Real-time warning system for malignant arrhythmia based on edge computing

Abstract

The present application relates to the technical field of electrocardio monitoring, and discloses a real-time warning system for malignant arrhythmia based on edge computing, which comprises an electrocardio collection channel, an impedance detection channel, an acceleration collection channel, a heart rhythm recognition module, an edge processing unit and an alarm unit. The edge processing unit performs time alignment and scale unification on the electrode-skin contact impedance, electrocardio voltage and patch normal acceleration in the same analysis window, performs time delay coupling analysis within a preset allowable time delay range, obtains the maximum coupling strength, optimal time delay, time delay triangle closure error and three-edge coupling strength, and forms an alarm decision accordingly. The system immediately alarms when the confidence of malignant arrhythmia reaches an emergency protection threshold, and rechecks or prompts in combination with the contact interface coupling risk under ordinary alarm conditions, and is mainly used for real-time monitoring of wearable or bedside electrocardio.

Description

Technical Field

[0001] This invention belongs to the field of electrocardiogram monitoring technology, specifically relating to a real-time early warning system for malignant arrhythmias on electrocardiograms based on edge computing. Background Technology

[0002] In clinical and outpatient management, continuous electrocardiogram (ECG) monitoring devices typically use surface electrodes or ECG patches to acquire ECG voltage, and perform filtering, QRS detection, rhythm recognition, alarm output, and event logging on a local processor. To ensure alarm reliability, the device also collects or generates quality information such as lead dislodgement, contact impedance, acceleration, saturation, noise, timestamps, and sampling sequence numbers. During deployment, patient turning, walking, coughing, patch loosening, or lead wire traction can simultaneously affect the contact interface, ECG channel, and patch movement, potentially resulting in ECG waveforms resembling wide, abnormal QRS complexes, rapid rhythm fragments, or short-duration baseline drift.

[0003] Existing edge-of-care ECG devices mostly use shunting quality thresholds for discrimination. For example, impedance exceeding the limit indicates contact abnormality, acceleration exceeding the limit indicates body motion, and ECG voltage saturation or noise exceeding the limit indicates waveform unusable. This type of shunting is easy to implement and suitable for running on low-power processors, but there is a lack of unified timing correlation between the judgments of each channel, making it difficult to distinguish between multiple synchronous abnormalities caused by the same contact interface disturbance and ECG voltage changes caused by real cardiac rhythm events.

[0004] If the system directly alarms when the heart rhythm recognition module gives a result with moderate confidence, it is easy to increase false alarms and the burden of manual review; if it simply relies on the quality threshold to delay the alarm, it may affect the handling of high-risk events. Summary of the Invention

[0005] The present invention aims to provide a real-time early warning system for malignant arrhythmias on electrocardiogram based on edge computing, which solves the technical problems mentioned in the background art.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution.

[0007] A real-time early warning system for malignant arrhythmias based on edge computing for electrocardiograms includes an electrocardiogram acquisition channel for acquiring electrocardiogram voltage, an impedance detection channel for acquiring electrode-skin contact impedance, an acceleration acquisition channel for acquiring patch normal acceleration, a heart rhythm recognition module, an edge processing unit, and an alarm unit. The edge processing unit is configured as follows: The electrode-skin contact impedance, the electrocardiogram voltage, and the patch normal acceleration are time-aligned and scaled within the same analysis window; Within a preset allowable time delay range, time delay coupling analysis is performed on the electrode-skin contact impedance, the electrocardiogram voltage, and the patch normal acceleration to determine the maximum coupling strength and the optimal time delay; The time delay triangle closure error is determined based on the optimal time delay, the three-sided coupling strength is determined based on the maximum coupling strength, and the electro-mechanical-impedance time delay triangle closure feature is generated from the time delay triangle closure error and the three-sided coupling strength. The heart rhythm recognition module is configured to output a malignant arrhythmia confidence level based on the ECG voltage; the edge processing unit generates an alarm decision based on the malignant arrhythmia confidence level and the electromechanical impedance time delay triangle closure feature, and controls the alarm unit to output a malignant arrhythmia alarm based on the alarm decision; Specifically, when the confidence level of the malignant arrhythmia reaches the emergency protection threshold, the alarm decision is to immediately output the malignant arrhythmia alarm, and the electromechanical impedance time delay triangle closure characteristic is not used as a condition for canceling the alarm.

[0008] Compared with the prior art, the present invention has the following substantial features and significant progress: This invention incorporates electrode-skin contact impedance, ECG voltage, and patch normal acceleration into the same analysis window, enabling the edge processing unit to identify temporal coupling relationships caused by disturbances at the same contact interface. Compared to relying solely on impedance thresholds, acceleration thresholds, or ECG quality thresholds, it reduces the impact of occasional fluctuations in a single auxiliary channel on alarm decisions, reduces false alarms for malignant arrhythmias caused by motion artifacts, and retains immediate alarms for high-confidence events through an emergency protection threshold. Furthermore, the event logging module saves ECG voltage, electrode-skin contact impedance, patch normal acceleration, optimal delay, maximum coupling strength, and alarm decisions, facilitating medical staff review and equipment performance verification. Attached Figure Description

[0009] Figure 1 This is a diagram of the system architecture and alarm decision mechanism of the present invention. Detailed Implementation

[0010] The following description is provided in conjunction with the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the invention; those skilled in the art can make equivalent substitutions or combinations for the specific implementations.

[0011] This implementation method is applicable to wearable ECG patch continuous monitoring scenarios and bedside ECG monitoring continuous monitoring scenarios.

[0012] like Figure 1As shown, the electrode-skin contact interface 100 is the common contact area formed between the ECG patch or ECG electrode and human skin. The ECG acquisition channel 110, impedance detection channel 120, and acceleration acquisition channel 130 all acquire monitoring data around this electrode-skin contact interface 100. The ECG acquisition channel 110 acquires the ECG voltage, the impedance detection channel 120 acquires the electrode-skin contact impedance, and the acceleration acquisition channel 130 acquires the patch normal acceleration. The three data streams are sent to the edge processing unit 140 respectively. The ECG acquisition channel 110 also provides the ECG voltage to the heart rhythm recognition module 150. The heart rhythm recognition module 150 outputs a malignant arrhythmia confidence level based on the ECG voltage and sends the malignant arrhythmia confidence level to the edge processing unit 140.

[0013] Edge processing unit 140 in Figure 1 Located between the three-channel data acquisition and alarm output, the system internally performs time alignment and scaling, time delay coupling analysis, generation of electromechanical-impedance time delay triangle closure characteristics, and alarm decision generation in sequence. The alarm unit 160 outputs a malignant arrhythmia alarm or lead contact abnormality alert based on the alarm decision output by the edge processing unit 140. The event recording module 170 associates and stores ECG voltage, electrode-skin contact impedance, patch normal acceleration, electromechanical-impedance time delay triangle closure characteristics, malignant arrhythmia confidence level, and alarm decision, enabling verification of the alarm cause.

[0014] The core processing object of this embodiment is not a single ECG voltage, nor an independent electrode-skin contact impedance overrun or patch normal acceleration overrun, but rather the temporal coupling relationship between electrode-skin contact impedance, ECG voltage, and patch normal acceleration within the same analysis window. When patient movement, electrode microslippage, patch edge lifting, or lead traction alters the contact area, local pressure, and interface charge state of the electrode-skin contact interface 100, this change may exhibit a common temporal structure in the electrode-skin contact impedance, ECG voltage, and patch normal acceleration. The edge processing unit 140 identifies the electrode-skin contact interface coupling risk through this temporal structure, thereby reducing false alarms for malignant arrhythmias caused by motion artifacts.

