A method and system for monitoring cardiac arrhythmias

Abstract

The application discloses a kind of arrhythmia monitoring method and system, it is related to arrhythmia monitoring technical field.The arrhythmia monitoring method, when monitoring object moves, based on setting time period, obtain the electrocardiosignal time series data and motion time series data of monitoring object, and pre-process input into pre-trained electrocardio feature extraction model and motion feature extraction model, extract electrocardio feature and motion feature;And respectively analyze electrocardio fluctuation index and motion state index;Based on electrocardio fluctuation index and motion state index, analyze arrhythmia fluctuation index, and judge whether monitoring object exists arrhythmia risk, the feature analysis of the present application through electrocardiosignal and motion time series data, it can accurately identify motion interference in electrocardiosignal and real arrhythmia, by the joint analysis of electrocardio feature and motion feature, can effectively distinguish the difference between motion caused electrocardio fluctuation and arrhythmia, to avoid the influence of artifact on arrhythmia risk identification.

Description

Technical Field

[0001] This invention relates to the field of cardiac arrhythmia monitoring technology, specifically to a cardiac arrhythmia monitoring method and system. Background Technology

[0002] With the increasing prevalence of cardiovascular diseases, arrhythmia has become one of the most significant health problems affecting humanity worldwide. Arrhythmia is usually caused by abnormal electrical activity in the heart, leading to irregular heartbeats and affecting the efficiency of blood circulation and oxygen supply to various parts of the body. The occurrence of arrhythmia not only impacts patients' quality of life but can also lead to serious health problems such as stroke, heart failure, and even sudden death. Therefore, early diagnosis and timely intervention of arrhythmia are crucial for improving patient survival rates and quality of life.

[0003] Currently, traditional methods for detecting cardiac arrhythmias largely rely on the acquisition and analysis of electrocardiogram (ECG) signals. ECG is a technique that directly reflects the heart's electrical activity; by recording fluctuations in the heart's electrical signals, doctors can determine the presence of arrhythmias. However, existing arrhythmia monitoring systems often only perform analysis in a resting state, making it difficult to handle artifacts and noise that occur during dynamic activities. Especially during exercise, ECG signals are affected by external factors such as body movement and postural changes, leading to signal distortion and reducing the accuracy and reliability of monitoring.

[0004] The limitations of existing technologies include at least the following problems: In practical applications, the monitored objects are usually in a non-resting state. Traditional methods are mostly based on resting electrocardiogram (ECG) signals for judgment, which fail to fully consider the interference caused by factors such as changes in body position and muscle disturbances during exercise. This makes it easy for ECG abnormalities during dynamic processes to be misjudged as physiological fluctuations, or for real abnormalities to be masked by interference signals, resulting in the failure of arrhythmia risk identification. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for monitoring cardiac arrhythmias, solving the problem that existing technologies struggle to accurately identify the risk of cardiac arrhythmias during exercise.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring cardiac arrhythmias, comprising the following steps: when the monitored object is moving, acquiring the electrocardiogram (ECG) signal time-series data and motion time-series data of the monitored object based on a set time period, and preprocessing them; inputting the preprocessed ECG signal time-series data and motion time-series data into a pre-trained ECG feature extraction model and a motion feature extraction model, respectively, to extract the ECG features and motion features of the monitored object; analyzing the ECG fluctuation index and motion state index of the monitored object based on the ECG features and motion features; analyzing the heart rhythm fluctuation index of the monitored object based on the ECG fluctuation index and motion state index; and determining whether the monitored object is at risk of cardiac arrhythmias based on the heart rhythm fluctuation index.

[0007] Furthermore, the specific formula for calculating the heart rate fluctuation index of the monitored subjects is as follows: Where XbD, DbD, and YdZ are the heart rate fluctuation index, electrocardiogram fluctuation index, and motion state index of the monitored object, respectively; e is a natural constant; and δ1 and δ2 are the electrocardiogram fluctuation sensitivity coefficient and motion state interference regulation coefficient stored in the database, respectively.

[0008] Furthermore, the ECG signal time series data includes ECG signal potential values ​​at several time points. The ECG feature extraction model includes an input layer, a time domain feature extraction layer, a waveform structure analysis layer, a spectral feature extraction layer, and a feature output layer. ECG features include heart rate interval fluctuations, the number of arrhythmia mutations, the proportion of QRS complex time span offsets, and the proportion of total energy outside the main frequency band.

