channels 401, 402, and 403, synchronizes the data across the three channels, and then sends it to the data management unit 4061 for standardized data storage and management. 407 uses a MySQL database for data storage and management, and updates the dataset 407 used for ECG reconstruction model training. The ECG reconstruction module 408 includes a data reading unit 4081, a data preprocessing unit 4082, a signal type determination unit 4083, a model building unit 4084, a training determination unit 4085, a model training unit 4086, and an ECG reconstruction inference unit 4087. The data reading unit 4081 is the software running in the laptop that reads ECG, PPG, and BCG signal datasets from memory. It converts the signal data from binary format into a software-processable data structure and then sends it to the data preprocessing unit 4082 for subsequent data preprocessing. The preprocessing unit 4082 is mainly used to convert the original data into a preprocessed data structure. The ECG, PPG, and BCG data are processed to meet the model input requirements. First, referring to the correspondence between ECG, PPG, and BCG waveforms and cardiac physiological indicators, the R wave of ECG, the trough of PPG, and the J wave of BCG are aligned in the time domain. Then, the ECG, PPG, and BCG signals are segmented into heartbeat-level data according to the above alignment points and paired. Normalization transformation is performed on the heartbeat-level data segments. The signal type determination unit 4083 determines the number of reconstructed source channels using an adaptive method. In this embodiment, the number of immersive sensing signals used is 2. The ECG channel undergoes a two-stage wavelet transform. ECG where φ a,b and ψ a,bLet represent the approximation function and detail function of the Bb5 wavelet basis function with scale 'a' and translation 'b', respectively. N represents the length of the ECG signal. In this embodiment, the wavelet transform translation is 0, and the transform scales are 1 and 2. First, the original ECG signal is used as input, and a transform with scale 1 is performed, outputting approximation components and first-level detail components. Then, using the approximation components as input, a transform with scale 2 is performed, outputting second-level approximation components and second-level detail components. Zero-value interpolation is performed on the second-level approximation components, first-level detail components, and second-level detail components to expand them into ECG data of equal length. Wavelet reconstruction is then performed, ultimately converting the data into ECG data composed of three channels: second-level approximation components, first-level detail components, and second-level detail components. The output data structure conforming to the reconstruction model interface format requirements is sent to the model building unit 4084. The model building unit 4084 mainly initializes the basic operator layer of the Flow-based ECG reconstruction model and connects the operator layers. The model consists of multiple Flow model basic building blocks TimeFlow and 1x1conv convolutional layers. Let x... ECG x BCG x PPG Let z represent the ECG signal path, BCG signal path, and PPG signal path, respectively. Let z represent the interpolation information path, where z ~ p(z) is the preset probability distribution of the interpolation information. In this embodiment, a Gaussian distribution is chosen as the preset probability distribution. ECG ~p(x ECG |x BCG x PPG The ECG signal is a conditional probability distribution dependent on the BCG and PPG signals. Let... These represent the reconstructed ECG signal path, the synthesized BCG signal path, and the synthesized PPG signal path, respectively. The Flow-based ECG reconstruction model will use x during forward propagation. ECG Mapped to And the interpolation vector z, which, during the reverse reasoning process, will be randomly sampled from a pre-defined Gaussian distribution, along with the original x. BCG x PPG Refactoring The forward propagation process can be represented as The reverse reasoning process can be represented as In this embodiment f θ A reversible base block f consisting of multiple stacked blocks i Composition, or TimeFlow. The flow of ECG data across these basic blocks can be described as follows: Where h iThis is the temporary output of the TimeFlow base block in the intermediate layer of the model. L is the number of stacked base blocks in the flow model. The more base blocks stacked, the stronger the model's representation ability, but the more time is consumed in the model training and inference process. In this embodiment, it is set to 50, which can achieve high-quality ECG reconstruction and maintain a reasonable inference time. The reverse inference process of the Flow-based reconstruction model can be described as follows: Where f i -1 This represents the reverse derivation of the TimeFlow basic block, ensuring the model's reversibility. The mapping relationships between ECG, BCG, and PPG are constructed within this reversibility framework, satisfying the correlation between the three signals in their sensing principles. The ECG signal senses the electrical activity of the heart, while both PPG and BCG sense changes in bodily organs caused by the heartbeat; the source of these changes is also the electrical activity of the heart. The data flow process of the TimeFlow basic block can be represented as follows: F(.), H(.), and G(.) are three identical sub-neural networks used to construct TimeFlow, mainly composed of recurrent neural operators and fully connected operators. In this embodiment, the recurrent neural operator is a Long Short-Term Memory (LSTM) unit. The initial operator of the sub-neural network is a recurrent neural operator with a hidden unit dimension of 200, followed by three cascaded fully connected operators. The first and second level fully connected operators have 200 units each, and the last level fully connected operator has 125 units, consistent with the dimension of the heartbeat level data of ECG, BCG, and PPG. After the model is built, the system configures the optimizer used for model training. In this embodiment, the Adam optimizer is configured, with a learning rate parameter of 0.0005 and a weight decay parameter of 0.0001. In this embodiment, the loss function of the Flow-based reconstruction model consists of three parts: supervised loss, i.e., the PPG and BCG synthesized by the model output during the model's forward inference process. The waveform should closely approximate the original PPG and BCG; the reconstruction loss, in order to generate a high-quality reconstructed ECG, is the reconstructed ECG output by the Flow-based reconstruction model during the back-inference process. It should approximate the original ECG; the stabilization loss (also known as the information entropy difference loss) is constructed using the cross-entropy of the information entropy difference vector to ensure that the Flow-based reconstruction model can stably update the model gradient during training. The total loss function used by the system is expressed as: Where ω1, ω2, ω3 are constant weight coefficients, q(ECG) is the distribution of the input ECG, and p(z) is the probability distribution of the information entropy difference vector z. This indicates that the value of z is The probability of time, where the third term, the stability loss (also known as the information entropy difference loss), represents... The similarity between the predicted distribution of z and the true distribution of z, E q(ECG) This represents the expected similarity between the probability of z, the output of the streaming model, and the actual probability of z, when all ECGs observed are taken as input and the probability distribution q(ECG) of the unknown ECG are unknown. The total loss function used by the system consists of three parts: the root mean square error between the reconstructed ECG and the original ECG, the root mean square error between the synthesized PPG and the original PPG, the root mean square error between the synthesized BCG and the original BCG, and the cross-entropy of the information entropy difference vector. The completed ECG reconstruction model, optimizer, loss function, and data packets are passed to the training decision unit 4085. If the training decision unit 4085 determines that training is required, the model training unit 4086 will then perform subsequent model training.