[0015] This implementation also preserves the safety boundaries of the malignant arrhythmia identification process. After the arrhythmia identification module 150 outputs the confidence level of the malignant arrhythmia, the edge processing unit 140 immediately generates an alarm decision to output a malignant arrhythmia alarm when the confidence level of the malignant arrhythmia reaches the emergency protection threshold, and does not use the electromechanical-impedance time delay triangle closure characteristic as a condition for canceling the alarm. This processing ensures that the immediate alarm of high-confidence malignant arrhythmia events is not unreasonably delayed by the alarm confirmation process.

[0016] Example 1: In a wearable ECG patch implementation, the ECG patch includes electrodes, an ECG analog front-end, an impedance detection circuit, an accelerometer, an edge processor, a memory, a communication unit, and a power supply. The ECG acquisition channel 110 can consist of electrodes, an ECG analog front-end, and an analog-to-digital conversion unit, used to acquire ECG voltage; the impedance detection channel 120 can consist of a conventional lead detachment detection circuit, a dedicated impedance measurement circuit, or an analog front-end integrating impedance detection functionality, used to acquire electrode-skin contact impedance; the acceleration acquisition channel 130 can consist of an axial output from an existing accelerometer aligned with the patch's normal direction, used to acquire the patch's normal acceleration.

[0017] In the bedside ECG monitoring implementation, electrodes are connected to the bedside ECG monitoring device via lead wires. ECG acquisition channel 110 obtains the ECG voltage, impedance detection channel 120 obtains the electrode-skin contact impedance, and acceleration acquisition channel 130 obtains the patch normal acceleration. If the accelerometer of the bedside ECG monitoring device is not directly installed at the patch, the patch normal acceleration is provided by an existing motion sensor in the attachment area; if no sensor exists to provide patch normal acceleration, the current analysis window does not calculate the electromechanical-impedance time delay triangle closure characteristic and performs a confirmation check or lead contact abnormality alert according to the baseline quality status.

[0018] ECG voltage, electrode-skin contact impedance, and patch normal acceleration are all timestamped with a unified device clock and written to a circular buffer accessible by the edge processing unit 140. When the edge processing unit 140 reads three data streams within the same analysis window, it simultaneously reads the corresponding timestamp, sampling sequence number, data saturation flag, and missing data flag. This approach allows the edge processing unit 140 to determine whether the data meets the processing requirements of the same analysis window and to stop calculating the electromechanical-impedance delay triangle closure characteristic of the current analysis window when there are abnormal timestamps, abnormal sampling sequence numbers, or the presence of data saturation or missing data flags.

[0019] The same analysis window is a time segment used by the edge processing unit 140 to process electrode-skin contact impedance, ECG voltage, and patch normal acceleration in parallel. The length of the analysis window can be set to a short time window that can cover several cardiac beat segments, based on the sampling rate of the ECG acquisition channel 110, the computing resources of the edge processing unit 140, and the delay requirements for real-time early warning of malignant arrhythmias. Analysis windows can slide continuously or in fixed steps, as long as the ECG voltage, electrode-skin contact impedance, and patch normal acceleration have comparable time indices within the same analysis window.

[0020] The length of the same analysis window is predetermined by a labeled sample set. The labeled sample set includes segments of malignant arrhythmias, normal stable segments, contact artifact segments, and body motion interference segments, with each type of segment bearing start and end times and event labels. The configuration tool validates the same analysis window length within a range of 6 to 12 seconds, with 8 seconds being the preferred length. The sliding step size for the same analysis window can be 1 to 2 seconds, preferably 2 seconds. If the sampling rate of the ECG acquisition channel 110 is low or the computational resources of the edge processing unit 140 are limited, a longer same analysis window length can be selected within the above candidate range; if a high real-time alarm delay is required, a shorter same analysis window length can be selected while ensuring the number of effective heartbeats and effective overlapping sample points.

[0021] The calibration conditions for the same analysis window length are as follows: For segments with labeled malignant arrhythmias, the detection sensitivity of the rhythm recognition module 150 in outputting the confidence level of the malignant arrhythmia and entering the alarm decision is not less than 0.95; for segments with contact artifacts, the edge processing unit 140 can form a stable electromechanical-impedance delay triangle closure feature; for normal stable segments, the false alarm rate caused by the same analysis window length is not higher than 0.05. If different candidate analysis window lengths all meet the above conditions, the candidate value with the shorter alarm response time and lower computational cost is selected. If the device model, the sampling rate of the ECG acquisition channel 110, the sampling rate of the impedance detection channel 120, the sampling rate of the acceleration acquisition channel 130, or the version of the rhythm recognition module 150 changes, the calibration for the same analysis window length must be performed again.

[0022] When the edge processing unit 140 processes the three data streams, it does not change the basic recognition algorithm of the heart rhythm recognition module 150. The heart rhythm recognition module 150 still performs QRS complex detection, routine ECG feature extraction, and malignant arrhythmia recognition based on ECG voltage, and outputs the confidence level of the malignant arrhythmia.

[0023] Example 2: Edge processing unit 140 performs time alignment and scale uniformity on electrode-skin contact impedance, ECG voltage, and patch normal acceleration within the same analysis window. The input for time alignment is the electrode-skin contact impedance sequence, ECG voltage sequence, and patch normal acceleration sequence acquired within the same analysis window; the output is the electrode-skin contact impedance, ECG voltage, and patch normal acceleration on a unified time grid. If the sampling rates of the three data streams are the same and the timestamps are consistent, edge processing unit 140 can directly align them according to the sampling sequence number; if the sampling rates of the three data streams are different, edge processing unit 140 resamples based on the timestamp of a unified device clock.

[0024] The unified time grid can be based on the sampling time point of the ECG voltage or the analysis time grid preset by the edge processing unit 140. When resampling the electrode-skin contact impedance and patch normal acceleration, linear interpolation, zero-order hold, or other conventional resampling methods can be used. When selecting a specific resampling method, the constraint should be to avoid changing the relative temporal relationship of the three data streams within the same analysis window, and to avoid misinterpreting interpolation fluctuations caused by resampling as a risk of electrode-skin contact interface coupling.

[0025] Scale uniformity is used to eliminate dimensional differences and individual variations in electrode-skin contact impedance, ECG voltage, and patch normal acceleration. Edge processing unit 140 performs robust normalization on the three data streams based on the median and median absolute deviation, respectively, to obtain normalized electrode-skin contact impedance, normalized ECG voltage, and normalized patch normal acceleration. For any sequence to be normalized... Robust normalization can be calculated using the following formula: In the above formula, Represents the first in the unified time grid Electrode-skin contact impedance sampling values ​​at each sampling point Indicates the first Standardized electrode-skin contact impedance sampling values ​​at each sampling point This indicates a zero-valued electrode-skin contact impedance. The dimension of this zero-valued electrode-skin contact impedance is consistent with the sampled electrode-skin contact impedance value, and can be one to three times the measurement resolution of impedance detection channel 120.

[0026] In the above formula, Represents the first in the unified time grid ECG voltage sampling values ​​at each sampling point Indicates the first Standardized ECG voltage sampling values ​​at each sampling point This indicates the zero-limit value of the ECG voltage. The dimension of the zero-limit value of the ECG voltage is consistent with the ECG voltage sampling value. It can be taken as one to three times the voltage resolution of the 110 analog-to-digital converter in the ECG acquisition channel, or as one-tenth to one-third of the standard amplitude of the high-frequency noise of the ECG in the quiet wearing sample.