[0009] Further, the specific steps for extracting the electrocardiogram (ECG) features of the monitored object are as follows: In the input layer of the ECG feature extraction model, preprocessed ECG signal time-series data are received; in the time-domain feature extraction layer of the ECG feature extraction model, time interval extraction and sequence trend judgment are performed on the ECG signal time-series data to obtain the heartbeat interval fluctuation value and the number of arrhythmia mutations of the monitored object; in the waveform structure analysis layer of the ECG feature extraction model, wave group region boundary identification and QRS time span analysis are performed on the ECG signal time-series data to obtain the proportion of QRS wave group time span offsets of the monitored object; in the spectral feature extraction layer of the ECG feature extraction model, frequency domain transformation and frequency band energy ratio analysis are performed on the ECG signal time-series data to obtain the proportion of total energy outside the main frequency band of the monitored object; in the feature output layer of the ECG feature extraction model, the heartbeat interval fluctuation value, the number of arrhythmia mutations, the proportion of QRS wave group time span offsets, and the proportion of total energy outside the main frequency band of the monitored object are output.

[0010] Further, the specific steps for analyzing the electrocardiogram fluctuation index of the monitored subjects are as follows: read the heartbeat interval fluctuation value of the monitored subjects and perform standardization processing; combine the standardized heartbeat interval fluctuation value of the monitored subjects with the corresponding number of arrhythmia mutations, the proportion of QRS complex time span offsets, and the proportion of total energy outside the main frequency band for comprehensive analysis to obtain the electrocardiogram fluctuation index of the monitored subjects.

[0011] Furthermore, the motion time series data includes triaxial acceleration values ​​at several time points, and the motion feature extraction model includes an input layer, an acceleration aggregation analysis layer, a time series mutation identification layer, a frequency domain perturbation extraction layer, and a feature output layer. The motion features include triaxial average acceleration values, the number of motion mutations, and the proportion of high-frequency energy in motion perturbations.

[0012] Further, the specific steps for extracting the electrocardiogram and motion features of the monitored object are as follows: In the input layer of the motion feature extraction model, preprocessed motion time-series data is received; in the acceleration aggregation analysis layer of the motion feature extraction model, the motion time-series data is weighted and synthesized on the coordinate axes and the mean is analyzed to obtain the triaxial average acceleration value of the monitored object; in the temporal mutation identification layer of the motion feature extraction model, the acceleration change amplitude is extracted and threshold is judged from the motion time-series data, and the number of mutations exceeding the set acceleration change threshold is counted to obtain the number of motion mutations of the monitored object; in the frequency domain perturbation extraction layer of the motion feature extraction model, the motion time-series data is subjected to fast Fourier transform, and the proportion of acceleration frequency band energy above the set frequency threshold is analyzed to obtain the proportion of high-frequency energy of motion perturbation of the monitored object; in the feature output layer of the motion feature extraction model, the triaxial average acceleration value, the number of motion mutations, and the proportion of high-frequency energy of motion perturbation of the monitored object are output.

[0013] Further, the specific steps for analyzing the motion state index of the monitored object are as follows: read the three-axis average acceleration value of the monitored object and perform standardization processing; combine the standardized three-axis average acceleration value of the monitored object with the corresponding number of motion mutations and the proportion of high-frequency energy of motion disturbances for comprehensive analysis to obtain the motion state index of the monitored object.

[0014] Furthermore, based on the heart rate fluctuation index, the specific steps for determining whether the monitored subject has a risk of arrhythmia are as follows: the heart rate fluctuation index is compared with the preset heart rate fluctuation arrhythmia range; and when the heart rate fluctuation index is within the preset heart rate fluctuation arrhythmia range, the monitored subject is considered to have a risk of arrhythmia.

[0015] A cardiac arrhythmia monitoring system includes: a data acquisition and processing unit, used to acquire and preprocess the electrocardiogram (ECG) signal time-series data and motion time-series data of the monitored object based on a set time period during the object's movement; a feature extraction unit, used to input the preprocessed ECG signal time-series data and motion time-series data into a pre-trained ECG feature extraction model and a motion feature extraction model, respectively, to extract the ECG features and motion features of the monitored object; a feature analysis unit, used to analyze the ECG fluctuation index and motion state index of the monitored object based on the ECG features and motion features; a cardiac rhythm fluctuation analysis unit, used to analyze the cardiac rhythm fluctuation index of the monitored object based on the ECG fluctuation index and the motion state index; and a cardiac arrhythmia determination unit, used to determine whether the monitored object has a risk of cardiac arrhythmia based on the cardiac rhythm fluctuation index.

[0016] The present invention has the following beneficial effects:

[0017] (1) This arrhythmia monitoring method can accurately identify motion interference and real arrhythmia in the electrocardiogram signal by simultaneously collecting electrocardiogram signals and motion time series data when the monitored object is moving, and combining the feature analysis of the two. By combining the analysis of electrocardiogram features and motion features, it can effectively distinguish the difference between electrocardiogram fluctuations caused by motion and arrhythmia, thereby avoiding the influence of artifacts on the identification of arrhythmia risk and ensuring that the monitoring results are more accurate and reliable in dynamic motion environment. This method can still maintain high accuracy under various motion states and effectively solves the problem of motion artifact interference in the prior art.