[0110] The model training unit 4086 first divides the preprocessed ECG, BCG, and PPG datasets into training and testing sets. In this embodiment, the ratio is 8:2. Using the training dataset as input, the Flow-based reconstruction model first performs forward propagation, outputting the synthesized BCG, synthesized PPG, and predicted interpolation vector. The system records the synthesized BCG, synthesized PPG, and predicted interpolation vector. Then, using the original BCG, PPG, and interpolation vector as input, the Flow-based reconstruction model performs reverse propagation, outputting the reconstructed ECG. The system calculates the loss function using the synthesized BCG, synthesized PPG, interpolation vector, and reconstructed ECG. Then, the system calculates the weights of the reconstructed model based on the loss and updates the gradient. The optimizer uses the gradient information to update and adjust the weights. This process constitutes the model training process. After model training is complete, the training judgment unit 4085 evaluates the model using a partitioned test set. If the model converges, training stops, and the model and updated model weights are output to the ECG reconstruction inference unit 4087 for ECG reconstruction applications. If the model fails to converge, the training process continues. In this embodiment, the ECG reconstruction inference unit 4087 primarily reads BCG and PPG signals and uses the interpolation vector obtained by the interpolation vector sampling module 4088 based on a preset interpolation vector distribution function as input to the ECG reconstruction inference unit 4087. It then performs model inference and outputs a high-fidelity reconstructed ECG for health monitoring.

