A distributed electrocardio monitoring and abnormality early warning method based on dynamic lead reconstruction

Abstract

The application discloses a kind of distributed electrocardiogram monitoring and abnormal early warning method based on dynamic lead reconstruction, it is related to wearable electrocardiogram technical field, including, establish the electrical connection of wearable host and disposable chest patch;Dynamic lead reconstruction is carried out to wearable host and disposable chest patch, and dynamic electrode switching timing is obtained;According to dynamic electrode switching timing, time division multiplexing switching is carried out, and the electrocardiogram signal segment of multiple leads is collected;Multiple lead electrocardiogram signal segments are reorganized into multi-lead electrocardiogram data according to timestamp, and multi-lead electrocardiogram data is preprocessed, and the output multi-lead heart beat segment;Multi-lead heart beat segment is input into deep learning model to predict and warn electrocardiogram anomaly of user.The application can obtain multi-lead electrocardiogram segment by contact impedance evaluation, user state classification and dynamic lead reconstruction, adaptively switch lead combination, and improve continuous acquisition stability.

Description

Technical Field

[0001] This invention relates to the field of wearable electrocardiogram (ECG) technology, and in particular to a distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction. Background Technology

[0002] With the development of wearable health monitoring technology, flexible electronics technology, low-power Bluetooth communication technology, and artificial intelligence medical analysis technology, electrocardiogram (ECG) monitoring devices are gradually evolving from large-scale ECG machines and Holter monitors used in hospitals to home-based, lightweight, and continuous monitoring. Traditional 12-lead ECGs provide relatively complete spatial information on cardiac electrical activity, suitable for clinical assessment of diseases such as arrhythmias, myocardial ischemia, and conduction abnormalities. Holter monitors, through longer continuous recording, improve the detection probability of occasional arrhythmias. In recent years, consumer-grade wearable devices such as smartwatches and smart rings, relying on PPG sensors, ECG analog front-ends, inertial sensors, and mobile terminal applications, have achieved convenient acquisition of heart rate, sleep, exercise, and short-term ECG signals. Meanwhile, disposable ECG patches, utilizing flexible substrates, gel electrodes, and miniaturized circuitry, can be attached to the chest for extended ECG recording. In addition, ECG abnormality identification methods based on machine learning and deep learning are gradually being applied to arrhythmia classification, ST segment abnormality analysis, and heart rate variability assessment, enabling ECG monitoring to evolve from simple data recording to automated assisted interpretation and risk alerts, providing a technological foundation for home-based chronic disease management, remote medical follow-up, and continuous screening of high-risk groups.

[0003] While the aforementioned wearable ECG monitoring technologies offer good portability and some data analysis capabilities, they still face challenges in addressing the difficulty of obtaining reliable dynamic multi-lead ECG data over long periods and providing real-time alerts for split-type wearable devices. Wearable main units, limited by size and electrode placement, typically only acquire single-lead or short-duration ECG signals, making it difficult to reflect the multi-directional spatial characteristics of the ECG on the chest. Disposable ECG patches, although closer to the heart, usually have lead configurations fixed after structural design, making it difficult to adaptively adjust based on changes in motion, electrode contact quality, and signal quality. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a distributed ECG monitoring and anomaly early warning method based on dynamic lead reconstruction to solve the problem that it is difficult for split-type wearable devices to obtain reliable dynamic multi-lead ECG data over long periods and provide real-time early warning.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a distributed electrocardiogram monitoring and abnormality early warning method based on dynamic lead reconstruction, comprising: A wearable main unit is provided, with at least two ECG contacts on the side that contacts the skin, and a magnetic attachment and at least four reusable contacts on the bottom; A disposable ECG chest patch is provided, which has a connecting structure and two electrically isolated conductive areas in the connecting structure; Establish an electrical connection between the wearable host and the disposable chest patch, perform identity verification, and switch the wearable host to ECG monitoring mode; In ECG monitoring mode, the user's physiological state characteristic parameters and contact impedance are acquired, the user's state is classified and the contact impedance is evaluated in real time, and the wearable host and disposable chest patch are dynamically reconstructed to obtain the dynamic electrode switching sequence. The physiological state characteristic parameters include motion acceleration and ECG signal segments. Based on the dynamic electrode switching sequence, time-division multiplexing switching is performed on different electrode pairs and differential channels between the wearable host and the disposable chest patch to collect ECG signal segments from multiple leads. The ECG signal segments from multiple leads are reassembled according to timestamps to form multi-lead ECG data, and the multi-lead ECG data is preprocessed to output multi-lead heartbeat segments. Multi-lead cardiac pulse segments are input into a deep learning model to predict and warn of ECG abnormalities in users.