[0018] (2) This arrhythmia monitoring method, by introducing a comprehensive analysis of electrocardiogram characteristics and motion characteristics, makes arrhythmia monitoring not only limited to static conditions, but also able to provide personalized assessments based on the motion state, health level and electrocardiogram status of the monitored object. This multi-dimensional feature fusion method enables the monitoring to flexibly respond to different individuals and situations, accurately reflect the cardiac rhythm health status of each monitored object, and in particular, can identify the interference of motion state on electrocardiogram signals, reduce the false alarm rate, and provide patients with more accurate risk assessment and health monitoring services.

[0019] (3) This arrhythmia monitoring method, by combining motion time-series data and electrocardiogram signal time-series data in real time, can achieve dynamic real-time monitoring and real-time feedback. When the monitored object is exercising or active, it can more quickly identify potential arrhythmia abnormalities. In particular, when the patient is exercising, it can promptly detect electrocardiogram fluctuations caused by exercise or external interference, thereby improving the early warning capability of arrhythmia. Through the dynamic calculation and fusion of electrocardiogram fluctuation index and motion state index, this method can not only achieve accurate monitoring of electrocardiogram signals during exercise, but also provide real-time arrhythmia risk assessment, significantly improving the sensitivity and response speed of the monitoring system, and providing patients with more comprehensive health protection.

[0020] (4) The arrhythmia monitoring system, through the combination of data acquisition and processing unit, feature extraction unit and feature analysis unit, enables the system to acquire and process electrocardiogram signals and motion data in real time, thereby accurately extracting the electrocardiogram and motion characteristics of the monitored object. The arrhythmia fluctuation analysis unit performs in-depth analysis based on the extracted features to generate electrocardiogram fluctuation index and motion state index, providing data support for arrhythmia judgment. Finally, the arrhythmia judgment unit makes a comprehensive judgment based on these indicators, timely identifying potential arrhythmia risks. The linkage of the entire system improves the monitoring accuracy and real-time performance, effectively ensuring early warning and prevention of arrhythmia risks.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a flowchart of a cardiac arrhythmia monitoring method according to the present invention.

[0023] Figure 2 This is a flowchart illustrating the specific steps involved in extracting electrocardiogram (ECG) features from a monitored object using a method for monitoring arrhythmias according to the present invention.

[0024] Figure 3 This is a block diagram of a cardiac arrhythmia monitoring system according to the present invention. Detailed Implementation

[0025] Please see Figure 1This invention provides a technical solution: a method for monitoring cardiac arrhythmias, comprising the following steps: When the monitored object is moving, based on a set time period, acquire the electrocardiogram (ECG) signal time-series data and motion time-series data of the monitored object, and perform preprocessing (identifying and removing outliers in the ECG signal time-series data and motion time-series data; outliers include abrupt changes or missing points caused by signal loss, extreme motion impact, or electrode loosening; perform data completion processing on the time-series data with gaps after removing outliers; data completion processing includes filling in missing time points using adjacent valid point interpolation or a sliding window averaging strategy); input the preprocessed ECG signal time-series data and motion time-series data into a pre-trained ECG feature extraction model and motion feature extraction model, respectively, to extract the ECG features and motion features of the monitored object; based on the ECG features and motion features, analyze the ECG fluctuation index and motion state index of the monitored object, respectively; based on the ECG fluctuation index and motion state index, analyze the heart rhythm fluctuation index of the monitored object; based on the heart rhythm fluctuation index, determine whether the monitored object has a risk of cardiac arrhythmias.

[0026] The specific formula for calculating the heart rate fluctuation index of the monitored subjects is as follows: Wherein, XbD, DbD, and YdZ are the heart rate fluctuation index, electrocardiogram fluctuation index, and motion state index of the monitored object, respectively; e is a natural constant, which is 2.718 in this embodiment; δ1 and δ2 are the electrocardiogram fluctuation sensitivity coefficient and motion state interference adjustment coefficient stored in the database, respectively, and are 1.2 and 0.7 in this embodiment.

[0027] Specifically, the ECG signal time series data includes ECG signal potential values ​​at several time points. The ECG feature extraction model includes an input layer, a time domain feature extraction layer, a waveform structure analysis layer, a spectral feature extraction layer, and a feature output layer. ECG features include heart rate interval fluctuations, the number of arrhythmia mutations, the proportion of QRS complex time span offsets, and the proportion of total energy outside the main frequency band.

[0028] Among them, the electrocardiogram (ECG) signal potential value refers to the voltage change value of electrophysiological activity generated at a specific lead position on the body surface of the monitored object within a set time period. Its unit is millivolt (mV), which is used to reflect the time-varying potential fluctuations formed by the depolarization and repolarization process of myocardial cells. It can be acquired by an ECG electrode sensor. The ECG electrode sensor can be a silver / silver chloride (Ag / AgCl) electrode or a dry flexible electrode array, and is used in conjunction with a low-noise amplifier and an analog-to-digital converter to construct an ECG signal acquisition unit to realize the measurement of the ECG activity of the monitored object.

