Intelligent analysis method and system for infant electrocardiosignal based on conductive fabric

Abstract

The application provides a kind of infantile electrocardiosignal intelligent analysis method and system based on conductive fabric, it is related to electrocardiosignal processing technical field, including: through flexible conductive fabric, multi-channel electrocardio data is collected, space correlation matrix is constructed to select optimal reference channel to obtain complete electrocardio data.Waveform reconstruction and feature extraction are carried out on complete electrocardio data, and physiological state feature vector is constructed.By querying heart rate baseline reference interval, the feature vector is calibrated and corrected, the heart rhythm type is determined by combining the interval stability and autonomic nervous state, and the analysis result is generated.The application realizes the stable collection and intelligent analysis of infantile electrocardiosignal, improves the analysis accuracy and reliability.

Description

Technical Field

[0001] This invention relates to the field of electrocardiogram (ECG) signal processing technology, and in particular to an intelligent method and system for analyzing ECG signals of young children based on conductive fabric. Background Technology

[0002] In existing technologies, using flexible conductive fabrics to collect electrocardiogram (ECG) signals from young children has become a routine method. Fabric electrodes, due to their softness and conformability, are more suitable for children's skin and frequent activities compared to traditional rigid electrodes. After acquiring multi-channel ECG signals using fabric electrodes, existing technologies typically select the single channel with the largest signal amplitude or the least noise for subsequent analysis, or perform simple averaging of signals from multiple channels to attempt to improve the signal-to-noise ratio. Conventional analysis processes rely on fixed thresholds or template matching algorithms to detect key waveform feature points, thereby calculating basic parameters such as heart rate and heart rate variability to assess the child's physiological state.

[0003] However, existing technologies still have limitations. The signals acquired by fabric electrodes are highly susceptible to motion artifacts, electromyography interference, and contact noise, making it difficult to effectively address the inconsistency in the quality of multi-channel signals. This may lead to the loss of key physiological information or the amplification of noise. Waveform detection methods that rely on fixed thresholds or universal templates are not well adapted to the characteristics of large individual differences in the morphology of children's electrocardiogram signals and complex noise backgrounds, which can easily lead to false detection or missed detection of feature points. Furthermore, directly using a wide range of normal values ​​for age groups to determine the state ignores the differences in the individual physiological baseline of children, which may lead to insufficient sensitivity to subtle changes in physiological state or false alarms. Summary of the Invention

[0004] This invention provides a method and system for intelligent analysis of electrocardiogram signals in young children based on conductive fabric, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides a method for intelligent analysis of electrocardiogram (ECG) signals in young children based on conductive fabric, comprising: Multi-channel electrocardiogram (ECG) data of young children are collected by flexible conductive fabric, and the cross-correlation coefficients and phase differences between different channels are calculated to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix is ​​decomposed by eigenvalue and the signal quality index of each channel is calculated. The channel with the highest signal quality index is used as the reference channel for amplitude scaling and phase shifting to obtain complete ECG data. The complete electrocardiogram (ECG) data is decomposed by wavelet packets and the ECG waveform is reconstructed. Based on the reconstructed ECG waveform, the peak position of the ventricular depolarization wave is detected and the continuous interval sequence is calculated. The boundary of the atrial depolarization wave is identified and the amplitude integral is calculated to obtain the atrial parameters. The time span from ventricular depolarization to repolarization is measured to obtain the ventricular parameters. The time-domain statistics and frequency-domain power index corresponding to the continuous interval sequence are calculated and the physiological state feature vector is constructed by combining the atrial and ventricular parameters. Query the current child's corresponding heart rate baseline reference interval and calculate the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, calculate the calibrated heart rate baseline and correct the physiological state feature vector to obtain the calibrated feature vector. The adjacent differences of the continuous interval sequence are calculated and the interval stability is determined. The power ratio in the calibration feature vector is extracted to determine the autonomic nervous state. The heart rhythm type is determined by combining the interval stability and the analysis results are generated.

