Electrocardio quality improving system based on lightweight electrocardio quality evaluation

Abstract

The invention discloses an electrocardio quality improving system based on lightweight electrocardio quality evaluation, and belongs to the technical field of electrocardio systems.The system comprises wearable composite eighteen-lead electrocardio equipment, and in the training process, the wearable composite eighteen-lead electrocardio equipment is in communication connection with an electrocardio reconstruction model training unit; in the using process, the wearable composite eighteen-lead electrocardio equipment is electrically connected with the electrocardio signal quality evaluation unit, and the electrocardio signal quality evaluation unit is in communication connection with the electrocardio signal reconstruction unit; the wearable composite eighteen-lead electrocardio device is used for collecting electrocardio signals, the wearable composite eighteen-lead electrocardio device sends the collected electrocardio signals to the electrocardio reconstruction model training unit in the training process, and a trained electrocardio self-adaptive reconstruction neural network is obtained and integrated to the electrocardio signal reconstruction unit; by means of the mode, reconstruction of the low-quality electrocardiosignals is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrocardiogram systems, and particularly relates to an electrocardiogram quality improvement system based on lightweight electrocardiogram quality assessment. Background Art

[0002] In the past, the method for improving the quality of dynamic electrocardiogram signals was electrocardiogram quality assessment (EAG signal quality assessment, ECG-SQA). Although this method improved the signal quality, filtering out low-quality signals would result in obvious blanks in electrocardiogram monitoring, which might lead to missed detection of abnormal electrocardiograms. Therefore, after ECG-SQA, it is crucial to adaptively synthesize low-quality electrocardiograms to truly improve the reliability of wearable electrocardiogram monitoring.

[0003] Based on this, the present invention designs an electrocardiogram quality improvement system based on lightweight electrocardiogram quality assessment to solve the above problems. Summary of the Invention

[0004] In view of the above-mentioned drawbacks of the prior art, the present invention provides an electrocardiogram quality improvement system based on lightweight electrocardiogram quality assessment.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0006] An electrocardiogram quality improvement system based on lightweight electrocardiogram quality assessment includes a wearable composite eighteen-lead electrocardiogram device, and the wearable composite eighteen-lead electrocardiogram device is electrically connected to an electrocardiogram reconstruction model training unit; the wearable composite eighteen-lead electrocardiogram device is electrically connected to an electrocardiogram signal evaluation unit, the electrocardiogram signal evaluation unit is communicatively connected to an electrocardiogram signal reconstruction unit, and the electrocardiogram reconstruction model training unit is deployed and connected to the electrocardiogram signal reconstruction unit;

[0007] The wearable composite eighteen-lead electrocardiogram device is used for collecting lead electrocardiogram signals, and the wearable composite eighteen-lead electrocardiogram device sends the lead electrocardiogram signals to the electrocardiogram reconstruction model training unit and the electrocardiogram signal evaluation unit;

[0008] The electrocardiogram reconstruction model training unit is used for training an electrocardiogram adaptive reconstruction neural network for the lead electrocardiogram signals;

[0009] The electrocardiogram signal evaluation unit is used for lightweight electrocardiogram quality assessment, giving high and low quality labels for the lead electrocardiogram signals, and the electrocardiogram signal evaluation unit sends the quality labels and electrocardiogram signals of the composite eighteen-lead electrocardiogram to the electrocardiogram signal reconstruction unit;

[0010] The electrocardiogram signal reconstruction unit reconstructs low-quality lead electrocardiogram signals using high-quality lead electrocardiogram signals based on the electrocardiogram adaptive reconstruction neural network trained by the electrocardiogram reconstruction model training unit according to the received quality tags of the composite 18-lead electrocardiogram.

[0011] Furthermore, the wearable composite 18-lead electrocardiogram device is provided with 18 leads, including 12 conventional leads and 6 additional leads; the wearable composite 18-lead electrocardiogram device sends the electrocardiogram signals of the 18 leads to the electrocardiogram signal evaluation unit.

[0012] Furthermore, the 12 conventional leads include I, II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5, and V6.

[0013] Furthermore, the 6 additional leads include V3R, V4R, V5R, E, S, and I.