[0029] The pre-training steps for the ECG feature extraction model are as follows:

[0030] First, a training sample set was constructed based on a multi-center ECG dataset. The training data included several sets of raw ECG signal sequences with rhythm annotations, and each set of ECG signal time-series data included potential value sequences at several consecutive time points, along with manually annotated heartbeat interval information, QRS complex boundary information, and spectral energy labels. The training samples covered various typical heart rhythm states (such as sinus rhythm, atrial fibrillation, premature ventricular contractions, etc.) to enhance the model's ability to recognize diverse ECG structures.

[0031] In the ECG feature extraction model, an input layer, a temporal feature extraction layer, a waveform structure analysis layer, a spectral feature extraction layer, and a feature output layer are constructed sequentially, and a layered supervised training strategy is used for layer-by-layer parameter optimization. Specifically, the temporal feature extraction layer extracts rhythm variation and abrupt fluctuation features through convolutional kernels; the waveform structure analysis layer focuses on QRS complex segment localization and structural stability change identification; the spectral feature extraction layer constructs a frequency domain response vector through Fourier transform and energy statistics; and finally, the feature output layer completes multi-dimensional modeling of target features such as inter-beat fluctuation values, number of arrhythmia abrupt changes, proportion of QRS complex time span shifts, and proportion of total energy outside the main frequency band.

[0032] During training, to improve the model's sensitivity to abnormal fluctuations and its ability to identify microstructural perturbations in ECG signals, a joint loss function was constructed, including a loss for minimizing intercardiac interval errors, a waveform boundary misjudgment penalty term, and a spectral offset balancing term. The model used the Adam optimizer for iterative parameter updates and cross-validation to ensure good generalization ability between the training and validation sets. Finally, it converged within a preset number of iterations and output the weight parameter set of the trained ECG feature extraction model.

[0033] like Figure 2 As shown, the specific steps for extracting the electrocardiogram (ECG) characteristics of the monitored subjects are as follows:

[0034] In the input layer of the ECG feature extraction model, preprocessed ECG signal time-series data is received;

[0035] In the time-domain feature extraction layer of the ECG feature extraction model, time interval extraction and sequence trend judgment are performed on the ECG signal time series data to obtain the heartbeat interval fluctuation value and the number of arrhythmia mutations of the monitored object. Specifically, each R wave peak is identified by threshold detection and zero-crossing method, and the RR interval sequence between adjacent R waves is extracted; the standard deviation of the RR interval sequence is calculated to obtain the heartbeat interval fluctuation value; further, the difference between adjacent RR intervals is analyzed by first-order difference analysis, and the number of times the difference change amplitude is greater than the set jump threshold is counted as the number of arrhythmia mutations.

[0036] In the waveform structure analysis layer of the ECG feature extraction model, wave group region boundary identification and QRS time span analysis are performed on the ECG signal time series data to obtain the proportion of QRS wave group time span deviations of the monitored object. Specifically, the continuous heartbeat cycle is segmented, the start and end points of the QRS wave group in each cycle are identified, the time span is measured, and it is compared with the upper and lower limits of the set normal range. The ratio between the number of times the range is exceeded and the total number of heartbeats is calculated to obtain the proportion of QRS wave group time span deviations.

[0037] In the spectral feature extraction layer of the ECG feature extraction model, frequency domain transformation and frequency band energy ratio analysis are performed on the ECG signal time series data to obtain the total energy ratio outside the main frequency band of the monitored object. Specifically, the original ECG signal sequence is subjected to fast Fourier transform to obtain the spectral energy distribution in the entire frequency range. The main frequency band range is defined as 0.5Hz to 40Hz. The ratio of the sum of the energies of all frequencies outside this frequency band to the total energy is calculated as the total energy ratio outside the main frequency band.

[0038] In the feature output layer of the ECG feature extraction model, the output includes the heart rate fluctuation value, the number of arrhythmia mutations, the proportion of QRS complex time span shifts, and the proportion of total energy outside the main frequency band for the monitored object.

[0039] In this implementation scheme, high-quality electrocardiogram (ECG) signal acquisition technology ensures data accuracy. Using silver / silver chloride electrodes or flexible electrode arrays, stable acquisition of surface potential changes is achieved. The ECG feature extraction model can extract key ECG features from the signal, including inter-beat fluctuations, the number of arrhythmia mutations, and QRS complex time span shifts, helping to better understand changes in the heart. By analyzing these features, the system can monitor and identify potential arrhythmia risks in real time. In addition, model training improves its ability to recognize different ECG signal patterns by processing various heart rhythm states, making monitoring more comprehensive and accurate. This method not only improves the accuracy of arrhythmia identification but also provides real-time monitoring data, providing strong support for the prevention and early intervention of heart health.