[0007] As a preferred embodiment of the distributed ECG monitoring and abnormal early warning method based on dynamic lead reconstruction described in this invention, the wearable host integrates ECG monitoring function and is in daily mode when worn independently for daily health monitoring. The wearable host is detachably fixed to the disposable chest patch through a connection structure to establish an electrical connection. The wearable host switches to ECG monitoring mode to collect multi-lead ECG data from the user.

[0008] As a preferred embodiment of the distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction described in this invention, the specific steps for classifying user states are as follows: After removing the gravitational component from the motion acceleration, the root mean square value of the motion acceleration is calculated as an index of motion intensity. Based on the historical motion acceleration distribution collected by the user in daily mode, a first motion intensity threshold and a second motion intensity threshold are set. When the exercise intensity index is not greater than the first exercise intensity threshold, the user is in a resting state. When the exercise intensity index is greater than the first exercise intensity threshold but less than the second exercise intensity threshold, the user is in a daily activity state. When the exercise intensity index is not less than the second exercise intensity threshold, the user is in a vigorous exercise state.

[0009] As a preferred embodiment of the distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction described in this invention, the real-time contact impedance assessment includes the following specific steps. An impedance detection signal of known amplitude is injected between the disposable chest patch electrode and the ECG analog front end of the wearable host. The current electrode pair to be tested is selected by the multiplexer inside the disposable chest patch. The response voltage generated at both ends of the selected electrode pair under the action of the impedance detection signal is collected. The ratio of the response voltage to the current value of the impedance detection signal is calculated to obtain the real-time contact impedance. The impedance threshold is set based on the disposable chest patch electrode material, the area of ​​the conductive adhesive, and the input impedance of the ECG analog front end; When the real-time contact impedance is not greater than the impedance threshold, it indicates that the currently selected electrode pair is in a reliable contact state and is marked as a usable electrode pair. When the real-time contact impedance is greater than the impedance threshold, it indicates that the currently selected electrode pair is in a poor contact state and is marked as an unusable electrode pair. A spare electrode pair is then selected for replacement.

[0010] As a preferred embodiment of the distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction described in this invention, the specific steps for dynamically reconstructing the wearable host and the disposable chest patch are as follows: Multiple ECG electrodes on a disposable chest patch are paired up to form a candidate electrode pair set, and electrode pairs marked as unusable are removed from the candidate electrode pair set; Set the electrode polling strategy based on the current user status; For multiple electrode pairs in the current candidate electrode pair set, polling is performed according to the electrode polling strategy to obtain the dynamic electrode switching timing.

[0011] As a preferred embodiment of the distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction described in this invention, the specific steps for time-division multiplexing switching of different electrode pairs and differential channels between the wearable host and the disposable chest patch are as follows: The wearable host sends a dynamic electrode switching sequence to the multiplexer of the disposable chest patch, driving the disposable chest patch to perform electrode switching according to the dynamic electrode switching sequence; By reusing the same physical contact point, the same multiplexer channel, and the same ECG analog front end between the wearable host and the disposable chest patch within different time windows, the voltage difference between each electrode pair is collected, and multiple ECG signal segments of multiple leads are obtained in time-division multiplexing.

[0012] As a preferred embodiment of the distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction described in this invention, the preprocessing of multi-lead ECG data includes the following specific steps. Active motion artifact compensation is performed on multi-lead ECG data by utilizing motion acceleration with the gravitational component removed. Baseline drift correction, bandpass filtering, and heartbeat segmentation were performed on the actively compensated multi-lead ECG data to obtain multi-lead heartbeat segments.

[0013] As a preferred embodiment of the distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction described in this invention, the specific steps for predicting and warning of ECG abnormalities in users are as follows: From the user's historical multi-lead heartbeat segments, obtain multi-lead heartbeat segments with ECG abnormality annotations, combine multiple consecutive training multi-lead heartbeat segments into a training sample, and calculate the corresponding RR interval sequence; The feature extraction layer of the deep learning model extracts the morphological features and inter-lead spatial features of multi-lead cardiac segments. The anomaly detection layer concatenates the morphological features and lead spatial features, performs linear transformation and nonlinear activation, and outputs the predicted ECG abnormality type. The temporal prediction layer performs temporal modeling on the feature vectors corresponding to multiple consecutive multi-lead cardiac segments and the corresponding RR interval sequences, and outputs the risk score corresponding to different ECG abnormality types. The deep learning model is trained using training samples and the corresponding RR interval sequences; A first risk threshold and a second risk threshold are set based on the user's historical electrocardiogram signals and past history of arrhythmia. When the risk score for any type of ECG abnormality is greater than or equal to the first risk threshold and less than the second risk threshold, the user is determined to have an abnormal risk and a general risk warning is issued through the mobile terminal. When the risk score for any type of ECG abnormality is greater than or equal to the second risk threshold, the user is determined to have a high-risk abnormal event, and a high-risk warning notification is issued via the mobile terminal.