[0014] Furthermore, based on the electrocardiogram quality assessment using permutation entropy, a signal quality threshold based on permutation entropy is formulated for each lead in the composite 18-lead electrocardiogram.

[0015] Furthermore, the lightweight electrocardiogram quality assessment uses permutation entropy (PE) assessment;

[0016] The electrocardiogram signal evaluation unit compares the PE value of each lead electrocardiogram signal with the PE threshold based on permutation entropy, finds out the lead electrocardiogram signals higher than the PE threshold and regards them as high-quality lead electrocardiogram signals, and finds out the lead electrocardiogram signals lower than the PE threshold and regards them as low-quality lead electrocardiogram signals;

[0017] The PE value is a measure of the signal-to-noise ratio or waveform integrity of the lead electrocardiogram signal.

[0018] Furthermore, the electrocardiogram adaptive reconstruction neural network (EASNN) in the electrocardiogram reconstruction model training unit includes a lead masking time-lead encoding module, a cardiac information extraction encoder, and a masked lead electrocardiogram mapping generation decoder;

[0019] The lead masking time-lead encoding module is divided into two parts. The first part randomly masks the leads of the input electrocardiogram. The masked lead electrocardiogram is regarded as a low-quality lead electrocardiogram, and the unmasked lead electrocardiogram is regarded as a high-quality lead electrocardiogram, enabling the EASNN to reconstruct the masked lead electrocardiogram through the unmasked lead electrocardiogram, which is used to train the electrocardiogram reconstruction performance of the EASNN. The second part performs time-lead position encoding and data encoding on the 18-lead electrocardiogram after the masking operation;

[0020] The cardiac information extraction encoder calculates the correlation relationship between electrocardiogram leads through an attention mechanism, thereby extracting the cardiac information in the unoccluded electrocardiogram signal to obtain electrocardiogram vector features;

[0021] The occluded lead electrocardiogram mapping generation decoder maps the obtained electrocardiogram vector features onto the occluded leads, reconstructs the occluded electrocardiogram signal, and calculates the error between the reconstructed signal and the truly recorded electrocardiogram to optimize the model.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: When the present invention is used, through the eighteen-lead electrocardiogram system composed of 16 electrodes of the wearable composite eighteen-lead electrocardiogram device, rich cardiac information can be collected;

[0023] Through the model training unit, low-quality electrocardiogram signals are simulated and large model reconstruction training is carried out to obtain the trained electrocardiogram signal reconstruction unit, realizing the adaptive reconstruction function of the eighteen-lead electrocardiogram signal;

[0024] When the present invention is used, the electrocardiogram signal evaluation unit realizes the identification of low-quality lead electrocardiogram signals in the eighteen-lead electrocardiogram signals collected by the wearable composite eighteen-lead electrocardiogram device;

[0025] When the present invention is used, the trained electrocardiogram signal reconstruction unit realizes the adaptive reconstruction of the low-quality lead electrocardiogram signals identified by the electrocardiogram signal evaluation unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0027] Figure 1 It is a block diagram of an electrocardiogram quality improvement system based on lightweight electrocardiogram quality assessment according to the present invention;

[0028] Figure 2 It is a block diagram of the electrocardiogram reconstruction model training unit of the present invention.

[0029] The reference numerals in the drawings respectively represent:

[0030] 1. Wearable composite eighteen-lead electrocardiogram device 2. Electrocardiogram reconstruction model training unit 21. Random occlusion data encoding 22. Cardiac information extraction encoder 23. Occluded lead electrocardiogram generation mapping decoder 3. Electrocardiogram signal evaluation unit 4. Electrocardiogram signal reconstruction unit. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] Embodiment 1: In some embodiments, refer to Figure 1 - Figure 2 of the accompanying drawings of the specification. A cardiac electrical quality improvement system based on lightweight cardiac electrical quality assessment, where the wearable composite eighteen-lead cardiac electrical device 1 is electrically connected to the cardiac electrical reconstruction model training unit 2; during use, the wearable composite eighteen-lead cardiac electrical device 1 is electrically connected to the cardiac electrical signal evaluation unit 3, the cardiac electrical signal evaluation unit 3 is communicatively connected to the cardiac electrical signal reconstruction unit 4, and the cardiac electrical reconstruction model training unit 2 is deployed and connected to the cardiac electrical signal reconstruction unit 4;