[0040] Specifically, the steps for analyzing the electrocardiogram fluctuation index of the monitored subjects are as follows: read the heartbeat interval fluctuation value of the monitored subjects and perform standardization processing (i.e., de-uniting); combine the standardized heartbeat interval fluctuation value of the monitored subjects with the corresponding number of arrhythmia mutations, the proportion of QRS complex time span offsets, and the proportion of total energy outside the main frequency band for comprehensive analysis to obtain the electrocardiogram fluctuation index of the monitored subjects.

[0041] The specific formula for calculating the electrocardiogram fluctuation index of the monitored subjects is as follows: Wherein, DbD, XtB, XsC, PzB, and ZpB are the electrocardiogram fluctuation index, the number of arrhythmia mutations, the proportion of QRS complex time span offsets, and the proportion of total energy outside the main frequency band of the monitored object, respectively. λ1, λ2, and λ3 are the arrhythmia mutation number adjustment coefficient, the complex structure offset adjustment coefficient, and the high-frequency energy abnormality adjustment coefficient stored in the database, respectively, and in this embodiment, they are taken as 2.0, 0.5, and 1.7, respectively.

[0042] In this implementation plan, by combining multiple ECG characteristics and performing standardized processing, interference between different units can be effectively eliminated, ensuring data consistency and accuracy. Simultaneously, the calculation of the ECG fluctuation index comprehensively considers multiple characteristics such as inter-beat fluctuations, the number of arrhythmia mutations, QRS complex time span shifts, and the proportion of energy outside the dominant frequency band, comprehensively reflecting the ECG state of the monitored object. This comprehensive analysis method not only accurately assesses abnormal fluctuations in the ECG signal but also allows for flexible adjustment of different characteristics through the introduction of adjustment coefficients, thereby improving the sensitivity and accuracy of the ECG fluctuation index. In particular, the setting of adjustment coefficients makes the system more adaptable, enabling dynamic adjustment of detection standards according to different monitored objects or environmental conditions, thus enhancing the early identification capability of arrhythmias.

[0043] Specifically, the motion time series data includes triaxial acceleration values ​​at several time points, and the motion feature extraction model includes an input layer, an acceleration aggregation analysis layer, a time series mutation identification layer, a frequency domain perturbation extraction layer, and a feature output layer. The motion features include the triaxial average acceleration value, the number of motion mutations, and the proportion of high-frequency energy in motion perturbations.

[0044] Among them, the triaxial acceleration value refers to the rate of change of velocity of the monitored object in three spatial directions (front-back X-axis, left-right Y-axis, and up-down Z-axis) per unit time within a set time period, and its unit is meters per square second (m / s²). 2 This device is used to characterize the motion intensity, directional changes, and disturbance fluctuations of the monitored object. It can be acquired in real time by a triaxial accelerometer, which can be a MEMS microelectromechanical accelerometer chip (such as ADXL345, MPU6050, etc.). It is fixed to the chest or waist of the monitored object and combined with a high-precision clock synchronization module and an anti-noise filtering module to complete the acquisition of continuous acceleration data.

[0045] The pre-training steps for the motion feature extraction model are as follows:

[0046] Before model pre-training, a training sample set is constructed based on motion data collected from multiple real-world scenarios. The training data consists of time series sequences of three-axis acceleration values ​​for several consecutive time periods. The data scenarios cover various states such as walking, running, violent disturbances, stillness, and sudden movements. The data is also equipped with manually labeled average intensity, mutation point labels, and disturbance frequency band energy labels to supervise the accuracy of each motion feature value extraction.

[0047] The motion feature extraction model comprises an input layer, an acceleration aggregation and analysis layer, a temporal abrupt change identification layer, a frequency domain perturbation extraction layer, and a feature output layer. The input layer receives the original triaxial acceleration sequence. The acceleration aggregation and analysis layer extracts acceleration intensity features using a triaxial squared weighted superposition and moving average method. The temporal abrupt change identification layer uses gradient detection and threshold judgment algorithms to extract points of drastic acceleration changes and counts the number of abrupt changes. The frequency domain perturbation extraction layer calculates high-frequency energy based on Fast Fourier Transform (FFT) and counts the proportion of energy in frequency bands above a set frequency (e.g., 10Hz). Finally, the feature output layer outputs three motion features: the triaxial average acceleration value, the number of motion abrupt changes, and the proportion of high-frequency energy in the motion perturbation.

[0048] To improve the model's sensitivity to sudden disturbances and its ability to discriminate high-frequency noise energy, a joint loss function was constructed, including an acceleration mean error term, a mutation detection accuracy loss term, and a frequency domain energy ratio deviation term. The parameters were trained using backpropagation. During the training process, strategies such as group sampling, noise enhancement, and temporal smoothing were employed to enhance the model's stability and generalization ability. Finally, a stable motion feature extraction model that is adaptable to motion state analysis in multiple scenarios was obtained.