[0027] In the above formula, Represents the first in the unified time grid The patch normal acceleration sampled at each sampling point Indicates the first Standardized patch normal acceleration sampled values ​​at each sampling point This represents the zero-limit value of the patch normal acceleration. The dimensions of the zero-limit value of the patch normal acceleration are consistent with the sampled value of the patch normal acceleration, and can be one to three times the resolution of the accelerometer.

[0028] when , or When approaching zero, the small quantities of electrode-skin contact impedance, ECG voltage, and patch normal acceleration each participate in the denominator stabilization of their respective channels and are not shared between different dimensions.

[0029] Example 3: The edge processing unit 140 performs time-delay coupling analysis on electrode-skin contact impedance, ECG voltage, and patch normal acceleration within a preset allowable time delay range. The time-delay coupling analysis includes normalized cross-correlation calculations between three pairs of data: electrode-skin contact impedance and ECG voltage, ECG voltage and patch normal acceleration, and patch normal acceleration and electrode-skin contact impedance. These three pairs of data correspond to… Figure 1 The data enters the three observation vertices of the edge processing unit 140 through the electrode-skin contact interface 100 via three acquisition channels.

[0030] The preset allowable latency range is used to limit the search interval for cross-correlation calculations. The edge processing unit 140 records the allowable latency length as... The unified time grid sampling interval is denoted as And based on this, the maximum allowable sample point offset is obtained. .

[0031] In the above formula, the allowable delay time length The unit of measurement is time, and the sampling interval of the unified time grid is... The unit of measurement is time, and the maximum allowable sample point offset is... The number of sample points is a dimensionless integer. The allowable delay time can be selected from 0.2s to 1.0s, preferably 0.5s, and is determined by a validation set of contact-prone segments, motion-interference segments, and normal wearing segments. This ensures that the inter-channel delay caused by actual contact disturbances is covered, while avoiding the introduction of random correlations by an excessively wide search interval.

[0032] For any pairwise sequence and Edge processing unit 140 has a latency The normalized cross-correlation is calculated based on the effective overlapping interval, and the formula is as follows: In the above formula, Indicates the sample point offset. Indicates the offset at the sample point The following conditions must be met simultaneously: both the first sequence sampling point and the second sequence sampling point fall into the set of valid overlapping sample points in the current analysis window. Indicates the number of samples in the current analysis window. (Using...) When summing, regardless of the sample point offset Is it a positive value, a zero value, or a negative value? and None of them will go beyond the current analysis window boundary.

[0033] In the above formula, This represents the offset between the first and second standardized sequences at sample points. The cross-correlation coefficients under the following conditions Indicates the first normalized sequence at the th Standardized sampled values ​​of each sampling point Indicates the second normalized sequence in the th order. Standardized sampled values ​​of each sampling point This represents the first normalized sequence in the set of effectively overlapping sample points. The mean of the above, This represents the second normalized sequence in the set of effectively overlapping sample points. The mean value at the corresponding offset position, This represents the zero-incremental quantity of cross-correlation. The zero-incremental quantity of cross-correlation is a dimensionless value and can be taken as... to This is only used to avoid the denominator being zero when the variances of two valid overlapping sequences are both close to zero.

[0034] Edge processing unit 140 only The cross-correlation coefficient is calculated only when the number of effective overlapping sample points is not less than the preset minimum number of overlapping sample points. The preset minimum number of overlapping sample points can be 80% of the number of samples in the current analysis window; when the number of effective overlapping sample points corresponding to a certain offset is insufficient, that sample point offset will not participate in the peak search.

[0035] Edge processing unit 140 searches for the position with the highest cross-correlation for each pair of sequences within a preset allowable delay range, and determines the delay corresponding to the highest cross-correlation as the optimal delay, and the cross-correlation corresponding to the optimal delay as the maximum coupling strength. The maximum coupling strength and the optimal delay can be determined by the following formula: In the above formula, This indicates the peak correlation strength between the first and second normalized sequences. This represents the peak sample point offset between the first and second normalized sequences. This represents the peak time delay between the first and second normalized sequences. Indicates the uniform time grid sampling interval. Peak sample point offset. The number of dimensionless integer sample points, peak delay The dimension of is time.

[0036] Edge processing unit 140 calculates the offset of the first peak sample point between the standardized electrode-skin contact impedance sequence and the standardized electrocardiogram voltage sequence. Correlation intensity with the first peak The offset of the second peak sample point was calculated between the standardized ECG voltage sequence and the standardized patch normal acceleration sequence. Second peak correlation intensity The offset of the third peak sample point was calculated based on the standardized patch normal acceleration sequence and the standardized electrode-skin contact impedance sequence. Correlation intensity with the third peak The offsets of the three peak sample points mentioned above are all based on the number of sample points, and the correlation intensities of the three peaks are all dimensionless values.

[0037] When multiple candidate delays have the same degree of cross-correlation, the edge processing unit 140 can select the candidate delay with the smaller absolute value as the optimal delay to reduce the impact of accidental edge samples within the same analysis window on the delay result.

[0038] Example 4: Edge processing unit 140 determines the time delay triangle closure error based on three optimal time delays. The three optimal time delays correspond to the optimal time delay between electrode-skin contact impedance and ECG voltage, the optimal time delay between ECG voltage and patch normal acceleration, and the optimal time delay between patch normal acceleration and electrode-skin contact impedance, respectively. Following a unified positive time delay direction, the above three optimal time delays form a closed loop from electrode-skin contact impedance to ECG voltage, from ECG voltage to patch normal acceleration, and then from patch normal acceleration back to electrode-skin contact impedance.

[0039] If the three data streams are primarily generated by the same electrode-skin contact interface perturbation, the three optimal time delays should be nearly closed under a unified positive time delay direction. If the sum of the three optimal time delays deviates significantly from zero, it indicates that while there may be some pairwise correlations among the three data streams, a closed time structure generated by the same electrode-skin contact interface perturbation is lacking. The time delay triangle closure error can be calculated using the following formula: In the above formula, This indicates the time delay triangle closure error. This represents the offset of the first peak sample point between the standardized electrode-skin contact impedance sequence and the standardized ECG voltage sequence. This represents the offset of the second peak sample point between the standardized ECG voltage sequence and the standardized patch normal acceleration sequence. This represents the offset of the third peak sample point between the standardized patch normal acceleration sequence and the standardized electrode-skin contact impedance sequence. This represents the maximum allowed sample point offset. This represents the normalized anti-zero decimal of the time delay. The dimension of the normalized anti-zero decimal of the time delay is the number of sample points, which can be fixed to 1 sample point to keep the denominator consistent with the dimension of the sum of the offsets of the three peak sample points.

[0040] Edge processing unit 140 determines the tri-coupling strength based on the three maximum coupling strengths. The tri-coupling strength represents the degree to which the three maximum coupling strengths coexist and suppresses the dominant role of any single pair of high correlation between electrode-skin contact impedance, ECG voltage, and patch normal acceleration in alarm decision-making. The tri-coupling strength can be calculated using the following formula: In the above formula, Indicates the three-sided coupling strength; This indicates the maximum coupling strength between the electrode-skin contact impedance and the ECG voltage. This represents the maximum coupling strength between the ECG voltage and the normal acceleration of the patch. This represents the maximum coupling strength between the patch normal acceleration and the electrode-skin contact impedance. When using the geometric mean, a lower maximum coupling strength at any point will reduce the three-sided coupling strength, thus avoiding the risk of assuming simultaneous electrode-skin contact interface coupling based on a single pair of data with accidental high correlation.