[0111] like Figure 6 As shown, a preferred execution process for an ECG reconstruction method is as follows:

[0112] Step 601: The ECG acquisition terminal, PPG acquisition smartwatch, and BCG acquisition scale use their integrated sensors to acquire ECG, PPG, and BCG data respectively, filter the data, and then transmit the data to the data management module on the computer.

[0113] Step 602: The data management module on the computer receives data from the ECG acquisition terminal, PPG acquisition smartwatch, and BCG acquisition scale, and stores the collected data in the computer's memory.

[0114] Step 603: Determine whether the ECG reconstruction model needs to be trained. If it does, proceed to step 604; otherwise, proceed to step 611.

[0115] Step 604: Determine whether the amount of data in the dataset stored on the computer is sufficient for training the reconstructed model. If the amount of data is not sufficient, return to step 601 to continue collecting ECG, PPG, and BCG data. If the amount of data is sufficient, proceed to step 604 for subsequent data processing.

[0116] Step 605: The ECG reconstruction module on the computer reads the stored dataset and preprocesses the ECG, PPG, and BCG data respectively, and finally outputs data that meets the input format requirements of the ECG reconstruction model.

[0117] Step 606: The system divides the preprocessed data samples into training and test sets in an 8:2 ratio.

[0118] Step 607: Construct an ECG reconstruction model based on Flow, randomly initialize its weights, construct the Adam optimizer, and set the learning rate and weight decay parameters.

[0119] Step 608: Using the training set as input, first, the original ECG is used as input, and the ECG reconstruction model performs forward inference, outputting synthesized PPG and BCG; then, the original BCG and original PPG are used as input, and the ECG reconstruction model performs reverse inference, outputting reconstructed ECG; using the above output and original data as input to the loss function calculation formula, the loss value is calculated, and the loss value is used to calculate the model update gradient and update the weights of the reconstruction model.

[0120] Step 609: Use the test set to test whether the ECG reconstruction model has converged. If it has converged, proceed to step 610; if it has not converged, return to step 608 to continue training the ECG reconstruction model.

[0121] Step 610: Store the Flow-based reconstructed model structure and the updated ECG reconstructed model weight data on the computer.

[0122] Step 611: Enter the model deployment process. First, the ECG reconstruction model structure and ECG reconstruction model weight data stored on the computer are sent to the mobile device. After the mobile device has collected the PPG and BCG data, it uses the model to perform ECG reconstruction inference.

[0123] Step 612: PPG data acquisition: The smartwatch collects PPG signals and sends them to the mobile phone via Bluetooth. BCG data acquisition: The scale collects BCG signals and sends them to the mobile phone via Bluetooth. The data management module on the mobile phone summarizes the PPG and BCG data.

[0124] Step 613: The ECG reconstruction module on the mobile device reads PPG and BCG data and performs data preprocessing, transforming the PPG and BCG data into a data structure that meets the input requirements of the ECG reconstruction model.

[0125] Step 614: The mobile device reads the Flow-based reconstructed model structure and model weight data, and uses PPG and BCG data as input to start inference for the reconstructed model, finally outputting the reconstructed ECG (see details). Figure 8 ).

[0126] In summary, this invention proposes a complete ECG reconstruction system. This system includes the dataset acquisition process required for building the ECG reconstruction method, the training of the ECG reconstruction model, and the deployment of the model, providing a complete and feasible ECG reconstruction process. The ECG reconstruction system of this invention provides a complete dataset acquisition process, including the selection of signal acquisition equipment, the setting of acquisition parameters, and the preprocessing of acquired data, ensuring the quality and reliability of the dataset. The ECG reconstruction system of this invention uses a Flow-based technical framework to build the ECG reconstruction model. The training process includes dataset partitioning, model structure implementation, loss function setting for the Flow model, and hyperparameter design, which can improve the model's accuracy and generalization performance. The ECG reconstruction system of this invention provides a complete model deployment process, including model distribution and data flow design for the deployment environment, ensuring that the model can run stably in a real-world environment.