[0049] The specific steps for extracting the electrocardiogram and motion characteristics of the monitored subjects are as follows:

[0050] In the input layer of the motion feature extraction model, preprocessed motion time-series data is received;

[0051] In the acceleration aggregation analysis layer of the motion feature extraction model, the motion time series data is weighted and analyzed by coordinate axes to obtain the triaxial average acceleration value of the monitored object. Specifically, the square root of the sum of the squares of the triaxial accelerations at each time point is used to calculate the resultant acceleration value. After obtaining the resultant acceleration sequence of all time points, its arithmetic mean is calculated as the triaxial average acceleration value, which is used to reflect the overall motion intensity level within the set period.

[0052] In the temporal mutation identification layer of the motion feature extraction model, the acceleration change amplitude is extracted and threshold judgment is performed on the motion time series data. The number of mutations exceeding the set acceleration change threshold is counted to obtain the number of motion mutations of the monitored object. Specifically, the first-order difference calculation is performed on the resultant acceleration sequence to obtain the acceleration change value at each adjacent time point. A mutation threshold (e.g., 0.8g) is set, and the number of changes greater than the threshold is counted as the number of motion mutations, which is used to identify unstable behaviors such as motion interruption and jumping.

[0053] In the frequency domain perturbation extraction layer of the motion feature extraction model, a fast Fourier transform is performed on the motion time series data, and the proportion of acceleration frequency band energy above the set frequency threshold is analyzed to obtain the high-frequency energy proportion of motion perturbation of the monitored object. Specifically, the resultant acceleration sequence is subjected to Fourier transform to obtain the spectral distribution of acceleration, a high-frequency perturbation interval (e.g., ≥5Hz) is set, and the proportion of energy in the high-frequency interval to the total energy is statistically analyzed as the high-frequency energy proportion of motion perturbation, which is used to reflect the intensity of severe or vibration-type motion interference.

[0054] In the feature output layer of the motion feature extraction model, the three-axis average acceleration value, the number of motion mutations, and the proportion of high-frequency energy of motion disturbances of the monitored object are output.

[0055] In this implementation scheme, a motion feature extraction model is used to combine the motion data of the monitored object with electrocardiogram (ECG) signal data, significantly improving the accuracy of arrhythmia monitoring under dynamic conditions. First, the model collects motion time-series data in real time using a triaxial accelerometer and extracts acceleration intensity features in the acceleration aggregation analysis layer, calculating the triaxial average acceleration value. This process accurately reflects the overall motion intensity of the monitored object within a set period, ensuring the system's sensitivity to different motion states (such as walking and running). Second, the time-series abrupt change recognition layer can identify drastic motion changes, analyzing the drastic acceleration through gradient detection and threshold judgment. The system analyzes and counts the number of motion mutations to accurately identify potential motion interruptions or abnormal fluctuations during motion. Finally, the frequency domain perturbation extraction layer uses Fast Fourier Transform (FFT) to analyze high-frequency energy and calculates the proportion of high-frequency energy of motion perturbations, reflecting the potential interference of vibrational or strenuous exercise on the electrocardiogram (ECG) signal. Through the extraction of this series of features, the present invention can accurately distinguish the relationship between ECG signals and motion interference, reduce the impact of motion-generated artifacts on arrhythmia monitoring results, and enhance the system's real-time monitoring capability of heart rhythm fluctuations under dynamic motion conditions, effectively improving the accuracy of motion interference suppression and early arrhythmia identification.

[0056] Specifically, the steps for analyzing the motion state index of the monitored object are as follows: read the three-axis average acceleration value of the monitored object and perform standardization processing (i.e., unit removal); combine the standardized three-axis average acceleration value of the monitored object with the corresponding number of motion mutations and the proportion of high-frequency energy of motion disturbances for comprehensive analysis to obtain the motion state index of the monitored object.

[0057] The specific formula for calculating the motion state index of the monitored object is as follows: Wherein, YdZ, SpJ, YtB, and YrZ are the motion state index, three-axis average acceleration value, number of motion mutations, and proportion of high-frequency energy of motion disturbance of the monitored object, respectively; e is a natural constant, which is 2.718 in this embodiment; and μ1, μ2, and μ3 are the motion intensity adjustment coefficient, motion mutation interference adjustment coefficient, and high-frequency disturbance amplification adjustment coefficient stored in the database, respectively, and are 0.8, 2.3, and 1.2 in this embodiment, respectively.