[0006] In one alternative implementation, Multi-channel electrocardiogram (ECG) data of young children were collected using flexible conductive fabric, and the cross-correlation coefficients and phase differences between different channels were calculated to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix was then subjected to eigenvalue decomposition, and the signal quality index of each channel was calculated, including: Raw electrocardiogram (ECG) signals are collected by contacting the skin of the infant's chest with a flexible conductive fabric electrode unit. The raw ECG signals are then converted from analog to digital to obtain digital ECG signals. The digital ECG signals are stored according to channel numbers and time-aligned to obtain multi-channel ECG data. A sliding window cross-correlation calculation is performed between any two channels in the multi-channel electrocardiogram data to obtain a cross-correlation function, and the peak value of the cross-correlation function is extracted as the cross-correlation coefficient. The phase difference is obtained by calculating the time delay of the peak position of the cross-correlation function. The cross-correlation coefficient is used as the correlation strength in the spatial dimension, and the phase difference is used as the propagation delay in the time dimension to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix is ​​decomposed into eigenvalues ​​to obtain an eigenvalue sequence. The eigenvalue sequence is sorted in descending order of numerical value and the first few eigenvalues ​​are extracted as dominant eigenvalues. The ratio between the total energy of the dominant eigenvalues ​​and the total energy of the entire eigenvalue sequence is calculated to obtain the dominant energy percentage. The interference energy is obtained by calculating the sum of the energy of the last few eigenvalues ​​in the eigenvalue sequence. The interference energy ratio is obtained by calculating the ratio between the interference energy and the sum of the energy of the dominant eigenvalues. The signal quality index of each channel is obtained by weighting and combining the dominant energy ratio and the interference energy ratio.

[0007] In one alternative implementation, Using the channel with the highest signal quality index as the reference channel, amplitude scaling and phase shifting are performed to obtain complete ECG data, including: The channel with the highest signal quality index is selected as the reference channel. The electrode position coordinates corresponding to the reference channel are obtained as the reference position coordinates. Each channel is traversed and the channel with a signal quality index lower than a preset threshold is identified as the channel to be reconstructed. The electrode position coordinates corresponding to the channel to be reconstructed are obtained as the position coordinates to be reconstructed. The Euclidean distance between the reference position coordinates and the position coordinates to be reconstructed is calculated to obtain the electrode spacing. An attenuation function is established based on the electrode spacing, and the attenuation amplitude of the ECG signal in the reference channel is calculated to obtain the attenuation amplitude. The attenuation amplitude is used as the initial reconstruction amplitude of the channel to be reconstructed. The spatial angle between the reference position coordinates and the position coordinates to be reconstructed is calculated to obtain the electrode position angle. The propagation delay of the cardiac electric field is calculated by weighting the electrode position angle to obtain the propagation delay. The ECG signal of the reference channel is shifted along the time axis according to the propagation delay to obtain a phase shift signal. The preliminary reconstructed amplitude is combined with the phase shift signal to obtain a reconstructed ECG signal. The reconstructed ECG signal is used to replace the signal segments in the channel to be reconstructed that have a signal quality index lower than a preset threshold, and the signal segments in the channel to be reconstructed that have a signal quality index higher than the preset threshold are retained to obtain complete ECG data.

[0008] In one alternative implementation, The complete ECG data is subjected to wavelet packet decomposition and the ECG waveform is reconstructed. Based on the reconstructed ECG waveform, the peak position of the ventricular depolarization wave is detected and the continuous interval sequence is calculated. The boundary of the atrial depolarization wave is identified and the amplitude integral is calculated to obtain the atrial parameters, including: The complete ECG data is decomposed using wavelet packets, and the dominant frequency band is identified by calculating the energy distribution of each frequency band. Based on the decomposition coefficients corresponding to the dominant frequency band, wavelet packets are reconstructed to obtain the reconstructed ECG waveform. Time-frequency joint analysis is then performed to obtain the instantaneous frequency sequence and instantaneous amplitude sequence, and a time-frequency feature matrix is ​​constructed. The time-frequency feature matrix is ​​then decomposed using singular value decomposition to obtain the dominant singular values. Based on the dominant singular values, feature enhancement is performed on the reconstructed ECG waveform to obtain the enhanced ECG waveform. Morphological gradient operations are performed on the enhanced ECG waveform to obtain a waveform gradient sequence, and extreme point detection is used to determine the candidate peak positions. Local maximum verification is performed on the amplitude of the enhanced ECG waveform corresponding to the candidate peak positions, and the temporal width feature and frequency energy distribution feature of the ventricular depolarization wave are extracted and filtered to obtain the peak position of the ventricular depolarization wave. The peak positions of the ventricular depolarization wave are arranged in chronological order and differential operations are performed to obtain an instantaneous interval value sequence. Outlier detection and correction are performed on the instantaneous interval value sequence to obtain a continuous interval sequence. In the enhanced ECG waveform, a reverse time window is constructed with the peak position of the ventricular depolarization wave as the anchor point, and the zero point search of the second derivative is performed to obtain the starting boundary of the atrial depolarization wave. The waveform curvature change rate analysis is performed within the reverse time window to obtain the ending boundary of the atrial depolarization wave. The atrial depolarization wave amplitude integral is calculated by combining the starting boundary of the atrial depolarization wave as the atrial parameter.

