Conversion model for generating ECG signal from PPG signal

Abstract

The invention discloses a conversion model for generating an ECG signal from a PPG signal, which comprises a period sensing module, a Transform extraction framework and an adaptive aggregation module, and is characterized in that the period sensing module is used for identifying and extracting a periodic component in an originally input PPG signal, determining a frequency related to a cardiac cycle, remodeling an input time sequence, and generating an ECG signal in the PPG signal; obtaining a two-dimensional periodic view formed by k different periods; a Transform extraction framework divides a two-dimensional period view formed by k different periods into a plurality of blocks Patchi corresponding to cardiac periods, and sequentially inputs each block Patchi into an embedding layer and a coding layer for processing, and captures output representation X ''i changing in the periods and between the periods in PPG signal data, and a feature sequence # imgabs0 # matched with the length of the original PPG signal is obtained after remodeling, and the adaptive aggregation module adopts a weight aggregation method to aggregate the feature sequence # imgabs1 # to obtain a reconstructed ECG signal X. According to the method, physiological feature capture is enhanced through signal alignment based on the heart cycle, so that the accuracy of generating the ECG signal is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrocardiograms, and in particular relates to a conversion model for generating an ECG signal from a PPG signal. Background Art

[0002] Electrocardiography (ECG), as a standard method for cardiac monitoring, is widely used in clinical diagnosis because it can provide comprehensive cardiac information. However, traditional ECG acquisition usually requires the sensor to be attached to the chest and may stimulate the skin through electrodes, which not only causes discomfort to the subject but also makes long-term monitoring difficult, thus limiting its applicability in daily health monitoring. Although some researchers have proposed more comfortable wearable devices, these solutions still focus on sensor design and have not yet been put into practical use in daily cardiac monitoring.

[0003] Photoplethysmography (PPG) has become an attractive solution for daily cardiac monitoring. Currently, the effectiveness of using PPG to estimate heart rate has been demonstrated. PPG is a non-invasive optical signal that measures changes in blood volume in the skin, usually through a sensor worn on the wrist (such as a smart watch), to continuously monitor the heart state. PPG is easier to obtain and more user-friendly than ECG, but it cannot provide detailed cardiac activity such as P waves or QRS complexes that ECG has, limiting its clinical application.

[0004] PPG and ECG signals have physiological correlation, ECG captures the electrophysiological activity of the heart, while PPG measures its mechanical movement, both reflecting the same cardiac process. In addition, the changes in peripheral blood volume recorded by PPG are directly affected by myocardial activity, which is controlled by the electrical signals of the sinoatrial node (SA). The inherent correlation between PPG and ECG has inspired research on converting PPG signals into ECG signals, combining the advantages of both to promote continuous and long-term cardiac monitoring without compromising diagnostic accuracy.

[0005] Existing PPG to ECG conversion studies rely on peak detection or simple mapping techniques, which often ignore the inherent periodicity of cardiovascular signals. This periodicity is crucial for generating high-fidelity ECG from PPG signals. On the one hand, PPG and ECG represent two signals of the same cardiac cycle. They are both generated by the rhythmic activity of the heart, and their onset, duration, and termination are closely related to the periodic fluctuations of the heart. Therefore, ignoring the periodic information will directly affect the accuracy of the generated ECG. On the other hand, accurate conversion requires retaining the physiological information of the signal when reconstructing the ECG signal from the PPG signal. If the characteristics of the period are ignored, key features such as the QRS complex wave, RR interval, and ST segment deviation cannot be accurately reconstructed, thereby reducing the clinical value of the generated ECG signal. Summary of the invention

[0006] In order to solve the above-mentioned technical problems, the present invention provides a conversion model for generating ECG signals from PPG signals, aiming to enhance the adaptability of input data modeling and strengthen the capture of physiological characteristics through signal alignment based on the cardiac cycle, thereby improving the accuracy of generated ECG signals and providing a convenient and effective heart monitoring technology.