[0014] As a preferred embodiment of the distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction described in this invention, the specific steps for actively compensating for motion artifacts in the ECG data are as follows: Remove the gravitational component from the motion acceleration to obtain the dynamic acceleration component. Then, concatenate the dynamic acceleration components of the current sampling time and several previous historical sampling times to form a reference input vector. The reference input vector is input into the adaptive filter, which outputs the pseudo-components of the ECG signal and subtracts them from the ECG signal.

[0015] As a preferred embodiment of the distributed ECG monitoring and abnormal early warning method based on dynamic lead reconstruction described in this invention, the step of setting the electrode polling strategy according to different user states means polling all electrode pairs in the candidate electrode pair set according to different user states and different frequencies, based on real-time contact impedance from low to high.

[0016] The beneficial effects of this invention are as follows: By coordinating the wearable host and the disposable chest patch, the daily-wearable device can be transformed into a long-term ECG monitoring device when needed, taking into account both user stickiness and the quality of chest ECG acquisition; through contact impedance assessment, user status classification, and dynamic lead reconstruction, the single-channel hardware can acquire multi-lead ECG segments in a time-sharing manner, and adaptively switch lead combinations when there is poor electrode contact or increased motion, thereby improving the stability of continuous acquisition; through timestamp reconstruction, active compensation for motion artifacts, and multi-lead heartbeat segmentation, the fidelity of ECG waveforms is improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a distributed ECG monitoring and abnormal early warning method based on dynamic lead reconstruction.

[0019] Figure 2 This is a flowchart for preprocessing multi-lead electrocardiogram data.

[0020] Figure 3 A flowchart for user status classification and real-time contact impedance assessment.

[0021] Figure 4 This is a flowchart for dynamic lead reconstruction and time-division multiplexing multi-lead acquisition. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a distributed electrocardiogram monitoring and abnormality early warning method based on dynamic lead reconstruction, including the following steps: S1. Establish an electrical connection between the wearable host and the disposable chest patch, perform identity verification, and switch the wearable host to ECG monitoring mode.

[0026] The wearable main unit integrates ECG monitoring function and can be worn independently. It has a built-in ECG analog front end, PPG sensor, three-axis accelerometer, Bluetooth and rechargeable battery. The side that contacts the skin has at least two ECG contacts, and the bottom has a magnetic attachment and at least four multiplexed contacts to form a charging / control multiplexed contact group (used to connect to the charging dock to charge the wearable main unit when charging, and to transmit dynamic electrode switching timing through the bottom magnetic attachment of the disposable chest patch in ECG monitoring mode).

[0027] The disposable chest patch includes a flexible substrate, a bottom magnetic closure, at least four ECG electrodes, a multiplexer (analog switch), an ultra-thin disposable battery, and an identification resistor. The disposable chest patch has a connection structure fixed to the upper surface of the flexible substrate, and the waist of the card holder has two electrically isolated conductive areas that cover the entire circumference of the waist of the card holder, enabling a 360° electrical connection between the wearable host and the disposable chest patch without alignment.

[0028] Furthermore, a wearable host refers to a portable electronic device that can be worn on the human body and has electrocardiogram signal processing and wireless communication functions. Its form includes, but is not limited to, smart rings, smart bracelets, smartwatches, and necklace-style hosts. The connection structure refers to the structure that can detachably connect the wearable host to the disposable chest patch, including an I-shaped bracket, or it can adopt a buckle, magnetic or other detachable connection method, which can be changed according to the different structures of the wearable host to ensure that the wearable host can be fixed to the chest patch and that the contacts can reliably contact the conductive area.

[0029] When a user wears the wearable device alone, it is in daily mode, collecting data such as heart rate and motion acceleration through a PPG sensor and a three-axis accelerometer for the user's daily health monitoring.

[0030] When a user needs to perform long-term ECG monitoring, a disposable chest patch is attached to the user's chest. The wearable host is connected to the disposable chest patch through a connection structure. The two ECG contacts of the wearable host make physical contact with the two conductive areas of the disposable chest patch to establish an electrical connection, thus switching from the daily mode to the ECG monitoring mode.