[0009] In one alternative implementation, Ventricular parameters are obtained by measuring the time span from ventricular depolarization to repolarization. The time-domain statistics and frequency-domain power indices corresponding to the continuous interval sequence are calculated, and a physiological state feature vector is constructed by combining atrial and ventricular parameters, including: In the pre-acquired enhanced electrocardiogram waveform, the waveform amplitude baseline regression detection is performed starting from the peak position of the ventricular depolarization wave, and the ventricular repolarization endpoint is identified. The time difference between the peak position of the ventricular depolarization wave and the ventricular repolarization endpoint is calculated to obtain the ventricular parameters. The central tendency measure and dispersion measure are calculated for the continuous interval series to obtain the interval central location parameter and the interval dispersion parameter. The adjacent element change rate measure is calculated for the continuous interval series to obtain the interval short-range fluctuation parameter. The interval central location parameter and the interval dispersion parameter are combined and combined with the interval short-range fluctuation parameter to calculate the time domain statistics. The continuous interval sequence is transformed in the frequency domain to obtain a frequency domain representation sequence. Energy concentration is calculated in the low-frequency component range and the high-frequency component range to obtain a low-frequency energy concentration index and a high-frequency energy concentration index. Based on the relative relationship between the low-frequency energy concentration index and the high-frequency energy concentration index, an autonomic nervous system regulation balance index is calculated and the frequency domain power index is obtained. The time-domain statistics and the frequency-domain power index are obtained and concatenated with the atrial parameters and the ventricular parameters in a preset physiological correlation order to obtain the original feature sequence. The original feature sequence is then subjected to numerical scaling to obtain the physiological state feature vector.

[0010] In one alternative implementation, The system queries the current child's corresponding heart rate baseline reference interval and calculates the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, it calculates a calibrated heart rate baseline and corrects the physiological state feature vector to obtain the calibrated feature vector, which includes: Obtain the current child's age information in months and determine the current child's corresponding developmental stage classification. Based on the developmental stage classification, query the corresponding heart rate baseline reference interval from the preset physiological database. Extract the interval concentration location parameter from the physiological state feature vector and convert it into a heart rate value. Calculate the deviation distance between the heart rate value and the heart rate baseline reference interval and select the smallest non-negative value as the deviation amount. Calculate the width of the heart rate baseline reference interval and determine whether the deviation is greater than the width. If the deviation exceeds the width of the heart rate baseline reference interval, obtain the historical heart rate sequence corresponding to the current child, calculate the individualized heart rate benchmark value corresponding to the current child, and use it as the calibration heart rate baseline. The baseline offset between the calibrated heart rate baseline and the center value of the heart rate baseline reference interval is calculated. Correlation analysis is performed on each feature component in the physiological state feature vector to obtain the feature coupling relationship matrix and determine the correlation strength coefficient of each feature component with the interval concentration position parameter. The decomposed offset is obtained based on the correlation strength coefficient and the baseline offset. Based on the decomposed offset, each feature component in the physiological state feature vector is coordinatedly adjusted to obtain the calibrated feature components. The calibrated feature vector is obtained by combining the interval concentration position parameter.

[0011] In one alternative implementation, Calculating the adjacent differences of the continuous interval sequence and determining interval stability, extracting the power ratio from the calibration feature vector to determine the autonomic nervous state, and combining the interval stability to determine the heart rhythm type and generate analytical results include: The interval change sequence is obtained by calculating the point-to-point adjacent difference of the continuous interval sequence, and the interval fluctuation amplitude sequence is obtained by extracting the absolute value. The variance of the interval fluctuation amplitude sequence within a preset sliding window is calculated to obtain the local fluctuation intensity sequence, and the interval stability is obtained by performing global statistical analysis. The low-frequency energy concentration index and the high-frequency energy concentration index are extracted from the calibration feature vector and their ratio is calculated to obtain the autonomic nervous system regulation balance index as the power ratio. The power ratio is then mapped to the sympathetic nervous system sensitive area and the parasympathetic nervous system sensitive area to obtain the quantification value of the sympathetic nervous system dominance and the quantification value of the parasympathetic nervous system dominance, and the autonomic nervous system state is obtained by solving the problem. The temporal regularity of the interval stability is quantified to obtain the rhythmic characteristics of the heart rhythm; the dynamic regulation of the autonomic nervous state is quantified to obtain the regulatory characteristics of the heart rhythm; the rhythmic characteristics of the heart rhythm are subjected to a periodicity test to obtain the periodicity test result; the regulatory characteristics of the heart rhythm are subjected to a consistency test to obtain the consistency test result; based on the periodicity test result and the consistency test result, a co-discrimination is performed to obtain the heart rhythm type; the heart rhythm type is data encapsulated and combined with the interval stability and the autonomic nervous state fields to obtain the parsing result.