[0007] The technical solutions adopted are as follows:

[0008] A conversion model for generating an ECG signal from a PPG signal, the conversion model comprising:

[0009] The cycle perception module is used to identify and extract the periodic components in the original input PPG signal, determine the frequency related to the cardiac cycle, reshape the input time series, and obtain a two-dimensional cycle view formed by k different cycles;

[0010] Transformer extraction framework, which divides the two-dimensional periodic view formed by k different cycles into multiple patches corresponding to the cardiac cycle i , and each block Patch i The output representation X″ is input into the embedding layer and the encoding layer in sequence to capture the intra-cycle and inter-cycle changes in the PPG signal data. i , and then reshaped to obtain a characteristic sequence that matches the length of the original PPG signal

[0011] Adaptive aggregation module, which uses weighted aggregation method to aggregate feature sequences Get the aggregated feature X;

[0012] Single-layer perceptron,The aggregated feature X is input into the linear layer of the single-layer perceptron,,which is linearly transformed, and the features extracted from the PPG signal are mapped to the ECG signal representation,,and the reconstructed ECG signal is obtained.

[0013] Specifically, the cycle perception module uses fast Fourier transform to identify and extract the periodic components in the original input PPG signal and determine the frequency associated with the basic cardiac cycle, which is expressed as:

[0014] A=Amp(FFT(X))

[0015] Among them: FFT(·) is the process of rapidly transforming a signal to determine its frequency components;

[0016] Amp(·) indicates the calculated amplitude;

[0017] is an amplitude array, where the jth value A j Describes the intensity of the periodic basis function with frequency j, corresponding to The length of the cycle;

[0018] Raw Input L is the length of the time series.

[0019] Furthermore, the largest k amplitude values ​​are selected from the PPG signal, and the corresponding frequencies {f1, ..., f k},amplitude Select k period lengths {c1, ..., c k}; get the period length {c1, c2, ..., c k} and frequencies {f1, f2, ..., f k}, construct the input data Cycle 2D view of the cycle i , whose expression is:

[0020]

[0021] in, is based on the frequency f i The i-th reshaped time series, the rows and columns represent the corresponding period lengths c i intra- and inter-cycle variations in the following;

[0022] Padding(·) is the filling constructor; Reshape(·) is the reshaping function.

[0023] Preferably, the cycle two-dimensional view Cycle i Reshape it to get a two-dimensional periodic view formed by k different periods, and its expression is:

[0024]

[0025] Will Reshape Where T represents a predetermined hyperparameter;

[0026] The resulting set of patches {Patch1, ..., Patch k} is a two-dimensional view formed by k different periods.

[0027] Preferably, multiple blocks Patch i Input into the embedding layer in the Transformer extraction framework to obtain the embedded representation where c i It's a patch i The period length, d model is the dimension of the model, which is expressed as:

[0028] X′ i =Embedding(Patch i ),i∈{1,2,…,k}

[0029] The resulting embedding is represented as X′ i After being processed by the encoding layer in the Transformer extraction framework, the output representation is obtained Its expression is:

[0030] X″ i = TransformerEncoder(X′ i ),i∈{1,2,…,k}

[0031] After reshaping by truncation method, the characteristic sequence matching the length of the original PPG signal is obtained. Its expression is:

[0032] Preferably, the encoding layer in the Transformer extraction framework adopts a dual mechanism to embed the representation X ′ i For processing, on the one hand, the self-attention mechanism is used to capture the inter-cycle dependencies in the PPG signal; on the other hand, a feedforward network is used to model the intra-cycle dependencies in the PPG signal to capture the detailed local features within each cycle.

[0033] Further preferably, a weighted aggregation method is used to aggregate k feature sequences The aggregated feature X is expressed as:

[0034]

[0035]

[0036] The technical solution of the present invention has the following advantages:

[0037] A. The PPG-to-ECG signal conversion model provided by the present invention enhances the adaptability of input data modeling and strengthens the capture of physiological characteristics through signal alignment based on the cardiac cycle, thereby improving the accuracy of generating ECG signals and providing convenient and effective cardiac monitoring.

[0038] B. Compared with the prior art, the heart monitoring technology designed by the present invention for generating ECG signals based on PPG can reconstruct the PPG signals acquired by wearable devices into ECG signals with more comprehensive information. Compared with the existing methods, the conversion model proposed by the present invention can use simpler and more portable PPG acquisition devices (such as electronic watches) to replace complex ECG acquisition devices in the signal acquisition stage, and obtain more real and comprehensive ECG signals with the same information for heart monitoring.