[0031] The disposable chest patch has a pre-installed identification element (resistor, capacitor, or ID chip); when the wearable host detects the disposable chest patch being connected, it reads the feature value of the identification element; if the feature value matches the authorized value stored in the wearable host, ECG data collection is allowed; if they do not match, the host refuses to work and prompts the user to use the original consumable through the application, in order to prevent third-party compatible chest patches and protect the subscription business model.

[0032] S2. In ECG monitoring mode, acquire the user's physiological state characteristic parameters and contact impedance, classify the user's state and evaluate the contact impedance in real time, and dynamically reconstruct the lead between the wearable host and the disposable chest patch to obtain the dynamic electrode switching sequence.

[0033] In ECG monitoring mode, the wearable host collects the user's motion acceleration and ECG signals (ECG waveforms) in real time through a built-in triaxial accelerometer and ECG simulation front end, and outputs motion acceleration and ECG signal segments as physiological state characteristic parameters.

[0034] Based on motion acceleration, user states are categorized into resting state, daily activity state, and vigorous exercise state. Specifically, after removing the gravitational component from motion acceleration, the root mean square value of acceleration is calculated as an indicator of exercise intensity. Based on the historical motion acceleration distribution collected from the user in daily mode, a first and second motion intensity threshold are set. For example, with a calculation window of 5 seconds, the first motion intensity threshold can be set to 0.03g, and the second motion intensity threshold can be set to 0.15g, where g represents gravitational acceleration. When the wearable device is stationary or the user is wearing it quietly, motion acceleration mainly comes from sensor noise and slight body movements caused by breathing, resulting in a typically low exercise intensity. Setting the first motion intensity threshold to 0.05g ensures that static noise and slight body movements are not misjudged as daily activities. However, when the user performs daily activities such as walking, raising their arms, or turning over, the root mean square value of dynamic acceleration after removing gravity is usually higher than 0.05g, but lower than that of continuous running, rapid arm swinging, or vigorous exercise scenarios. Therefore, setting the second motion intensity threshold to 0.30g distinguishes between daily activities and vigorous exercise.

[0035] When the exercise intensity index is not greater than the first exercise intensity threshold, the user is in a resting state. When the exercise intensity index is greater than the first exercise intensity threshold but less than the second exercise intensity threshold, the user is in a daily activity state. When the exercise intensity index is not less than the second exercise intensity threshold, the user is in a vigorous exercise state.

[0036] An impedance detection signal of known amplitude (an AC constant current signal whose frequency avoids the effective frequency band of the ECG signal) is injected between the disposable chest patch electrode and the ECG simulation front end of the wearable host. The real-time contact impedance of the currently selected electrode pair is evaluated in real time. Specifically, the electrode pair to be tested is selected by a multiplexer in the disposable chest patch, and the response voltage generated at both ends of the currently selected electrode pair under the action of the impedance detection signal is collected. The ratio of the response voltage to the current value of the impedance detection signal is calculated to obtain the real-time contact impedance. The impedance threshold is set according to the material of the disposable chest patch electrode, the area of ​​the conductive adhesive, and the input impedance of the ECG simulation front end. For example, when the disposable chest patch uses Ag / AgCl printed electrodes and medical conductive adhesive, the impedance threshold of a single electrode is set to 10kΩ. Ag / AgCl electrodes with medical conductive adhesive are wet or semi-wet bioelectric electrodes. After using gel, the skin-electrode impedance can be reduced to the order of 5kΩ to 10kΩ. The contact impedance of the electrode pair approximately includes the series contribution of the impedance of the two electrode-skin interfaces. Therefore, the impedance threshold of the electrode pair can be set to 20kΩ.

[0037] When the real-time contact impedance is not greater than the impedance threshold, it indicates that the currently selected electrode pair is in a reliable contact state and is marked as a usable electrode pair. When the real-time contact impedance is greater than the impedance threshold, it indicates that the currently selected electrode pair is in a poor contact state and is marked as an unusable electrode pair. A spare electrode pair is then selected for replacement.

[0038] Based on user status classification results and real-time contact impedance assessment, the wearable host performs dynamic lead reconstruction by generating a dynamic electrode switching timing sequence. The specific steps are as follows. Multiple ECG electrodes on a disposable chest patch are paired to form a candidate electrode pair set. Each candidate electrode pair corresponds to an equivalent lead or lead variant. Based on the evaluation results of real-time contact impedance (available / unavailable electrode pairs), electrode pairs marked as unavailable are removed from the candidate electrode pair set.