[0012] A second aspect of this invention provides an intelligent analysis system for electrocardiogram (ECG) signals in young children based on conductive fabric, comprising: The ECG acquisition unit is used to acquire multi-channel ECG data of children through flexible conductive fabric, calculate the cross-correlation coefficient and phase difference between different channels to construct a spatiotemporal correlation matrix, perform eigenvalue decomposition on the spatiotemporal correlation matrix and calculate the signal quality index of each channel, and use the channel with the highest signal quality index as the reference channel to perform amplitude scaling and phase shifting to obtain complete ECG data. The feature extraction unit is used to perform wavelet packet decomposition on the complete electrocardiogram data and reconstruct the electrocardiogram waveform. Based on the reconstructed electrocardiogram waveform, it detects the peak position of the ventricular depolarization wave and calculates the continuous interval sequence. It identifies the boundary of the atrial depolarization wave and calculates the amplitude integral to obtain the atrial parameters. It measures the time span from ventricular depolarization to repolarization to obtain the ventricular parameters. It calculates the time domain statistics and frequency domain power index corresponding to the continuous interval sequence and constructs a physiological state feature vector by combining the atrial and ventricular parameters. The heart rate calibration unit is used to query the heart rate baseline reference interval corresponding to the current child and calculate the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, the calibration heart rate baseline is calculated and the physiological state feature vector is corrected to obtain the calibration feature vector. The heart rhythm analysis unit is used to calculate the adjacent differences of the continuous interval sequence and determine the interval stability, extract the power ratio in the calibration feature vector to determine the autonomic nervous state, and combine the interval stability to determine the heart rhythm type and generate analysis results.

[0013] A third aspect of the present invention provides an electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] In this invention, a spatiotemporal correlation matrix is ​​constructed by calculating the cross-correlation coefficients and phase differences between multiple channels, effectively quantifying the spatiotemporal dependencies between channels. The optimal channel is selected as a reference for amplitude scaling and phase shifting, which can compensate for signal attenuation and distortion caused by fabric contact or movement, thereby synthesizing high-fidelity complete ECG data and laying the foundation for subsequent accurate analysis. Wavelet packet decomposition and reconstruction technology is used to effectively separate noise and baseline drift in ECG signals, highlight key waveforms such as ventricular depolarization waves, and measure the time span from ventricular depolarization to repolarization, enabling the assessment of the ventricular repolarization process. The physiological state feature vector constructed by integrating time-domain statistics, frequency-domain power indices, and atrial and ventricular parameters comprehensively characterizes the electrophysiological state of a child's heart. By querying and comparing individualized heart rate baseline reference intervals, the deviation of the current physiological characteristics from the normal range can be calculated. When the deviation exceeds a reasonable range, the heart rate baseline is calibrated and the feature vector is corrected. This effectively reduces misjudgments caused by transient physiological fluctuations or measurement errors, and improves the stability and personalization of state assessment. By calculating the adjacent differences of continuous interval sequences, the stability of the heartbeat interval can be quantitatively assessed, enabling intelligent discrimination of a child's heart rhythm type and autonomic nervous system state. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the intelligent analysis method for electrocardiogram signals of young children based on conductive fabric, as described in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the individualized calibration process for heart rate monitoring using an intelligent analysis method for electrocardiogram signals in young children based on conductive fabric, as described in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0019] Figure 1 This is a flowchart illustrating the intelligent analysis method for electrocardiogram signals in young children based on conductive fabric, as described in an embodiment of the present invention. Figure 1 As shown, the method includes: Multi-channel electrocardiogram (ECG) data of young children are collected by flexible conductive fabric, and the cross-correlation coefficients and phase differences between different channels are calculated to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix is ​​decomposed by eigenvalue and the signal quality index of each channel is calculated. The channel with the highest signal quality index is used as the reference channel for amplitude scaling and phase shifting to obtain complete ECG data. The complete electrocardiogram (ECG) data is decomposed by wavelet packets and the ECG waveform is reconstructed. Based on the reconstructed ECG waveform, the peak position of the ventricular depolarization wave is detected and the continuous interval sequence is calculated. The boundary of the atrial depolarization wave is identified and the amplitude integral is calculated to obtain the atrial parameters. The time span from ventricular depolarization to repolarization is measured to obtain the ventricular parameters. The time-domain statistics and frequency-domain power index corresponding to the continuous interval sequence are calculated and the physiological state feature vector is constructed by combining the atrial and ventricular parameters. Query the current child's corresponding heart rate baseline reference interval and calculate the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, calculate the calibrated heart rate baseline and correct the physiological state feature vector to obtain the calibrated feature vector. The adjacent differences of the continuous interval sequence are calculated and the interval stability is determined. The power ratio in the calibration feature vector is extracted to determine the autonomic nervous state. The heart rhythm type is determined by combining the interval stability and the analysis results are generated.