[0039] C. The present invention performs supervised conversion model training on the PPG-ECG paired data set, and migrates it to the PPG data set to generate ECG signals, and finally realizes multi-dimensional generation effect evaluation and analysis from two dimensions: upstream and downstream. Through this research, the present invention can effectively combine the convenience and speed of PPG equipment, capture physiological characteristics within the cycle, infer accurate and clinically valuable ECG signals, and finally achieve more accurate heart monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the specific embodiments of the present invention, the drawings required for use in the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 The present invention provides a periodic PPG-to-ECG signal conversion model;

[0042] Figure 2 This is a heart monitoring technology framework and flow chart for generating ECG signals based on PPG provided by the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] like Figure 1As shown, the present invention provides a conversion model for generating ECG signals from PPG signals. The conversion model includes four parts: a cycle perception module, a Transformer extraction framework, an adaptive aggregation module, and a single-layer perceptron. The cycle perception module is used to identify and extract the periodic components in the original input PPG signal, determine the frequency related to the cardiac cycle, reshape the input time series, and obtain a two-dimensional cycle view formed by k different cycles; the Transformer extraction framework divides the two-dimensional cycle view formed by k different cycles into multiple patches corresponding to the cardiac cycle. i , and each block Patch i The output representation X″ is input into the embedding layer and the encoding layer in sequence to capture the intra-cycle and inter-cycle changes in the PPG signal data. i , and then reshaped to obtain a characteristic sequence that matches the length of the original PPG signal The adaptive aggregation module uses weighted aggregation method to aggregate feature sequences The aggregated feature X is obtained; the single-layer perceptron inputs the aggregated feature X into the linear layer of the single-layer perceptron, performs a linear transformation on it, maps the features extracted from the PPG signal to the ECG signal representation, and obtains the reconstructed ECG signal.

[0045] The following mainly describes the three parts of the period perception module, Transformer extraction framework and adaptive aggregation module one by one.

[0046] 1) Cycle Perception Module: Considering the potential periodicity of cardiovascular activity is crucial to generating a more realistic ECG from PPG. The main goal of this invention is to detect and extract these periodic components. Fast Fourier Transform (FFT) is used to identify and extract periodic components in the input signal, which can accurately determine the frequency associated with the basic cardiac cycle. L is the length of the time series. The FFT formula used is as follows:

[0047] A=Amp(FFT(X))

[0048] Where FFT(·) refers to the process of rapidly transforming a signal to determine its frequency components, and Amp(·) means calculating the amplitude. is an amplitude array, where the jth value A j Describes the intensity of the periodic basis function with frequency j, corresponding to The cycle length.

[0049] At the same time, considering the sparsity of the frequency domain, in order to avoid high-frequency noise, the present invention preferably selects only the largest k amplitude values ​​and finds the corresponding most important frequencies {f1, ..., f k} and its amplitude where k is a hyperparameter. These selected frequencies correspond to k cycle lengths {c1, ..., c k}.

[0050] Due to the conjugate nature of the frequency domain, only frequencies within the range The equation used is as follows:

[0051]

[0052]

[0053] Among them, argTopk(·) is the top K index selection function, which refers to the index of the largest k amplitudes from a set of amplitudes A, and the index corresponds to the frequency. k} and frequencies {f1, f2, ..., f k}, construct the input data Specifically, the input time series is adaptively reshaped into a two-dimensional tensor according to the following formula:

[0054]

[0055] in, Represents frequency f i The i-th reshaped time series of , which semantically corresponds to the cardiac cycle view. The rows and columns represent the corresponding cycle lengths c i It should be noted that if and If there is no match, it is filled with zeros.

[0056] Furthermore, the period extracted via FFT will fluctuate in real time, so The characteristic dimension f i Will follow the input In order to facilitate its input into the embedding layer, this variable-size Cycle must be i Convert to blocks of fixed feature dimension.

[0057] The present invention uses the last value to fill in to make up for the deficiency, or cuts off the excess part to achieve this. The specific process is as follows:

[0058]

[0059] Padding(·) is a filling constructor. If the length is less than T, the last value is used to fill it. Trunc(·) is a truncation function. If the length exceeds T, it is truncated directly.

[0060] Accordingly, the present invention will Reshape where T represents a pre-determined hyperparameter. The final result is a set of patches {Patch1, ..., Patch k}, which represents a two-dimensional view formed by k different cycles, and the process is as follows Figure 1 As shown in the figure on the right.