[0033] The wearable composite eighteen-lead cardiac electrical device 1 is used for collecting lead cardiac electrical signals, and the wearable composite eighteen-lead cardiac electrical device 1 sends the lead cardiac electrical signals to the cardiac electrical reconstruction model training unit 2 and the cardiac electrical signal evaluation unit 3;

[0034] The cardiac electrical reconstruction model training unit 2 is used for training the cardiac electrical adaptive reconstruction neural network for the lead cardiac electrical signals;

[0035] The cardiac electrical signal evaluation unit 3 is used for lightweight cardiac electrical quality assessment, giving high and low quality labels for the lead cardiac electrical signals, and the cardiac electrical signal evaluation unit 3 sends the quality labels and cardiac electrical signals of the composite eighteen-lead cardiac electricity to the cardiac electrical signal reconstruction unit 4;

[0036] The cardiac electrical signal reconstruction unit 4, according to the received quality labels of the composite eighteen-lead cardiac electricity, based on the cardiac electrical adaptive reconstruction neural network trained by the cardiac electrical reconstruction model training unit 2, uses the high-quality lead cardiac electrical signals to reconstruct the low-quality lead cardiac electrical signals.

[0037] When the present invention is used, through the eighteen-lead system composed of the eighteen electrodes of the wearable composite eighteen-lead cardiac electrical device 1, rich cardiac information can be collected;

[0038] Through the cardiac electrical reconstruction model training unit 2, the simulation of low-quality cardiac electrical signals in the eighteen-lead cardiac electrical signals is realized and large model reconstruction training is carried out;

[0039] When the present invention is used, through the cardiac electrical signal evaluation unit 3, the identification of low-quality cardiac electrical signals in the eighteen-lead cardiac electrical signals collected by the wearable composite eighteen-lead cardiac electrical device 1 is realized;

[0040] When the present invention is in use, the electrocardiogram (ECG) signal reconstruction unit 4 zeros the low-quality ECG signal and reconstructs the zeroed low-quality ECG signal based on the adaptive reconstruction neural network trained by the ECG reconstruction model training unit 2.

[0041] During the training process, the wearable composite eighteen-lead ECG device 1 is used for collecting ECG signals. During the training process, the wearable composite eighteen-lead ECG device 1 sends the collected ECG signals to the ECG reconstruction model training unit 2 to obtain the trained ECG adaptive reconstruction neural network, which is integrated into the ECG signal reconstruction unit 4. During the use process, the wearable composite eighteen-lead ECG device 1 sends the collected ECG signals to the ECG signal quality evaluation unit 3, and the ECG signal quality evaluation unit 3 sends the quality label of the composite eighteen-lead ECG and the ECG signal to the ECG signal reconstruction unit 4, and the ECG signal reconstruction unit 4 reconstructs the low-quality lead ECG through the high-quality lead ECG.

[0042] Based on the quality label of the composite eighteen-lead ECG given by the lightweight ECG quality evaluation system and the ECG adaptive reconstruction neural network EASNN trained by the ECG reconstruction model training unit 2, the low-quality lead ECG signal is reconstructed using the high-quality lead ECG signal.

[0043] The wearable composite eighteen-lead ECG device 1 is provided with eighteen leads, including twelve conventional leads and six additional leads;

[0044] The twelve conventional leads include I, II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5, and V6;

[0045] The six additional leads include V3R, V4R, V5R, E, S, and I;

[0046] The wearable composite eighteen-lead ECG device 1 sends the eighteen-lead ECG signals to the ECG signal evaluation unit 3;

[0047] The lightweight ECG quality evaluation uses permutation entropy (PE) evaluation;

[0048] The ECG signal evaluation unit 3 compares the PE value of each lead ECG signal with the PE threshold based on permutation entropy, and finds out the lead ECG signals higher than the PE threshold and regards them as low-quality lead ECG signals;

[0049] The PE value is a measure of the signal-to-noise ratio or waveform integrity of the lead ECG signal.