[0039] Furthermore, the electrode polling strategy is set according to the user's current ECG acquisition environment. Specifically, when the user is in a resting state, all electrode pairs in the candidate electrode pair set are polled at a low frequency (e.g., 1Hz) to obtain complete multi-lead information (e.g., leads I, II, and III, and variations in chest leads), focusing on diagnostic accuracy and power consumption optimization. When the user is in a daily activity state, the candidate electrode pair set is polled at a medium frequency (e.g., 4Hz) to balance signal quality and real-time performance. When the user is in a strenuous exercise state, the candidate electrode pair set is polled rapidly at a high frequency (e.g., 16Hz) to shorten the single conduction duration to suppress motion artifacts. At the same time, the bandwidth and gain of the ECG analog front-end are simultaneously increased on the wearable host.

[0040] For multiple electrode pairs in the current candidate electrode pair set, polling is performed according to the real-time contact impedance from low to high and the electrode polling strategy to obtain the dynamic electrode switching sequence (including the set of electrode pairs participating in the polling, the polling order of the electrode pairs, and the polling frequency). During the electrode pair polling process, when the contact impedance of a certain electrode pair is detected to exceed the impedance threshold, the corresponding electrode pair is dynamically skipped and switched to the backup electrode pair (electrode pairs with the same or similar lead directions) to ensure the continuity of acquisition.

[0041] S3. Based on the dynamic electrode switching sequence, time-division multiplexing is performed on different electrode pairs and differential channels between the wearable host and the disposable chest patch to collect ECG signal segments from multiple leads.

[0042] Time-division multiplexing switching refers to multiplexing the same ECG contact, the same differential output channel of the multiplexer, and the same ECG analog front end between the wearable host and the disposable chest patch within different time windows, and sequentially performing electrode pair gating, real-time contact impedance assessment, and ECG signal acquisition. It should be noted that what is multiplexed is the differential output channel, what is switched is the input electrode selection relationship, and the input side of the multiplexer changes the selected chest patch electrode pair according to the dynamic electrode switching sequence.

[0043] The wearable host sends the dynamic electrode switching sequence to the multiplexer of the disposable chest patch through the multiplexed contact, driving the disposable chest patch to perform electrode switching according to the dynamic electrode switching sequence, realizing online real-time reconstruction of lead configuration (the same physical differential channel corresponds to different electrode pair combinations at different times), thereby acquiring multi-dimensional ECG signals with single-channel hardware.

[0044] The dynamic electrode switching timing is transmitted in the form of serial control signals, clock / data signals or electrode address control signals, and is used to drive the disposable chest patch to perform electrode switching according to the dynamic electrode switching timing.

[0045] The selected electrode pair is connected to two conductive areas via the differential output channel of the multiplexer. The wearable host then collects the voltage difference between the two conductive areas through ECG contacts to obtain the ECG signal.

[0046] The ECG signals between each electrode pair are collected sequentially, and ECG signal segments from multiple leads are obtained in a time-division manner (including at least leads I, II, and III, depending on the dynamic configuration). The wearable host then sends the collected multi-lead ECG signal segments to the mobile terminal in real time via Bluetooth.

[0047] S4. Reassemble the ECG signal segments from multiple leads according to the timestamps to form multi-lead ECG data, preprocess the multi-lead ECG data, and output multi-lead heartbeat segments.

[0048] The mobile terminal reassembles the received multi-lead signal segments according to timestamps to form continuous multi-lead electrocardiogram data.

[0049] By utilizing motion acceleration with the gravity component removed, motion artifacts in multi-lead ECG data are actively compensated. Specifically, motion acceleration is linearly interpolated based on the acquisition timestamp, and the motion acceleration synchronously acquired by the wearable host is aligned with the ECG signal in the time domain. The motion acceleration is used as a reference reference for constructing an adaptive filter, and the reference input vector is input into the adaptive filter. During the ECG signal acquisition stage, the artifact components generated by the user's motion are actively predicted and canceled.

[0050] In this process, the reference input vector for constructing the adaptive filter using motion acceleration as the feedforward noise reference benchmark refers to removing the gravitational component from the motion acceleration to obtain the dynamic acceleration component. The dynamic acceleration components at the current sampling time and several previous historical sampling times are concatenated to form the reference input vector. The reference input vector is then input into the adaptive filter, which outputs the pseudo-components of the ECG signal. For example, an adaptive transverse filter with a finite impulse response (FIR) structure is used, with an FIR order of 32 and an initial weight vector value of 0. The weight vector is updated online using the normalized minimum mean method. The pseudo-components are obtained by weighted summation of the components in the reference input vector. By subtracting the pseudo-components from the ECG signal, the pseudo-components generated by the user's motion are actively predicted and canceled during the ECG signal acquisition phase.