[0041] The time delay triangle closure error and the three-sided coupling strength describe two aspects of the same time structure. The time delay triangle closure error reflects the degree of deviation when the three optimal time delays form a closed loop in the same positive time delay direction, while the three-sided coupling strength reflects the degree to which the three maximum coupling strengths exist simultaneously. The edge processing unit 140 combines these two quantities to distinguish between different cases such as unilateral strong correlation, bilateral strong correlation, and three-sided closure strong correlation.

[0042] Example 5: Edge processing unit 140 generates an electro-mechanical impedance time-delay triangle closure feature based on the time-delay triangle closure error and the triangular coupling strength. This electro-mechanical impedance time-delay triangle closure feature is used to characterize the electrode-skin contact interface coupling risk in alarm decision-making. Its calculation result increases with increasing triangular coupling strength and decreases with increasing time-delay triangle closure error. The electro-mechanical impedance time-delay triangle closure feature can be calculated using the following formula: In the above formula, This indicates the closed-loop characteristic of the electromechanical impedance time delay triangle; Indicates the three-sided coupling strength; This indicates the time delay triangle closure error; This represents an exponential function. The higher the closure characteristic of the electromechanical-impedance time delay triangle, the higher the probability that the three data streams—electrode-skin contact impedance, ECG voltage, and patch normal acceleration—have strong coupling and that the optimal time delay forms a closed loop. The lower the closure characteristic of the electromechanical-impedance time delay triangle, the weaker the coupling edge or the unclosed time delay loop.

[0043] The physical significance of the electromechanical impedance time delay triangle closure characteristic lies in the fact that when the same electrode-skin contact interface 100 is affected by body movement, traction, or patch micro-slippage, changes in contact area and local pressure can affect the electrode-skin contact impedance. Mechanical disturbances can be reflected in the patch normal acceleration, and interface charge disturbances and common-mode changes can manifest as artifacts in the ECG voltage. If the three data streams are driven by the same disturbance, a stable time delay propagation relationship may emerge within the preset allowable time delay range, thus increasing the electromechanical impedance time delay triangle closure characteristic.

[0044] Real malignant arrhythmic events mainly alter the rhythm and morphology of the electrocardiogram voltage, and usually do not necessarily cause a strong and closed three-sided relationship between electrode-skin contact impedance and patch normal acceleration.

[0045] The artifact coupling threshold is used to determine whether the electromechanical-impedance delay triangular closure characteristic meets the contact artifact coupling condition. The artifact coupling threshold is pre-determined by a calibration sample set, which includes manually annotated or synchronously annotated segments of poor contact, body motion interference, normal wear, and malignant arrhythmias. The configuration tool calculates the contact artifact detection rate, malignant arrhythmia false suppression rate, and normal wear false indication rate for each candidate artifact coupling threshold, and selects the candidate value with the highest contact artifact detection rate provided that the malignant arrhythmia false suppression rate does not exceed a preset upper limit.

[0046] In the wearable ECG patch embodiment, the artifact coupling threshold can be selected between 0.60 and 0.85. When the ECG acquisition channel 110 has weak resistance to motion artifacts, a lower artifact coupling threshold can be used to improve the alerting sensitivity. When the patch is fixed stably and the impedance detection channel 120 has low noise, a higher artifact coupling threshold can be used to reduce false alerts. The artifact coupling threshold is re-verified after device factory calibration, software version update, or sensor model change, and does not adaptively change with a single analysis window during daily operation.

[0047] The calibration criteria for the artifact coupling threshold are as follows: the contact artifact detection rate of contact artifact segments is not less than 0.90; the false suppression rate of malignant arrhythmias in marked malignant arrhythmia segments that are delayed or transferred for review due to the electro-mechanical-impedance time delay triangle closure characteristic being not less than the artifact coupling threshold is not higher than 0.01; and the lead contact abnormality indication rate of normal stable segments is not higher than 0.05. If all three conditions cannot be met simultaneously, the false suppression rate of malignant arrhythmias not higher than 0.01 is taken as the priority condition, and then the artifact coupling threshold with the higher contact artifact detection rate is selected from the candidate values ​​that meet this priority condition.

[0048] After calculating the electromechanical-impedance time delay triangle closure characteristics, the edge processing unit 140 can write the three pairs of normalized cross-correlations, three optimal time delays, three maximum coupling strengths, time delay triangle closure error, three-sided coupling strength, and electromechanical-impedance time delay triangle closure characteristics as intermediate results into the event logging module 170. The event logging module 170 does not participate in the identification of malignant arrhythmias, but its stored content can indicate whether the alarm decision is affected by the electrode-skin contact interface coupling risk.

[0049] Example 6: The heart rhythm recognition module 150 can employ an embedded ECG classification model, a rule classifier, or a combination of both to identify malignant arrhythmias from the ECG voltage sequence received from the ECG acquisition channel 110. The heart rhythm recognition module 150 outputs a confidence score for the malignant arrhythmia. The confidence level for malignant arrhythmias ranges from 0 to 1. The higher the value, the higher the probability that the current analysis window contains a malignant arrhythmia.

[0050] When the heart rhythm recognition module 150 directly outputs the probability, the edge processing unit 140 uses this probability as the confidence level for malignant arrhythmia; when the heart rhythm recognition module 150 outputs an uncalibrated score... At that time, the offline configuration tool uses a labeled validation set to perform monotonic calibration on the score and writes the calibration coefficients into the edge processing unit 140. When using logical calibration, the confidence level of malignant arrhythmia can be obtained by the following formula.

[0051] In the above formula, This indicates the uncalibrated score output by the heart rhythm recognition module 150. Indicates the confidence level calibration slope. This represents the confidence calibration intercept. The confidence calibration slope and confidence calibration intercept are obtained by fitting malignant arrhythmia labels and non-malignant arrhythmia labels in the validation set and are used as a fixed configuration in the edge processing unit 140. If equidistant binning calibration or ordinal-preserving regression calibration is used, the edge processing unit 140 uses the corresponding lookup table to obtain the confidence level of malignant arrhythmias in the range of 0 to 1.

[0052] Uncalibrated score The confidence calibration slope for the dimensionless score output by the heart rhythm recognition module 150. and confidence calibration intercept Used to make the independent variable of the exponential function Maintain dimensionless dimensions. When using logical calibration, the configuration tool uses the labels of malignant and non-malignant arrhythmias in the validation set as supervisory information, minimizes the binary classification cross-entropy loss, and uses a binning error of no more than 0.05 for the confidence level of malignant arrhythmias after calibration as the calibration pass criterion. The validation set does not overlap with the training set of the heart rhythm recognition module 150; when the version of the heart rhythm recognition module 150 is updated or the sampling configuration of the ECG acquisition channel 110 changes, the confidence calibration slope and confidence calibration intercept are redefined.

[0053] The heart rhythm recognition window of the heart rhythm recognition module 150 and the analysis window of the edge processing unit 140 use the same unified time grid. When the heart rhythm recognition model requires a shorter rhythm segment, the heart rhythm recognition window is located inside the current analysis window, and the confidence level of malignant arrhythmia is generated using the effective R peak, beat interval, and ECG waveform segment within the current analysis window. When the heart rhythm recognition model requires an input of the same length as the analysis window, the heart rhythm recognition window coincides with the current analysis window. Regardless of the method used, the confidence level of malignant arrhythmia output to the alarm decision generation process always includes the start and end times of the current analysis window.