[0058] In this implementation plan, by comprehensively analyzing the triaxial average acceleration value, the number of motion mutations, and the proportion of high-frequency energy of motion disturbances, the motion state index of the monitored object is accurately calculated, thereby improving the accuracy of arrhythmia monitoring in dynamic environments. First, the triaxial average acceleration value is standardized to remove the influence of units, making the measurement of exercise intensity have a uniform scale and ensuring the comparability of data between different individuals. Second, the number of motion mutations reflects sudden changes during exercise, which can effectively identify violent fluctuations or unstable behaviors during exercise and reduce misjudgments caused by motion artifacts. The proportion of high-frequency energy of motion disturbances is analyzed by Fourier transform to determine the intensity of high-frequency interference during exercise, accurately quantifying the potential impact of vigorous exercise on electrocardiogram signals. By introducing adjustment coefficients, the system can dynamically adjust the weights of each feature according to different exercise states, ensuring personalized and accurate monitoring. The application of this method effectively improves the monitoring accuracy under exercise conditions and reduces erroneous judgments caused by motion interference.

[0059] Specifically, based on the heart rate fluctuation index, the specific steps to determine whether the monitored subject has a risk of arrhythmia are as follows: the heart rate fluctuation index is compared with the preset heart rate fluctuation arrhythmia range; and when the heart rate fluctuation index is within the preset heart rate fluctuation arrhythmia range, the monitored subject is considered to have a risk of arrhythmia.

[0060] In this implementation plan, by analyzing the heart rhythm fluctuation index and the preset heart rhythm fluctuation arrhythmia range, the electrocardiogram fluctuation of the monitored subject can be effectively compared with known health standards to automatically determine whether there is a risk of arrhythmia. By setting the heart rhythm fluctuation arrhythmia range, the normal range of heart rhythm fluctuation can be defined and the possibility of arrhythmia can be quantified. This method does not rely on the traditional single judgment standard, but combines the heart rhythm fluctuation index with the range comparison, which increases the sensitivity and accuracy of monitoring. By judging whether the heart rhythm fluctuation exceeds the arrhythmia range in real time, potential arrhythmia risks can be identified in a timely manner, reducing the possibility of misdiagnosis and missed diagnosis.

[0061] Please see Figure 3 This invention provides a technical solution: a cardiac arrhythmia monitoring system, comprising: a data acquisition and processing unit, used to acquire, based on a set time period, the time-series data of the electrocardiogram (ECG) signal and the time-series data of the monitored object during movement, and perform preprocessing; a feature extraction unit, used to input the preprocessed time-series data of the ECG signal and the time-series data of the movement into a pre-trained ECG feature extraction model and a movement feature extraction model, respectively, to extract the ECG features and movement features of the monitored object; a feature analysis unit, used to analyze the ECG fluctuation index and the movement state index of the monitored object based on the ECG features and the movement features; a heart rhythm fluctuation analysis unit, used to analyze the heart rhythm fluctuation index of the monitored object based on the ECG fluctuation index and the movement state index; and a cardiac arrhythmia determination unit, used to determine whether the monitored object has a risk of cardiac arrhythmia based on the heart rhythm fluctuation index.

[0062] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0063] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for monitoring cardiac arrhythmias, characterized in that, Includes the following steps: When the monitored object is moving, based on a set time period, the monitoring object's electrocardiogram signal time series data and motion time series data are acquired and preprocessed. The preprocessed ECG signal time series data and motion time series data are input into the pre-trained ECG feature extraction model and motion feature extraction model, respectively, to extract the ECG features and motion features of the monitored object; Based on electrocardiogram (ECG) characteristics and motion characteristics, the ECG fluctuation index and motion state index of the monitored subjects were analyzed, respectively. Based on the electrocardiogram fluctuation index and the exercise state index, the heart rate fluctuation index of the monitored subjects was analyzed; Based on the heart rate fluctuation index, it is determined whether the monitored subject is at risk of arrhythmia.

2. The arrhythmia monitoring method according to claim 1, characterized in that, The specific formula for calculating the heart rate fluctuation index of the monitored subjects is as follows: Where XbD, DbD, and YdZ are the heart rate fluctuation index, electrocardiogram fluctuation index, and motion state index of the monitored object, respectively; e is a natural constant; and δ1 and δ2 are the electrocardiogram fluctuation sensitivity coefficient and motion state interference regulation coefficient stored in the database, respectively.

3. The method for monitoring cardiac arrhythmias according to claim 1, characterized in that, ECG signal time series data includes ECG signal potential values ​​at several time points. The ECG feature extraction model includes an input layer, a time domain feature extraction layer, a waveform structure analysis layer, a spectral feature extraction layer, and a feature output layer. ECG features include heart rate interval fluctuations, the number of arrhythmia mutations, the percentage of QRS complex time span offsets, and the percentage of total energy outside the main frequency band.