[0051] Furthermore, For actively compensated multi-lead ECG data, baseline drift correction, bandpass filtering, and heartbeat segmentation are performed. Specifically, Baseline drift correction refers to estimating the baseline drift component in the ECG signal using methods such as moving average filtering, median filtering, morphological filtering, or low-order trend fitting, and then subtracting the baseline drift component from the ECG signal in the corresponding lead. Bandpass filtering refers to selecting bandpass filter parameters according to the subsequent analysis task, preserving the ECG components within the effective frequency band of ECG, and suppressing residual low-frequency noise, high-frequency electromyography interference, and sampling noise. When used for arrhythmia detection and QRS complex identification, a bandpass filter range suitable for highlighting the QRS complex (5–40 Hz) can be used. The main energy of the QRS complex is concentrated in the mid-frequency band, and a 5 Hz high-pass filter can suppress the baseline. Drift (a 40Hz low-pass filter can suppress electromyography and high-frequency sampling noise, suitable for highlighting QRS complexes). When used for ST segment or myocardial ischemia risk assessment, filtering parameters that can preserve low-frequency morphological information of the ST segment are used to avoid excessive attenuation of effective ECG morphology. Beat segmentation refers to QRS complex detection and R-wave peak localization of the bandpass-filtered ECG signal. With each R-wave peak time point as the center, ECG data windows are extracted before and after it to form a single beat segment. For multi-lead ECG data, the system synchronously extracts the corresponding ECG data windows of each lead at the same R-wave peak time point to form a multi-lead beat segment.

[0052] S5. Input multi-lead cardiac pulse segments into a deep learning model to predict and warn of ECG abnormalities in users.

[0053] From the user's historical multi-lead heartbeat segments, multi-lead heartbeat segments marked with ECG abnormalities (ECG abnormalities include arrhythmia, ST segment abnormalities, and HRV abnormalities) are obtained as training multi-lead heartbeat segments, and the corresponding RR interval sequence is calculated based on the R wave peak time points of consecutive training multi-lead heartbeat segments.

[0054] A training sample is formed by combining N consecutive multi-lead cardiac beat segments. Different samples correspond to different types of ECG abnormalities. The training sample and the corresponding RR interval sequence are used as input to the deep learning model. The corresponding ECG abnormality labels are used as output labels of the deep learning model to supervise the training of the deep learning model.

[0055] Furthermore, deep learning models include feature extraction layers, anomaly detection layers, and temporal prediction layers.

[0056] The feature extraction layer includes two levels of one-dimensional convolution processing and one level of pooling processing. The first convolution processing has a kernel length of 7 and 32 channels, while the second convolution processing has a kernel length of 5 and 64 channels. The pooling processing is used to obtain morphological features and inter-lead spatial features. The anomaly detection layer concatenates the morphological features and inter-lead spatial features into an anomaly detection feature vector, and performs linear transformation and nonlinear activation on the anomaly detection feature vector to output the predicted ECG abnormality type. The temporal prediction layer uses LSTM to perform temporal modeling on the feature vectors corresponding to N consecutive multi-lead cardiac beat segments and the corresponding RR interval sequences, and normalizes them to 0 to 1 using Sigmoid to output the risk score corresponding to each type of ECG abnormality (representing the probability that the current multi-lead cardiac beat segment has a corresponding ECG abnormality).

[0057] A training loss function is constructed based on the difference between the output of the deep learning model and the corresponding ECG abnormality label. Since the deep learning model is used to perform multiple abnormality detection tasks, the training loss function includes at least one of arrhythmia classification loss, ST segment abnormality detection loss, HRV abnormality detection loss, and time series risk prediction loss. That is, the training loss function of the deep learning model is a multi-task weighted loss function. During the training process, the model parameters of the feature extraction layer, abnormality detection layer, and time series prediction layer are updated through backpropagation until the maximum number of training rounds is reached.

[0058] Among them, the arrhythmia classification loss adopts cross-entropy loss, the ST segment abnormality detection loss and HRV abnormality detection loss adopt binary cross-entropy loss, and the time series risk prediction loss adopts cross-entropy loss.

[0059] Input the current user's multi-lead heartbeat segments and corresponding RR interval sequences into the deep learning model, and output the arrhythmia classification results, ST segment abnormality status, HRV abnormality status, and risk scores corresponding to various ECG abnormalities.