[0054] Edge processing unit 140 generates alarm decisions based on the confidence level of malignant arrhythmia and the closed-loop characteristics of the electro-mechanical-impedance time delay triangle, and controls alarm unit 160 to output a malignant arrhythmia alarm according to the alarm decisions. Alarm decisions include an emergency protection path, a normal alarm path, and a contact anomaly indication path. The emergency protection path has the highest priority; the normal alarm path is used for continuous confirmation of events with medium confidence; and the contact anomaly indication path is used for verification or indication when the closed-loop characteristics of the electro-mechanical-impedance time delay triangle indicate a strong contact artifact.

[0055] When the confidence level of a malignant arrhythmia reaches the emergency protection threshold, the edge processing unit 140 immediately generates an alarm decision to output a malignant arrhythmia alarm. At this time, regardless of whether the electro-mechanical impedance time delay triangle closure characteristic is not lower than the artifact coupling threshold, the edge processing unit 140 does not use the electro-mechanical impedance time delay triangle closure characteristic as a condition for canceling the alarm. The alarm unit 160 can output a malignant arrhythmia alarm through buzzer, vibration, screen, bedside light, or communication notification, and can simultaneously record the electrode-skin contact interface coupling risk for verification.

[0056] When the confidence level of a malignant arrhythmia is lower than the emergency protection threshold but reaches the normal alarm threshold, the edge processing unit 140 generates a continuous confirmation flag based on the malignant arrhythmia confidence level meeting the normal alarm threshold within a continuous analysis window. The continuous confirmation flag can indicate that several consecutive analysis windows meet the normal alarm threshold, or it can indicate that the duration of the same event reaches a preset duration requirement. The continuous confirmation flag is used to prevent the accidental classification result of a single, short-lived analysis window from directly triggering a normal alarm.

[0057] When the electromechanical-impedance delay triangle closure characteristic is below the artifact coupling threshold and the continuous confirmation flag indicates that continuous confirmation has passed, the edge processing unit 140 generates an alarm decision to output a malignant arrhythmia alarm. This path indicates that the confidence level of the malignant arrhythmia has reached the ordinary alarm threshold and passed continuous confirmation, while the electromechanical-impedance delay triangle closure characteristic does not indicate a strong contact artifact. Therefore, the alarm unit 160 outputs a malignant arrhythmia alarm.

[0058] When the confidence level of a malignant arrhythmia is lower than the emergency protection threshold and reaches the normal alarm threshold, and the electro-mechanical impedance time delay triangle closure characteristic is not lower than the artifact coupling threshold, the edge processing unit 140 performs a short confirmation window review and updates the alarm decision based on the result of the short confirmation window review. Simultaneously, it controls the alarm unit 160 to output a lead contact abnormality alert. The short confirmation window review can add a short-time analysis window immediately after the current analysis window to recalculate the confidence level of the malignant arrhythmia and the electro-mechanical impedance time delay triangle closure characteristic to determine whether the moderate confidence event persists and whether the electrode-skin contact interface coupling risk still exists.

[0059] Alarm decisions can be represented by the following logic: In the above formula, Indicates the decision to call the police. This indicates that a malignant arrhythmia alarm will be output. Indicates the confidence level of malignant arrhythmia. Indicates the emergency protection threshold. Indicates the normal alarm threshold. Indicates the pseudo-coupling threshold. This represents the closed-loop characteristic of the electromechanical impedance time delay triangle. This indicates a continuous confirmation flag. In the second formula, This indicates a short confirmation window for alarm decision-making. This indicates the confidence level of malignant arrhythmias within a short confirmation window. This indicates the closed-loop characteristic of the short confirmation window electromechanical-impedance time delay triangle. This indicates a short confirmation window with continuous confirmation.

[0060] If the confidence level of a malignant arrhythmia is not lower than the emergency protection threshold, the edge processing unit 140 directly outputs a malignant arrhythmia alarm through the alarm unit 160, and saves the electro-mechanical-impedance time delay triangle closure characteristics as event interpretation information. This branch is used to cover situations requiring immediate response, such as ventricular fibrillation and sustained ventricular tachycardia, and the alarm is not delayed due to contact artifact coupling conditions.

[0061] If the confidence level of malignant arrhythmia is lower than the emergency protection threshold but not lower than the normal alarm threshold, the edge processing unit 140 determines whether the electro-mechanical impedance time delay triangle closure characteristic is lower than the pseudocoupling threshold and whether the continuous confirmation flag is 1. Only when the electro-mechanical impedance time delay triangle closure characteristic is lower than the pseudocoupling threshold and the continuous confirmation flag is 1, the alarm unit 160 outputs a malignant arrhythmia alarm; if the electro-mechanical impedance time delay triangle closure characteristic is not lower than the pseudocoupling threshold, the alarm unit 160 outputs a lead contact abnormality indication or motion interference indication, and temporarily does not output a malignant arrhythmia alarm.

[0062] The short confirmation window length is used to supplement a quick verification within the normal alarm range. The short confirmation window length can be 1s to 3s, preferably 2s, and should be consistent with or not exceed twice the sliding step size of the analysis window. The short confirmation window extends backward from the end of the current analysis window. Within the short confirmation window, the edge processing unit 140 recalculates the short confirmation window malignant arrhythmia confidence, the short confirmation window electromechanical-impedance time delay triangle closure characteristic, and the short confirmation window continuous confirmation flag.

[0063] The short confirmation window decision has three possible outcomes. The first outcome is that the confidence level of the malignant arrhythmia within the short confirmation window is not lower than the emergency protection threshold, and the alarm unit 160 immediately outputs a malignant arrhythmia alarm. The second outcome is that the confidence level of the malignant arrhythmia within the short confirmation window is not lower than the normal alarm threshold, the electrical-mechanical-impedance delay triangle closure characteristic of the short confirmation window is lower than the pseudo-coupling threshold, and the continuous confirmation flag of the short confirmation window is 1, and the alarm unit 160 outputs a malignant arrhythmia alarm. The third outcome is that the confidence level of the malignant arrhythmia within the short confirmation window is lower than the normal alarm threshold, or the electrical-mechanical-impedance delay triangle closure characteristic of the short confirmation window is not lower than the pseudo-coupling threshold, and the alarm unit 160 does not output a malignant arrhythmia alarm; if the electrical-mechanical-impedance delay triangle closure characteristic of the short confirmation window is not lower than the pseudo-coupling threshold in the third outcome, then the alarm unit 160 outputs a lead contact abnormality indication or a motion interference indication.

[0064] The emergency protection threshold is used to determine whether the confidence level of a malignant arrhythmia is sufficient to directly trigger an alarm, while the general alarm threshold is used to determine whether the confidence level of a malignant arrhythmia has entered an interval requiring further discrimination based on the closed characteristics of the electromechanical-impedance time delay triangle. Both the emergency protection threshold and the general alarm threshold are predetermined by a labeled validation set, which includes at least ventricular fibrillation, sustained ventricular tachycardia, torsades de pointes, severe bradycardia, normal sinus rhythm, atrial or ventricular premature beats, and contact artifact fragments.

[0065] The offline configuration tool selects the highest threshold that meets the emergency alarm sensitivity target from the candidate emergency protection thresholds to reduce direct alarms from non-emergency segments; and selects the lowest threshold that meets the early warning sensitivity target from the candidate general alarm thresholds to allow potentially malignant arrhythmia segments to enter a short confirmation process. For wearable ECG patches, the emergency protection threshold can be selected between 0.90 and 0.98, preferably 0.95; the general alarm threshold can be selected between 0.50 and 0.80, preferably 0.60, and the general alarm threshold is lower than the emergency protection threshold.