[0050] The ECG reconstruction model training unit 2 includes a lead masking time-lead coding module 21, a cardiac information extraction encoder 22, and a masked lead ECG mapping generation decoder 23;

[0051] The randomly masked data encoding 21 is divided into two parts. The first part randomly masks the leads of the input electrocardiogram (ECG) to enable the model to reconstruct the ECG of the masked leads, thereby training the ECG reconstruction performance of the model. The second part performs position encoding and data encoding on the 18-lead ECG after masking;

[0052] The cardiac information extraction encoder 22 calculates the correlation relationship between ECG leads through the attention mechanism, thereby extracting the cardiac information in the unmasked ECG signal and obtaining the ECG vector features;

[0053] The masked lead ECG mapping generation decoder 23 maps the obtained ECG vector features to the masked leads, reconstructs the masked ECG signal, and calculates the error between the reconstructed signal and the truly recorded ECG to optimize the model.

[0054] Training steps of the ECG adaptive reconstruction neural network:

[0055] During actual use, when some leads in the dynamic 12-lead ECG present low quality, the low-quality lead ECG is detected through ECG quality assessment, and the detected low-quality lead ECG is set to 0. Then, the EASNN model is used to adaptively compensate and generate the low-quality lead ECG.

[0056] Therefore, during the training process, the lead ECG is actively set to 0 through masking to simulate the situation of low-quality leads.

[0057] Step 1, random masking of the input data: Denote the input twelve conventional leads and six additional leads as where T represents the total time step Time, L represents the total number of leads Lead, represents the amplitude on the l-th lead at the t-th moment, and R represents the set of real numbers;

[0058] Step 2, randomly select L 1 leads from the twelve conventional leads and six additional leads respectively for masking, and the remaining L 2 leads are not masked, where L1 + L2 = 18;

[0059] 22 Extract the ECG vector features from the unmasked additional leads;

[0060] Step 3, the masking method of is as follows:

[0061]

[0062] Masked - ECG represents the masked ECG vector matrix, where the ECG signal on each j-th lead is 0, and the original ECG signal is retained on each i-th lead; represents the masked lead number, L 1 represents the number of masked leads, Indicates the unmasked lead number, L 2 Indicates the number of unmasked leads;

[0063] Step 4: Perform position encoding and data encoding: Since EASNN extracts cardiac information through the attention mechanism, before extracting the cardiac information in the Masked-ECG variable, time-lead position encoding must be performed. Therefore, we designed a time-lead encoding method to generate time-lead position variables and add these variables to the masked electrocardiogram to obtain the initial input of the model:

[0064] l is an even number

[0065] l is an odd number

[0066] where L is the number of leads, Indicates the absolute position of lead l at time t. The position encoding for the same time and the same lead in each segment remains fixed. The position information is added to the masked electrocardiogram signal Masked-ECG to obtain the initial input of the model

[0067]

[0068] Indicates adding the position information on the basis of to the variable. For the variable after position encoding The data encoding method is to use one-dimensional convolution to convert the input containing electrocardiogram information into a high-dimensional electrocardiogram vector:

[0069]

[0070] In the formula, the first dimension of the input X of Conv1D is time T, the second dimension is lead L, and the output dimension is T×D. When using one-dimensional convolution with a convolution kernel size of 18:

[0071] In the electrocardiogram reconstruction model training unit 2, EASNN uses D = 256 trainable convolution kernels;

[0072] In the electrocardiogram reconstruction model training unit 2, EASNN performs convolution once at each sampling point in the time dimension T, with a step size of 1;

[0073] In the electrocardiogram reconstruction model training unit 2, EASNN fuses the electrocardiogram information of all 18 leads through convolution at once in the lead dimension L.

[0074] For the input X D, the extraction of the unoccluded lead feature information is achieved through the heart information extraction encoder Encoder22 in EASNN, and then the reconstruction of the occluded lead ECG is realized through a layer of occluded lead ECG mapping generation decoder Decoder23:

[0075] ECG = Decoder(Encoder(X D ))

[0076] The dimension of the output ECG variable is T×L, where is the reconstructed ECG, and the EASNN model will calculate the error between the reconstructed and the recorded ECG signals, and train the parameters of the model according to the error.