[0060] Based on the user's historical ECG signals and past history of arrhythmia, a first risk threshold and a second risk threshold are set. For example, the first risk threshold is 0.60 and the second risk threshold is 0.85. The risk score is the probability corresponding to each type of ECG abnormality, i.e., the range is 0~1. 0.5 can be used as the theoretical dividing line between abnormality and normality. The first risk threshold is 0.60, which is an uncertain dividing line above 0.50. This ensures that the model output only triggers ordinary risk warnings when there is a certain abnormal tendency, thereby reducing false alarms caused by short-term noise, abnormal single cardiac segment, or low-confidence output of the model. The second risk threshold is 0.85, which indicates that the model has a high confidence in abnormal events and is suitable for triggering high-risk warnings to reduce the false alarm rate of high-risk warnings.

[0061] Specifically, the second risk threshold is greater than the first risk threshold; when the risk score for any ECG abnormality type is greater than or equal to the first risk threshold and less than the second risk threshold, the user is determined to have an abnormal risk, and a general risk warning is issued through the mobile terminal; when the risk score for any ECG abnormality type is greater than or equal to the second risk threshold, the user is determined to have a high-risk abnormal event, and a high-risk warning notification is issued through the mobile terminal.

[0062] In summary, this invention achieves the following: First, the wearable host and disposable chest patch work together to transform a daily-wearable device into a long-term ECG monitoring device when needed, balancing user stickiness and the quality of chest ECG acquisition. Second, through contact impedance assessment, user status classification, and dynamic lead reconstruction, single-channel hardware can acquire multi-lead ECG segments in a time-sharing manner, and adaptively switch lead combinations when electrode contact is poor or motion is enhanced, improving continuous acquisition stability. Third, through timestamp reconstruction, active compensation for motion artifacts, and multi-lead heartbeat segmentation, the fidelity of ECG waveforms is improved.

[0063] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction, characterized in that: include, A wearable main unit is provided, with at least two ECG contacts on the side that contacts the skin, and a magnetic attachment and at least four reusable contacts on the bottom; A disposable ECG chest patch is provided, which has a connecting structure and two electrically isolated conductive areas in the connecting structure; Establish an electrical connection between the wearable host and the disposable chest patch, perform identity verification, and switch the wearable host to ECG monitoring mode; In ECG monitoring mode, the user's physiological state characteristic parameters and contact impedance are acquired, the user's state is classified and the contact impedance is evaluated in real time, and the wearable host and disposable chest patch are dynamically reconstructed to obtain the dynamic electrode switching sequence. The physiological state characteristic parameters include motion acceleration and ECG signal segments. Based on the dynamic electrode switching sequence, time-division multiplexing switching is performed on different electrode pairs and differential channels between the wearable host and the disposable chest patch to collect ECG signal segments from multiple leads. The ECG signal segments from multiple leads are reassembled according to timestamps to form multi-lead ECG data, and the multi-lead ECG data is preprocessed to output multi-lead heartbeat segments. Multi-lead cardiac pulse segments are input into a deep learning model to predict and warn of ECG abnormalities in users.

2. The distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction as described in claim 1, characterized in that: The wearable device integrates ECG monitoring functionality and operates in daily mode when worn independently for routine health monitoring. The wearable host is detachably fixed to the disposable chest patch through a connection structure to establish an electrical connection. The wearable host switches to ECG monitoring mode to collect multi-lead ECG data from the user.

3. The distributed ECG monitoring and abnormal early warning method based on dynamic lead reconstruction as described in claim 2, characterized in that: The specific steps for classifying user states are as follows: After removing the gravitational component from the motion acceleration, the root mean square value of the motion acceleration is calculated as an index of motion intensity. Based on the historical motion acceleration distribution collected by the user in daily mode, a first motion intensity threshold and a second motion intensity threshold are set. When the exercise intensity index is not greater than the first exercise intensity threshold, the user is in a resting state. When the exercise intensity index is greater than the first exercise intensity threshold but less than the second exercise intensity threshold, the user is in a daily activity state. When the exercise intensity index is not less than the second exercise intensity threshold, the user is in a vigorous exercise state.