[0066] The calibration criteria for the emergency protection threshold are as follows: the immediate alarm sensitivity for segments of ventricular fibrillation, sustained ventricular tachycardia, torsades de pointes, and severe bradycardia should be no less than 0.99, and the proportion of non-malignant arrhythmia segments directly triggering malignant arrhythmia alarms should be no more than 0.05. The calibration criteria for the normal alarm threshold are as follows: the sensitivity for segments of ventricular arrhythmia marked as malignant to enter the normal alarm candidate state should be no less than 0.95, and the proportion of normal stable segments entering the normal alarm candidate state should be no more than 0.20. If the candidate thresholds cannot simultaneously meet the sensitivity and false alarm constraints, the immediate alarm sensitivity corresponding to the emergency protection threshold should be prioritized, and then false alarms should be reduced through continuous confirmation flags, artifact coupling thresholds, and short confirmation windows.

[0067] Emergency protection thresholds and general alarm thresholds are used consistently under the same device model, the same version of the heart rhythm recognition module 150, and the same sampling configuration. When the version of the heart rhythm recognition module 150 is updated, the sampling rate of the ECG acquisition channel 110 changes, a new type of malignant arrhythmia is added to the validation set, or the output strategy of the alarm unit 160 changes, the emergency protection thresholds and general alarm thresholds are recalibrated. After recalibration, the event logging module 170 saves the new threshold configuration version, ensuring that alarm cause verification corresponds to the emergency protection thresholds and general alarm thresholds used at that time.

[0068] The continuous acknowledgment flag indicates whether the confidence level of a malignant arrhythmia continuously meets the ordinary alarm conditions over time. The edge processing unit 140 can set the continuous acknowledgment flag to 1 when a malignant arrhythmia confidence level is not lower than the ordinary alarm threshold detected in two or more consecutive analysis windows; alternatively, it can set the continuous acknowledgment flag to 1 when the cumulative duration reaches 6 to 10 seconds. The generation rule for the continuous acknowledgment flag remains fixed during the same device configuration and is written to the event logging module 170 with each analysis window.

[0069] Example 7: The event logging module 170 is used to correlate and store the ECG voltage before and after the alarm, routine ECG characteristics, confidence level of malignant arrhythmias, electrode-skin contact impedance, patch normal acceleration, maximum coupling strength, optimal time delay, electromechanical-impedance time delay triangle closure characteristics, and alarm decision, for use in alarm cause review or equipment performance verification. The correlated storage can be indexed according to the timestamp of a unified equipment clock, allowing medical staff or equipment maintenance personnel to simultaneously view cardiac rhythm evidence and contact interface coupling evidence during review.

[0070] The event logging module 170 can save the original ECG voltage, electrode-skin contact impedance, and patch normal acceleration for a fixed time period before and after an alarm event. It can also save intermediate sequences after time alignment and scaling. For each analysis window, the event logging module 170 can save the corresponding maximum coupling strength, optimal time delay, time delay triangle closure error, triangular coupling strength, electromechanical-impedance time delay triangle closure characteristics, confidence level of malignant arrhythmia, and alarm decision.

[0071] The basic quality state is a quality indicator used to determine whether the current analysis window meets the calculation prerequisites. It is generated by the edge processing unit 140 before entering the heart rhythm recognition module 150 and time delay coupling analysis. The basic quality state includes at least the following: timestamp abnormality state, sampling sequence number abnormality state, ECG voltage saturation state, ECG voltage noise state, electrode-skin contact impedance exceeding the limit state, and patch normal acceleration exceeding the limit state. All of the above states are judged based on the same analysis window and are written to the event logging module 170 along with the data frame of that analysis window.

[0072] Timestamp anomaly status is used to reflect the reliability of sampling time before the establishment of a unified time grid. When the deviation of adjacent raw sampling time intervals from the target sampling interval exceeds one-tenth of the target sampling interval, or when time reversal, duplicate timestamps, or missing timestamps occur, the edge processing unit 140 marks the corresponding channel as having a timestamp anomaly status. For embodiments using hardware synchronization clocks, timestamp anomaly status can also be triggered by a synchronization loss flag given by the acquisition chip.

[0073] The sampling sequence number abnormality status is used to reflect the integrity of the continuous sampling sequence. When the sampling sequence number of any channel is not incremented by one, or when the number of missing sampling points in an analysis window exceeds one percent of the number of samples in that analysis window, the edge processing unit 140 marks the channel as having an abnormal sampling sequence number. If the number of missing sampling points does not exceed the above proportion, linear interpolation of adjacent valid sampling points can be used to fill in the gaps, and the interpolation position is recorded in the event logging module 170.

[0074] The ECG voltage saturation state is used to reflect whether the ECG acquisition channel 110 is at the range boundary of the analog-to-digital converter. When two or more consecutive ECG voltage sampling points are equal to or close to the upper or lower limit code value of the ECG acquisition channel 110, or when the acquisition chip outputs a saturation flag, the edge processing unit 140 marks the corresponding analysis window as ECG voltage saturation state. The judgment margin for approaching the range boundary can be taken as one-thousandth to one-hundredth of the full range of the ECG acquisition channel 110, and is fixed within the same equipment model.

[0075] The ECG voltage noise state is used to reflect whether the ECG voltage sequence can still be used for malignant arrhythmia identification. The edge processing unit 140 can calculate the root mean square amplitude of the ECG high-frequency residual after removing baseline drift and compare it with the effective amplitude of the ECG signal; when this ratio exceeds a pre-determined ECG noise ratio threshold based on validation data, the ECG voltage noise state is marked. The ECG noise ratio threshold is typically selected between 0.2 and 0.5 and is calibrated with the primary constraint of not omitting any malignant arrhythmia segments.

[0076] The electrode-skin contact impedance out-of-bounds status is used to reflect whether the electrode-skin contact impedance sequence acquired by the impedance detection channel 120 is within the interpretable range. When the electrode-skin contact impedance sample value exceeds the effective measurement range specified by the patch electrode and impedance sampling circuit, or when the abrupt change amplitude within an analysis window exceeds the impedance abrupt change threshold determined by the contact state verification data, the edge processing unit 140 marks the electrode-skin contact impedance out-of-bounds status. The impedance abrupt change threshold can be taken as two to three times the upper quantile of the impedance change amplitude in the quiet wearing sample.

[0077] The patch normal acceleration out-of-bounds status is used to reflect whether the motion data of the acceleration acquisition channel 130 can be used for artifact detection. When the patch normal acceleration sample value reaches the range boundary of the accelerometer, is continuously missing, or the patch normal acceleration amplitude after gravity removal exceeds 80% of the sensor range, the edge processing unit 140 marks the patch normal acceleration out-of-bounds status. If only a short-term violent motion occurs and the range boundary is not reached, this status does not directly prohibit calculation, but is saved with the event as the basis for artifact indication.

[0078] For example, in a complete workflow embodiment, after the patient wears the wearable ECG patch, the system first establishes an association between the patient identifier, device identifier, and monitoring task, and reads the device's battery level, storage, clock, and network status. Then, electrodes or the ECG patch are attached in standard positions, and polarity, adhesion, and comfort are checked. Self-tests are performed for lead dislodgement, electrode-skin contact impedance, analog front-end, and analog-to-digital conversion. After the self-tests pass, ECG acquisition channel 110, impedance detection channel 120, and acceleration acquisition channel 130 begin continuous acquisition.