[0061] 2) Cycle-aware Transformer Extraction Framework

[0062] Transformer is used as the main feature extraction architecture because of its ability to capture long-range dependencies and local features in the data. The uniqueness of the design of this invention lies in the cycle-based PPG signal block, which divides the PPG signal into multiple blocks corresponding to the cardiac cycle. This adaptive partitioning ensures that the model accurately captures the changes within and between cycles, which is crucial for accurate physiological analysis.

[0063] The adaptively divided patches form a multi-resolution, multi-branch network, where each branch processes the calculation as follows:

[0064] In the Transformer extraction framework, an embedding layer and an encoding layer are set to transform multiple patches i Input into the embedding layer in the Transformer extraction framework to obtain the embedded representation where c i It's a patch i The period length, d model is the dimensionality of the model. This embedding process maps the original input to a higher dimensional space, making it easier for Transformer to capture complex patterns in the data. Its expression is:

[0065] X′ i =Embedding(Patch i ), i∈{1, 2, …, k}

[0066] Embedding(·) is the embedding layer function, which transforms the block Patch i The signal is converted into a dense continuous vector representation that captures the semantic relationship of the input items.

[0067] The resulting embedding is represented as X′ iAfter being processed by the encoding layer in the Transformer extraction framework, the self-attention mechanism in the encoding layer focuses on capturing inter-cycle dependencies (the relationship between different cardiac cycles) by simultaneously attending to various parts of the signal. On the other hand, a feed-forward network is used to model intra-cycle dependencies and capture detailed local features within each cycle. This dual mechanism enables the Transformer to learn a rich set of features that represent global trends and local changes in the PPG signal. After being processed by the Transformer, the output representation X″ i The size is Its expression is:

[0068] X″ i = TransformerEncoder(X′ i ), i∈{1, 2, …, k}

[0069] TransformerEncoder(·) is the encoder function that encodes the input using a multi-head self-attention mechanism and a feed-forward network to capture global dependencies in the input data.

[0070] In order to combine k features {X″1,...,X″ k} to a format suitable for further analysis or forecasting, and to restore the time series shape of the signal, do the following:

[0071]

[0072] The above is done by reshaping the two-dimensional representation into a one-dimensional feature sequence This is then accomplished by truncation to ensure that the final sequence length matches the original signal length L.

[0073] 3) Adaptive aggregation module

[0074] Amplitude {A1, ..., A k} reflects the relative importance of the most prominent k cycles, corresponding to the importance of patches at different resolutions. For example, when the i-th frequency A in the PPG signal i When the amplitude is large, it indicates that the corresponding period c f is more important, therefore, the relevant scale features are given greater weight w i .

[0075] The calculation process can be formalized as follows:

[0076]

[0077] {ω1, ..., ω k}=Softmax({A1,...,A k})

[0078] Topk(·) is a top-K selection function that selects the top K largest amplitudes from the input data; Softmax(·) is a normalized exponential function that converts the input vector into a probability distribution, and each value of the output is between 0 and 1, and the sum is 1.

[0079] Therefore, the present invention aggregates k features according to the weights The aggregated feature X is obtained and used for transformation in the linear layer of a single-layer perceptron. The expression formula used is as follows:

[0080]

[0081] in is the feature sequence representation learned based on the i-th cycle, and ω i express Since the weights are derived from the amplitudes of the input frequencies, this adaptive aggregation method ensures that the weighting process is closely coupled to the signal characteristics.

[0082] The aggregated feature X is input into the linear layer of the single-layer perceptron and linearly transformed, thereby mapping the feature X extracted from the PPG signal to the ECG signal representation, and realizing the conversion of the PPG signal to the ECG signal. The present invention generates ECG signals based on PPG signals for cardiac monitoring, and can reconstruct the PPG signals obtained by the wearable device into ECG signals with more comprehensive information. Compared with the existing methods, the present invention can use simpler and more portable PPG acquisition devices, such as electronic watches, to replace complex ECG acquisition devices during the signal acquisition stage, and obtain more real and comprehensive ECG signals with the same information for cardiac monitoring.