[0127] It should be noted that although the steps are described in a specific order above, it does not mean that the steps must be executed in the above specific order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required function can be achieved.

[0128] This invention can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0129] Computer-readable storage media can be tangible devices that hold and store instructions for use by an instruction execution device. Computer-readable storage media can include, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof.

[0130] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A training method for an electrocardiogram reconstruction model, characterized in that, The method includes: S1. Obtain a training dataset consisting of multiple training data pairs, wherein each training data pair includes ECG signal conversion data and immersive sensing signal data collected by the user at the same time. The ECG signal conversion data is obtained by converting the channel number of the ECG signal data collected by the user according to the channel number of the immersive sensing signal data, so that the channel number of the ECG signal conversion data is the channel number of the immersive sensing signal data plus 1. The channel number conversion is performed as follows: perform multiple wavelet transforms on the ECG signal data, each wavelet transform yielding an approximate component and a detail component; perform zero-value interpolation on the approximate component and detail component obtained from the last wavelet transform, as well as the detail component from the wavelet transform preceding the last wavelet transform, to obtain the ECG signal conversion data. S2. Obtain the ECG reconstruction model, which is a flow model; S3. Train the ECG reconstruction model using the training dataset. The training includes forward inference and backward inference processes. During forward inference, the ECG reconstruction model generates synthesized immersive sensing signal data and an information entropy difference vector based on the ECG signal conversion data. The information entropy difference vector indicates the information entropy difference between the ECG signal conversion data and the immersive sensing signal data. During backward inference, the ECG reconstruction model generates synthesized ECG signal conversion data based on the immersive sensing signal data and an interpolation vector randomly sampled from a preset probability distribution of the information entropy difference information. The parameters of the ECG reconstruction model are updated using the information entropy difference loss corresponding to the information entropy difference vector. The information entropy difference loss is determined based on the similarity between the probability distribution of the information entropy difference vector of each training data and a preset Gaussian probability distribution.

2. The method according to claim 1, characterized in that, During training, the parameters of the ECG reconstruction model are updated based on the total loss determined by the information entropy difference loss, the reconstruction loss between the synthesized ECG signal conversion data and the ECG signal conversion data, and the supervision loss between the synthesized immersive sensing signal data and the immersive sensing signal data.

3. The method according to claim 2, characterized in that, The integrated sensing signal data includes cardiac impaction signal data and photoplethysmography (PPG) signal data, and the total loss is determined as follows: in, Indicates the total loss. Indicates the first hyperparameter. Indicates the second hyperparameter. Indicates the third hyperparameter. Indicates the reconstruction loss. This represents the synthesized ECG signal conversion data. This represents ECG signal conversion data. Indicates monitoring losses, This represents the first supervised loss corresponding to the cardiac impaction signal data. This represents the synthesized cardiac impaction signal data. This represents the cardiac impaction signal data. This represents the second supervisory loss corresponding to the photoplethysmography (PPG) signal data. This represents the synthesized photoplethysmography (PPG) signal data. This represents photoplethysmography (PPG) signal data. This represents the loss due to differences in information entropy.

4. The method according to any one of claims 1-3, characterized in that, The electrocardiogram reconstruction model includes at least one recurrent neural network for extracting time-domain features and at least one fully connected neural network for extracting spatial-domain features.