[0077] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A system for improving ECG quality based on lightweight ECG quality assessment, comprising a wearable composite 18-lead ECG device (1), characterized in that: The wearable composite 18-lead ECG device (1) is electrically connected to the ECG reconstruction model training unit (2); the wearable composite 18-lead ECG device (1) is electrically connected to the ECG signal evaluation unit (3); the ECG signal evaluation unit (3) is communicatively connected to the ECG signal reconstruction unit (4); and the ECG reconstruction model training unit (2) is deployed and connected to the ECG signal reconstruction unit (4); The wearable composite 18-lead ECG device (1) is used for collecting lead ECG signals. The wearable composite 18-lead ECG device (1) sends the lead ECG signals to an ECG reconstruction model training unit (2) and an ECG signal evaluation unit (3); The ECG reconstruction model training unit (2) is used to train the ECG adaptive reconstruction neural network on the lead ECG signals; The ECG signal evaluation unit (3) is used for lightweight ECG quality evaluation and provides high and low quality labels for lead ECG signals. The ECG signal evaluation unit (3) sends the quality label of the composite 18-lead ECG and the ECG signal to the ECG signal reconstruction unit (4); The ECG signal reconstruction unit (4) reconstructs low-quality lead ECG signals using high-quality lead ECG signals based on the ECG adaptive reconstruction neural network trained by the ECG reconstruction model training unit (2) according to the quality label of the received composite 18-lead ECG.

2. The ECG quality improvement system based on lightweight ECG quality assessment according to claim 1, characterized in that: The wearable composite 18-lead ECG device (1) is provided with 18 leads, including 12 conventional leads and 6 additional leads; the wearable composite 18-lead ECG device (1) sends the 18-lead ECG signals to the ECG signal evaluation unit (3).

3. The ECG quality improvement system based on lightweight ECG quality assessment according to claim 2, characterized in that: The twelve conventional leads include Ⅰ, Ⅱ, Ⅲ, aVR, aVL, aVF, V1, V2, V3, V4, V5, and V6.

4. The ECG quality improvement system based on lightweight ECG quality assessment according to claim 3, characterized in that: The six additional leads include V3R, V4R, V5R, E, S, and I.

5. The ECG quality improvement system based on lightweight ECG quality assessment according to claim 4, characterized in that: ECG quality assessment based on permutation entropy, a signal quality threshold based on permutation entropy is established for each lead in the composite 18-lead ECG.

6. The ECG quality improvement system based on lightweight ECG quality assessment according to claim 5, characterized in that: The lightweight ECG quality assessment adopts permutation entropy (PE) assessment; The ECG signal evaluation unit (3) compares the PE value of each lead ECG signal with the PE threshold based on the permutation entropy, and finds out the lead ECG signals that are higher than the PE threshold and uses them as high-quality lead ECG signals, and finds out the lead ECG signals that are lower than the PE threshold and uses them as low-quality lead ECG signals.

7. The ECG quality improvement system based on lightweight ECG quality assessment according to claim 6, characterized in that: The electrocardiogram adaptive reconstruction neural network (EASNN) in the electrocardiogram reconstruction model training unit (2) comprises a lead cover time-lead encoding module (21), a cardiac information extraction encoder (22) and a cover lead electrocardiogram mapping generation decoder (23); The lead masking time-lead encoding module (21) is divided into two parts. The first part randomly masks the leads of the input ECG, and the masked lead ECG is a low-quality lead ECG, while the uncovered lead ECG is a high-quality lead ECG, so that the EASNN reconstructs the masked lead ECG through the uncovered lead ECG, which is used to train the ECG reconstruction performance of the EASNN. The second part is to perform time-lead position encoding and data encoding on the eighteen-lead ECG after the masking operation. The cardiac information extraction encoder (22) calculates the correlation relationship between ECG leads through an attention mechanism, thereby extracting cardiac information from the unmasked ECG signal and obtaining ECG vector features; The masked lead ECG mapping generation decoder (23) maps the acquired ECG vector features onto the masked lead, reconstructs the masked ECG signal, calculates the error between the masked ECG signal and the real recorded ECG, and optimizes the model.

8. The ECG quality improvement system based on lightweight ECG quality assessment according to any one of claims 6-7, characterized in that: The PE value is a measurement value of the signal-to-noise ratio or waveform integrity of the lead ECG signal.

Images