[0083] like Figure 2 As shown, the present invention designs a PPG-to-ECG conversion model based on the cycle. First, supervised training is performed on the PPG-ECG paired data set, and then it is transferred to the PPG data set for ECG signal generation, and finally multi-dimensional generation effect evaluation and analysis is achieved from the upstream and downstream dimensions. Through this study, the conversion model provided by the present invention can effectively combine the convenience and speed of PPG equipment, capture the physiological characteristics within the cycle, obtain accurate and clinically valuable ECG signals through reasoning, generate ECG signals through PPG signals, and use them for cardiovascular task analysis, and finally achieve more accurate heart monitoring.

[0084] Anything not described in the present invention is applicable to the prior art.

[0085] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. However, the obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A conversion model for generating an ECG signal from a PPG signal, characterized in that: The conversion model includes: The cycle perception module is used to identify and extract the periodic components in the original input PPG signal, determine the frequency related to the cardiac cycle, reshape the input time series, and obtain a two-dimensional cycle view formed by k different cycles; Transformer extraction framework, which divides the two-dimensional periodic view formed by k different cycles into multiple patches corresponding to the cardiac cycle i , and each block Patch i The output representation X″ is input into the embedding layer and the encoding layer in sequence to capture the intra-cycle and inter-cycle changes in the PPG signal data. i , and then reshaped to obtain a characteristic sequence that matches the length of the original PPG signal Adaptive aggregation module, which uses weighted aggregation method to aggregate feature sequences Get the aggregated feature X; Single-layer perceptron,The aggregated feature X is input into the linear layer of the single-layer perceptron,,which is linearly transformed, and the features extracted from the PPG signal are mapped to the ECG signal representation,,and the reconstructed ECG signal is obtained.

2. The conversion model for generating an ECG signal from a PPG signal according to claim 1, characterized in that: The cycle perception module uses fast Fourier transform to identify and extract the periodic components in the original input PPG signal and determine the frequency associated with the basic cardiac cycle, which is expressed as: A=Amp(FFT(X)) Among them: FFT(·) is the process of rapidly transforming a signal to determine its frequency components; Amp(·) indicates the calculated amplitude; is an amplitude array, where the jth value A j Describes the intensity of the periodic basis function with frequency j, corresponding to The length of the cycle; Raw Input L is the length of the time series.

3. The conversion model for generating an ECG signal from a PPG signal according to claim 2, characterized in that: Select the largest k amplitude values ​​from the PPG signal and find the corresponding frequencies {f1,…,f k },amplitude Select k period lengths {c1,…,c k }; get the period length {c1,c2,…,c k } and frequencies {f1,f2,…,f k }, construct the input data Cycle 2D view of the cycle i , whose expression is: in, is based on the frequency f i The i-th reshaped time series, the rows and columns represent the corresponding period lengths c i intra- and inter-cycle variations in the following; Padding(·) is the filling constructor; Reshape(·) is the reshaping function.

4. The conversion model for generating an ECG signal from a PPG signal according to claim 3, characterized in that: Cycle 2D View i Reshape it to get a two-dimensional periodic view formed by k different periods, and its expression is: Will Reshape Where T represents a predetermined hyperparameter; The resulting set of patches {Patch1,…,Patch k } is a two-dimensional view formed by k different periods.

5. The conversion model for generating an ECG signal from a PPG signal according to claim 1, characterized in that: Patching multiple blocks i Input into the embedding layer in the Transformer extraction framework to obtain the embedded representation where c i It's a Patch i The period length, d model is the dimension of the model, which is expressed as: X′ i =Embedding(Patch i ),i∈{1,2,…,k} The resulting embedding is represented as X′ i After being processed by the encoding layer in the Transformer extraction framework, the output representation is obtained Its expression is: X″ i =TransformerEncoder(X′ i ),i∈{1,2,…,k} After reshaping by truncation method, the characteristic sequence matching the length of the original PPG signal is obtained. Its expression is:

6. The conversion model for generating an ECG signal from a PPG signal according to claim 5, characterized in that: The encoding layer in the Transformer extraction framework adopts a dual mechanism to embed the representation X′ i For processing, on the one hand, the self-attention mechanism is used to capture the inter-cycle dependencies in the PPG signal; on the other hand, a feedforward network is used to model the intra-cycle dependencies in the PPG signal to capture the detailed local features within each cycle.

7. The conversion model for generating an ECG signal from a PPG signal according to claim 1, characterized in that: Aggregate k feature sequences using weighted aggregation method Get the aggregated feature X, Its expression is:

Images