4. The distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction as described in claim 3, characterized in that: The specific steps for the real-time contact impedance assessment are as follows. An impedance detection signal of known amplitude is injected between the disposable chest patch electrode and the ECG analog front end of the wearable host. The current electrode pair to be tested is selected by the multiplexer inside the disposable chest patch. The response voltage generated at both ends of the selected electrode pair under the action of the impedance detection signal is collected. The ratio of the response voltage to the current value of the impedance detection signal is calculated to obtain the real-time contact impedance. The impedance threshold is set based on the disposable chest patch electrode material, the area of ​​the conductive adhesive, and the input impedance of the ECG analog front end; When the real-time contact impedance is not greater than the impedance threshold, it indicates that the currently selected electrode pair is in a reliable contact state and is marked as a usable electrode pair. When the real-time contact impedance is greater than the impedance threshold, it indicates that the currently selected electrode pair is in a poor contact state and is marked as an unusable electrode pair. A spare electrode pair is then selected for replacement.

5. The distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction as described in claim 4, characterized in that: The specific steps for dynamically reconstructing the connection between the wearable host and the disposable chest patch are as follows. Multiple ECG electrodes on a disposable chest patch are paired up to form a candidate electrode pair set, and electrode pairs marked as unusable are removed from the candidate electrode pair set; Set the electrode polling strategy according to the current user status; For multiple electrode pairs in the current candidate electrode pair set, polling is performed according to the electrode polling strategy to obtain the dynamic electrode switching timing.

6. The distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction as described in claim 5, characterized in that: The specific steps for time-division multiplexing and switching of different electrode pairs and differential channels between the wearable host and the disposable chest patch are as follows: The wearable host sends a dynamic electrode switching sequence to the multiplexer of the disposable chest patch, driving the disposable chest patch to perform electrode switching according to the dynamic electrode switching sequence; By reusing the same physical contact point, the same multiplexer channel, and the same ECG analog front end between the wearable host and the disposable chest patch within different time windows, the voltage difference between each electrode pair is collected, and multiple ECG signal segments of multiple leads are obtained in time-division multiplexing.

7. The distributed ECG monitoring and abnormality early warning method based on dynamic lead reconstruction as described in claim 6, characterized in that: The specific steps for preprocessing multi-lead ECG data are as follows. Active motion artifact compensation is performed on multi-lead ECG data by utilizing motion acceleration after removing the gravitational component; Baseline drift correction, bandpass filtering, and heartbeat segmentation were performed on the actively compensated multi-lead ECG data to obtain multi-lead heartbeat segments.

8. The distributed ECG monitoring and abnormal early warning method based on dynamic lead reconstruction as described in claim 7, characterized in that: The specific steps for predicting and issuing early warnings of ECG abnormalities in users are as follows. From the user's historical multi-lead heartbeat segments, obtain multi-lead heartbeat segments with ECG abnormality annotations, combine multiple consecutive training multi-lead heartbeat segments into a training sample, and calculate the corresponding RR interval sequence; The feature extraction layer of the deep learning model extracts the morphological features and inter-lead spatial features of multi-lead cardiac segments. The anomaly detection layer concatenates the morphological features and lead spatial features, performs linear transformation and nonlinear activation, and outputs the predicted ECG abnormality type. The temporal prediction layer performs temporal modeling on the feature vectors corresponding to multiple consecutive multi-lead cardiac segments and the corresponding RR interval sequences, and outputs the risk score corresponding to different ECG abnormality types. The deep learning model is trained using training samples and the corresponding RR interval sequences; A first risk threshold and a second risk threshold are set based on the user's historical electrocardiogram signals and past history of arrhythmia. When the risk score for any type of ECG abnormality is greater than or equal to the first risk threshold and less than the second risk threshold, the user is determined to have an abnormal risk and a general risk warning is issued through the mobile terminal. When the risk score for any type of ECG abnormality is greater than or equal to the second risk threshold, the user is determined to have a high-risk abnormal event, and a high-risk warning notification is issued via the mobile terminal.

9. The distributed ECG monitoring and abnormal early warning method based on dynamic lead reconstruction as described in claim 7, characterized in that: The specific steps for actively compensating for motion artifacts in electrocardiogram data are as follows. Remove the gravitational component from the motion acceleration to obtain the dynamic acceleration component. Then, concatenate the dynamic acceleration components of the current sampling time and several previous historical sampling times to form a reference input vector. The reference input vector is input into the adaptive filter, the components of the reference input vector are weighted and summed, the pseudo-components of the ECG signal are output, and the pseudo-components are subtracted from the ECG signal.

10. The distributed ECG monitoring and abnormal early warning method based on dynamic lead reconstruction as described in claim 5, characterized in that: The electrode polling strategy based on different current user states refers to polling all electrode pairs in the candidate electrode pair set according to different user states and different frequencies, based on real-time contact impedance from low to high.

Images