[0079] In the At the start of each analysis window, the edge processing unit 140 reads the ECG voltage, electrode-skin contact impedance, and patch normal acceleration within the same analysis window from the circular buffer, and reads the corresponding timestamp, sampling number, data saturation flag, and missing flag. The edge processing unit 140 first determines if any of the electrode-skin contact impedance, ECG voltage, or patch normal acceleration is missing, and whether there are any abnormal timestamps, abnormal sampling numbers, data saturation, missing flags, or values ​​exceeding the preset quality range. If quality is unavailable, the current analysis window does not calculate the electromechanical-impedance time delay triangle closure characteristic and generates a basic quality state.

[0080] If the data in the current analysis window is available, the edge processing unit 140 resamples the electrode-skin contact impedance, ECG voltage, and patch normal acceleration to a unified time grid, and performs robust normalization based on the median and median absolute deviation, respectively, to obtain normalized electrode-skin contact impedance, normalized ECG voltage, and normalized patch normal acceleration. Subsequently, within a preset allowable time delay range, the edge processing unit 140 calculates the normalized cross-correlation between the three pairs of data based on the effective overlap interval.

[0081] Edge processing unit 140 searches for the position with the highest cross-correlation degree among the three pairs of normalized cross-correlation results, determines the time delay corresponding to the highest cross-correlation degree as the optimal time delay, and determines the cross-correlation degree corresponding to the optimal time delay as the maximum coupling strength. Subsequently, edge processing unit 140 determines the time delay triangle closure error based on the three optimal time delays, determines the three-sided coupling strength based on the three maximum coupling strengths, and generates the electromechanical-impedance time delay triangle closure feature from the time delay triangle closure error and the three-sided coupling strength.

[0082] Within the same analysis window or adjacent rhythm recognition windows, the rhythm recognition module 150 generates routine ECG characteristics based on ECG voltage and outputs a malignant arrhythmia confidence level. After receiving the malignant arrhythmia confidence level, the edge processing unit 140 first determines whether the confidence level reaches the emergency protection threshold. If the emergency protection threshold is reached, the edge processing unit 140 immediately generates an alarm decision to output a malignant arrhythmia alarm and controls the alarm unit 160 to output the malignant arrhythmia alarm.

[0083] If the confidence level of a malignant arrhythmia is lower than the emergency protection threshold but reaches the normal alarm threshold, the edge processing unit 140 determines the continuous confirmation flag and the electro-mechanical impedance time delay triangle closure characteristic. If the electro-mechanical impedance time delay triangle closure characteristic is lower than the pseudocoupling threshold and the continuous confirmation flag indicates that continuous confirmation has passed, the edge processing unit 140 generates an alarm decision to output a malignant arrhythmia alarm. If the electro-mechanical impedance time delay triangle closure characteristic is not lower than the pseudocoupling threshold, the edge processing unit 140 performs a short confirmation window review and controls the alarm unit 160 to output a lead contact abnormality indication.

[0084] After alarm unit 160 outputs an alarm or notification, event recording module 170 saves the ECG voltage before and after the alarm, routine ECG characteristics, confidence level of malignant arrhythmia, electrode-skin contact impedance, patch normal acceleration, maximum coupling strength, optimal time delay, electromechanical-impedance time delay triangular closure characteristics, and alarm decision. If the communication link is available, the system can send the alarm and event record to the nurse station, doctor terminal, or remote platform; if the communication link is unavailable, the system retains the local alarm and queues it for retransmission.

[0085] As an optional implementation, the patient is at rest or in a state of slight activity, and the ECG patch maintains stable contact with the skin. The ECG voltage acquired by the ECG acquisition channel 110 has a stable QRS complex and a relatively flat baseline, the electrode-skin contact impedance acquired by the impedance detection channel 120 is within the normal quality range, and the patch normal acceleration acquired by the acceleration acquisition channel 130 does not show any significant abrupt changes. After the edge processing unit 140 completes time alignment and scale unification within the same analysis window, the normalized cross-correlation between the three pairs of data will not simultaneously form a strong and closed time delay structure.

[0086] In this implementation, the triangular coupling strength typically does not continuously increase, or even if a pair of data is accidentally correlated, the time delay triangle closure error will not remain low simultaneously. The electromechanical-impedance time delay triangle closure feature generated by the edge processing unit 140 is lower than the pseudo-coupling threshold. If the confidence level of the malignant arrhythmia output by the heart rhythm recognition module 150 is lower than the normal alarm threshold, the edge processing unit 140 does not generate an alarm decision to output a malignant arrhythmia alarm; if the confidence level of the malignant arrhythmia reaches the normal alarm threshold and is continuously confirmed, and the electromechanical-impedance time delay triangle closure feature is lower than the pseudo-coupling threshold, the alarm unit 160 outputs a malignant arrhythmia alarm according to the normal alarm path.

[0087] As an optional implementation, when the patient turns over or the lead wires are slightly stretched, a short-term disturbance occurs at the electrode-skin contact interface 100. This disturbance causes fluctuations in the electrode-skin contact impedance, resulting in corresponding mechanical changes in the patch normal acceleration and artifact segments in the ECG voltage. After reading the three data streams within the same analysis window, the edge processing unit 140 eliminates dimensional differences through time alignment and scale unification, and then calculates the normalized cross-correlation between the three pairs of data within a preset allowable time delay range.

[0088] If the electrode-skin contact impedance, ECG voltage, and patch normal acceleration are caused by the same perturbation, then high maximum coupling strengths may simultaneously exist between the electrode-skin contact impedance and ECG voltage, the ECG voltage and patch normal acceleration, and the patch normal acceleration and electrode-skin contact impedance. Furthermore, the three optimal time delays form a nearly closed loop along a unified positive time delay direction. In this case, the three-sided coupling strength increases, the time delay triangle closure error decreases, and the electromechanical-impedance time delay triangle closure characteristic is not lower than the artifact coupling threshold.

[0089] If the heart rhythm recognition module 150 provides a malignant arrhythmia confidence level below the emergency protection threshold but reaching the normal alarm threshold under this operating condition, the edge processing unit 140 does not directly cancel the event, but instead performs a short confirmation window review and controls the alarm unit 160 to output a lead contact abnormality alert. The short confirmation window review is used to determine whether the abnormal morphology in the ECG voltage persists and whether the electrode-skin contact interface coupling risk still exists. If the malignant arrhythmia confidence level decreases after review and the electro-mechanical-impedance delay triangle closure characteristic still indicates a contact abnormality, the alarm decision can remain a lead contact abnormality alert or an event pending review; if the malignant arrhythmia confidence level increases to the emergency protection threshold after review, a malignant arrhythmia alarm is immediately output.

[0090] As an optional implementation, the impedance detection channel 120 may experience missing electrode-skin contact impedance, the acceleration acquisition channel 130 may experience patch normal acceleration data saturation, or the ECG acquisition channel 110 may experience ECG voltage noise exceeding the preset quality range. When reading data from the same analysis window, the edge processing unit 140 first checks whether the three data streams are complete and whether the quality markers are available. If any of the electrode-skin contact impedance, ECG voltage, or patch normal acceleration is missing, or if there is at least one quality unavailability condition such as abnormal timestamp, abnormal sampling sequence number, data saturation, missing markers, or exceeding the preset quality range, the calculation of the electromechanical-impedance delay triangular closure characteristic of the current analysis window is stopped.