4. The arrhythmia monitoring method according to claim 3, characterized in that, The specific steps for extracting the electrocardiogram characteristics of the monitored subjects are as follows: In the input layer of the ECG feature extraction model, preprocessed ECG signal time-series data is received; In the time-domain feature extraction layer of the ECG feature extraction model, time interval extraction and sequence trend judgment are performed on the time series data of ECG signals to obtain the heartbeat interval fluctuation value and the number of arrhythmia mutations of the monitored object. In the waveform structure analysis layer of the ECG feature extraction model, wave group region boundary identification and QRS time span analysis are performed on the ECG signal time series data to obtain the proportion of QRS wave group time span offset times of the monitored object. In the spectral feature extraction layer of the ECG feature extraction model, frequency domain transformation and frequency band energy ratio analysis are performed on the ECG signal time series data to obtain the total energy ratio outside the main frequency band of the monitored object; In the feature output layer of the ECG feature extraction model, the output includes the heart rate fluctuation value, the number of arrhythmia mutations, the proportion of QRS complex time span shifts, and the proportion of total energy outside the main frequency band for the monitored object.

5. The method for monitoring cardiac arrhythmias according to claim 3, characterized in that, The specific steps for analyzing the electrocardiogram fluctuation index of the monitored subjects are as follows: Read the heart rate fluctuation values ​​of the monitored object and perform standardization processing; The standardized cardiac interval fluctuation values ​​of the monitored subjects are combined with the corresponding number of arrhythmia mutations, the proportion of QRS complex time span shifts, and the proportion of total energy outside the main frequency band for comprehensive analysis to obtain the electrocardiogram fluctuation index of the monitored subjects.

6. The method for monitoring cardiac arrhythmias according to claim 1, characterized in that, The motion time series data includes triaxial acceleration values ​​at several time points. The motion feature extraction model includes an input layer, an acceleration aggregation analysis layer, a time series mutation identification layer, a frequency domain perturbation extraction layer, and a feature output layer. The motion features include triaxial average acceleration values, the number of motion mutations, and the proportion of high-frequency energy in motion perturbations.

7. The method for monitoring cardiac arrhythmias according to claim 6, characterized in that, The specific steps for extracting the electrocardiogram and motion characteristics of the monitored subjects are as follows: In the input layer of the motion feature extraction model, preprocessed motion time-series data is received; In the acceleration aggregation analysis layer of the motion feature extraction model, the motion time series data is weighted and synthesized on the coordinate axes and the mean analysis is performed to obtain the three-axis average acceleration value of the monitored object. In the temporal mutation identification layer of the motion feature extraction model, the acceleration change amplitude is extracted and threshold judgment is performed on the motion temporal data, and the number of mutations exceeding the set acceleration change threshold is counted to obtain the number of motion mutations of the monitored object. In the frequency domain perturbation extraction layer of the motion feature extraction model, a fast Fourier transform is performed on the motion time series data, and the proportion of acceleration frequency band energy above the set frequency threshold is analyzed to obtain the proportion of high-frequency energy of motion perturbation of the monitored object. In the feature output layer of the motion feature extraction model, the three-axis average acceleration value, the number of motion mutations, and the proportion of high-frequency energy of motion disturbances of the monitored object are output.

8. The method for monitoring cardiac arrhythmias according to claim 6, characterized in that, The specific steps for analyzing the motion state index of the monitored object are as follows: Read the three-axis average acceleration values ​​of the monitored object and perform standardization processing; The motion state index of the monitored object is obtained by comprehensively analyzing the standardized triaxial average acceleration value of the monitored object, combined with the corresponding number of motion mutations and the proportion of high-frequency energy of motion disturbances.

9. The method for monitoring cardiac arrhythmias according to claim 1, characterized in that, The specific steps for determining whether a monitored individual is at risk of arrhythmia based on the heart rate fluctuation index are as follows: The heart rhythm fluctuation index is compared with the preset heart rhythm fluctuation and arrhythmia interval for analysis. When the heart rate fluctuation index is within the preset heart rate fluctuation and arrhythmia range, the monitored subject is considered to have a heart rate arrhythmia risk.

10. A cardiac arrhythmia monitoring system, employing the cardiac arrhythmia monitoring method according to any one of claims 1-9, characterized in that, include: The data acquisition and processing unit is used to acquire the electrocardiogram signal time series data and motion time series data of the monitored object based on a set time period when the monitored object is moving, and to perform preprocessing. The feature extraction unit is used to input the preprocessed ECG signal time series data and motion time series data into the pre-trained ECG feature extraction model and motion feature extraction model, respectively, to extract the ECG features and motion features of the monitored object; The feature analysis unit is used to analyze the electrocardiogram fluctuation index and motion state index of the monitored object based on electrocardiogram characteristics and motion characteristics, respectively. The heart rhythm fluctuation analysis unit is used to analyze the heart rhythm fluctuation index of the monitored object based on the electrocardiogram fluctuation index and the exercise state index; The arrhythmia determination unit is used to determine whether the monitored subject is at risk of arrhythmia based on the heart rhythm fluctuation index.

Images