5. An electrocardiogram (ECG) reconstruction method based on an ECG reconstruction model, characterized in that, The method includes the following steps: C1. Obtain the user's historical data, which includes the user's ECG signal data and immersive sensing signal data at multiple times; C2. The physiological signals obtained in step C1 are preprocessed according to a preset processing method to obtain training starting data consisting of multiple data pairs, wherein each data pair includes ECG signal data and immersive sensing signal data of the user in at least one heartbeat cycle. C3. The number of conversion channels is obtained by adding 1 to the number of channels of the integrated sensing signal data in step C2; C4. Convert the channel number of the ECG signal data in the data pair obtained in step C2 to obtain ECG signal conversion data. A training data pair is composed of the ECG signal conversion data and the immersive sensing signal data in a data pair. A training dataset is composed of multiple training data pairs. C5. The electrocardiogram reconstruction model is trained multiple times using the training method described in any one of claims 1-4 to obtain the trained electrocardiogram reconstruction model. C6. Obtain the user's current immersive sensing signal, preprocess the user's current immersive sensing signal according to the preset processing method to obtain the user's current immersive sensing signal data, and sample from the preset information entropy difference information probability distribution between the immersive sensing signal and the ECG signal to generate an insertion vector. C7. Input the user's current immersive sensing signal data and the sampling-generated insertion vector into the trained ECG reconstruction model to obtain the user's current reconstructed ECG signal conversion data; C8. Perform an inverse channel number transformation on the user's current reconstructed ECG signal conversion data to obtain the reconstructed ECG signal data.

6. The method according to claim 5, characterized in that, The preset processing method includes one or more of the following methods: filtering, alignment, segmentation, and normalization.

7. An electrocardiogram reconstruction system, characterized in that, The system includes: An information collection terminal is configured to collect a user's physiological signals, and to perform noise reduction processing on the collected physiological signals and output them; wherein, the physiological signals include ECG signals and immersive sensing signals; A data management module is configured to receive data output by the information collection terminal and process it according to a preset data structure; A database configured to store data processed by the data management module; A reconstruction module is configured to perform the method as described in claim 5 or 6 to obtain reconstructed ECG signal data.

8. The system according to claim 7, characterized in that, The reconstruction module includes: The data reading unit is configured to acquire and output the training physiological signal of the detection target, wherein the training physiological signal is the ECG signal and immersive sensing signal of the detection target in multiple consecutive heartbeat cycles; The data preprocessing unit is configured to receive the physiological signals output by the data reading unit, preprocess them according to a preset processing method to obtain and output training starting data composed of data pairs of the detection target in each heartbeat cycle, wherein each data pair is ECG signal data and immersive sensing signal data of the detection target in the same heartbeat cycle. The signal type determination unit is configured to: use the number of channels of the immersive sensing signal data plus 1 as the number of conversion channels, and convert ECG signal data into ECG signal conversion data, wherein the number of channels of the ECG signal conversion data is equal to the number of conversion channels; The model building unit is configured to store a pre-initialized electrocardiogram reconstruction model. The training judgment unit is configured to: determine whether the ECG reconstruction model stored in the model building unit needs to be trained, and output the training judgment result; The model training unit is configured to: when the training judgment unit outputs a training judgment result indicating that training is required, execute the training method as described in any one of claims 1-4 to update the parameters of the model in the model building unit; The insertion vector sampling module is configured to generate an insertion vector by sampling from a preset probability distribution of information entropy difference information between the immersive sensing signal and the ECG signal. The reconstruction inference unit is configured to: when the output result of the training judgment unit is that no training is required, use the ECG reconstruction model stored in the model building unit as the trained ECG reconstruction model, and obtain the immersive sensing signal in the physiological signal of the detection target and process it according to the preset processing method to obtain the immersive sensing signal data, and obtain the insertion vector obtained by the insertion vector sampling module, and use the trained ECG reconstruction model to perform ECG reconstruction.

9. A computer-readable storage medium, characterized in that, It stores a computer program that can be executed by a processor to implement the steps of the method according to any one of claims 1 to 6.

10. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the electronic device to perform the steps of the method as described in any one of claims 1 to 6.

Images