[0091] After ceasing calculation of the electromechanical impedance time delay triangle closure characteristics of the current analysis window, the edge processing unit 140 generates a baseline quality status based on lead detachment, electrode-skin contact impedance exceeding limits, patch normal acceleration exceeding limits, ECG voltage saturation, and ECG voltage noise. When the baseline quality status indicates a contact abnormality, the alarm decision includes outputting a lead contact abnormality prompt or performing a confirmation review. If the heart rhythm recognition module 150 simultaneously outputs a malignant arrhythmia confidence level that reaches the emergency protection threshold, the edge processing unit 140 still immediately outputs a malignant arrhythmia alarm according to the emergency protection path.

[0092] As an optional implementation, the heart rhythm recognition module 150 bases its confidence level on the malignant arrhythmia when the ECG voltage output reaches the emergency protection threshold. Even if the electro-mechanical-impedance delay triangle closure feature within the same analysis window is not lower than the artifact coupling threshold, the edge processing unit 140 does not use the electro-mechanical-impedance delay triangle closure feature as a condition for canceling the alarm, but instead immediately generates an alarm decision to output a malignant arrhythmia alarm. The alarm unit 160 outputs a malignant arrhythmia alarm, and the event recording module 170 saves the electro-mechanical-impedance delay triangle closure feature and the alarm decision for subsequent determination of whether the event is accompanied by contact interface disturbance.

[0093] This implementation illustrates that the electromechanical-impedance time delay triangle closure feature is only used for alarm confirmation and electrode-skin contact interface coupling risk assessment, and does not replace the high-confidence malignant arrhythmia identification results of the heart rhythm recognition module 150. Through the emergency protection threshold, the system can prevent quality gating from causing unreasonable delays for high-risk events.

[0094] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A real-time warning system for malignant arrhythmia based on edge computing, comprising an electrocardiogram acquisition channel for acquiring electrocardiogram voltage, an impedance detection channel for acquiring electrode-skin contact impedance, an acceleration acquisition channel for acquiring patch normal acceleration, a heart rhythm recognition module, an edge processing unit and an alarm unit, characterized in that, The edge processing unit is configured as follows: The electrode-skin contact impedance, the electrocardiogram voltage, and the patch normal acceleration are time-aligned and scaled within the same analysis window; Within a preset allowable time delay range, time delay coupling analysis is performed on the electrode-skin contact impedance, the electrocardiogram voltage, and the patch normal acceleration to determine the maximum coupling strength and the optimal time delay; The time delay triangle closure error is determined based on the optimal time delay, the three-sided coupling strength is determined based on the maximum coupling strength, and the electro-mechanical-impedance time delay triangle closure feature is generated from the time delay triangle closure error and the three-sided coupling strength. The heart rhythm recognition module is configured to output a malignant arrhythmia confidence level based on the ECG voltage; the edge processing unit generates an alarm decision based on the malignant arrhythmia confidence level and the electromechanical impedance time delay triangle closure feature, and controls the alarm unit to output a malignant arrhythmia alarm based on the alarm decision; Specifically, when the confidence level of the malignant arrhythmia reaches the emergency protection threshold, the alarm decision is to immediately output the malignant arrhythmia alarm, and the electromechanical impedance time delay triangle closure characteristic is not used as a condition for canceling the alarm.

2. The edge computing based real-time warning system for malignant arrhythmia according to claim 1, characterized in that, The time alignment and scale uniformity include: The electrode-skin contact impedance, ECG voltage, and patch normal acceleration collected within the same analysis window are resampled to a unified time grid and robustly normalized based on the median and median absolute deviation, respectively, to obtain the normalized electrode-skin contact impedance, normalized ECG voltage, and normalized patch normal acceleration. 3.The real-time warning system for malignant arrhythmia based on edge computing electrocardiogram according to claim 2, characterized in that, The time delay coupling analysis includes: Within the preset allowable time delay range, the normalized cross-correlation between the electrode-skin contact impedance and the ECG voltage, the ECG voltage and the patch normal acceleration, and the patch normal acceleration and the electrode-skin contact impedance are calculated based on the effective overlap interval. The time delay corresponding to the highest cross-correlation degree is determined as the optimal time delay, and the cross-correlation degree corresponding to the optimal time delay is determined as the maximum coupling strength.

4. The real-time early warning system for malignant arrhythmias based on edge computing in electrocardiogram according to claim 3, characterized in that, The time delay triangle closure error is used to represent the degree of deviation when the three optimal time delays form a closed loop in the same positive time delay direction.

5. The real-time early warning system for malignant arrhythmias based on edge computing in electrocardiogram according to claim 4, characterized in that, The trilateral coupling strength is used to represent the degree to which the three maximum coupling strengths exist simultaneously, and to suppress the dominant role of any single high correlation of any pair of the electrode-skin contact impedance, the ECG voltage, and the patch normal acceleration on the alarm decision.

6. The real-time early warning system for malignant arrhythmias based on edge computing in electrocardiogram according to claim 5, characterized in that, The electromechanical impedance time delay triangular closure feature is used to characterize the electrode-skin contact interface coupling risk in the alarm decision.

7. The real-time early warning system for malignant arrhythmias based on edge computing in electrocardiogram according to claim 1, characterized in that, The edge processing unit is further configured to: When the confidence level of the malignant arrhythmia is lower than the emergency protection threshold and reaches the normal alarm threshold, a continuous confirmation flag is generated based on the situation where the confidence level of the malignant arrhythmia meets the normal alarm threshold within the continuous analysis window. When the electromechanical impedance time delay triangle closure characteristic is lower than the pseudo-coupling threshold and the continuous confirmation flag indicates that continuous confirmation has passed, the alarm decision is generated to output the malignant arrhythmia alarm.

8. The real-time early warning system for malignant arrhythmias based on edge computing in electrocardiogram according to claim 7, characterized in that, The edge processing unit is further configured to: When the confidence level of the malignant arrhythmia is lower than the emergency protection threshold and reaches the normal alarm threshold, and the electromechanical impedance time delay triangle closure characteristic is not lower than the pseudocoupling threshold, a short confirmation window review is performed, and the alarm decision is updated according to the result of the short confirmation window review. At the same time, the alarm unit is controlled to output a lead contact abnormality prompt.

9. The real-time early warning system for malignant arrhythmias based on edge computing in electrocardiogram according to claim 1, characterized in that, It also includes an event logging module, and the heart rhythm recognition module is further configured to generate conventional electrocardiogram features based on the electrocardiogram voltage; The event recording module is used to associate and store the ECG voltage before and after the alarm, the normal ECG characteristics, the confidence level of the malignant arrhythmia, the electrode-skin contact impedance, the patch normal acceleration, the maximum coupling strength, the optimal time delay, the electromechanical-impedance time delay triangular closure characteristics, and the alarm decision, for use in alarm cause verification or equipment performance verification.

10. The real-time early warning system for malignant arrhythmias based on edge computing in electrocardiograms according to claim 1, characterized in that, The edge processing unit is further configured to: If any of the electrode-skin contact impedance, the ECG voltage, and the patch normal acceleration are missing, or if there is at least one quality unavailability condition such as abnormal timestamp, abnormal sampling sequence number, data saturation, missing marker, or exceeding the preset quality range, the calculation of the electromechanical-impedance time delay triangle closure feature of the current analysis window is stopped, and the basic quality state is generated based on lead detachment, electrode-skin contact impedance exceeding the limit, patch normal acceleration exceeding the limit, ECG voltage saturation, and ECG voltage noise. When the basic quality status indicates a contact abnormality, the alarm decision includes outputting a contact abnormality prompt or performing a confirmation review.

Images