[0020] In one alternative implementation, Multi-channel electrocardiogram (ECG) data of young children were collected using flexible conductive fabric, and the cross-correlation coefficients and phase differences between different channels were calculated to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix was then subjected to eigenvalue decomposition, and the signal quality index of each channel was calculated, including: Raw electrocardiogram (ECG) signals are collected by contacting the skin of the infant's chest with a flexible conductive fabric electrode unit. The raw ECG signals are then converted from analog to digital to obtain digital ECG signals. The digital ECG signals are stored according to channel numbers and time-aligned to obtain multi-channel ECG data. A sliding window cross-correlation calculation is performed between any two channels in the multi-channel electrocardiogram data to obtain a cross-correlation function, and the peak value of the cross-correlation function is extracted as the cross-correlation coefficient. The phase difference is obtained by calculating the time delay of the peak position of the cross-correlation function. The cross-correlation coefficient is used as the correlation strength in the spatial dimension, and the phase difference is used as the propagation delay in the time dimension to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix is ​​decomposed into eigenvalues ​​to obtain an eigenvalue sequence. The eigenvalue sequence is sorted in descending order of numerical value and the first few eigenvalues ​​are extracted as dominant eigenvalues. The ratio between the total energy of the dominant eigenvalues ​​and the total energy of the entire eigenvalue sequence is calculated to obtain the dominant energy percentage. The interference energy is obtained by calculating the sum of the energy of the last few eigenvalues ​​in the eigenvalue sequence. The interference energy ratio is obtained by calculating the ratio between the interference energy and the sum of the energy of the dominant eigenvalues. The signal quality index of each channel is obtained by weighting and combining the dominant energy ratio and the interference energy ratio.

[0021] Electrode units made of flexible conductive fabric are attached to the skin of the infant's chest. The flexible conductive fabric is woven from conductive and elastic fibers, with a thickness of 0.5 mm and a surface resistivity of less than 5 ohms per square centimeter. Each electrode unit has an area of ​​2 square centimeters and is arranged in a circle around the heart area. The electrode units are connected to the data acquisition module via anti-interference shielded wires, with a sampling frequency set to 500 Hz and a resolution of 16 bits. After the electrodes contact the skin, the raw ECG signal is initially amplified by a preamplifier with a gain of 100 times, and then filtered by a bandpass filter with a bandwidth of 0.5 to 100 Hz to remove baseline drift and high-frequency interference.

[0022] The raw analog ECG signal is converted into a digital signal by an analog-to-digital converter (ADC). The ADC uses a successive approximation converter with a conversion time of less than 2 microseconds. Data from each channel is stored in a 2-megabyte circular buffer according to its channel number. Time alignment correction is performed on the data from each channel based on a preset synchronization pulse. During correction, the first channel's data is used as a reference channel, and the timestamp of the synchronization pulse in each channel is recorded. The timestamp difference between each channel and the reference channel is calculated. Other channel data are shifted and compensated based on the calculated time difference. The compensation method involves adding appropriate zero values ​​before and after the data sequence or truncating redundant data points to align the effective signals of all channels in the time dimension, achieving a compensation accuracy within 2 milliseconds. After alignment, a multi-channel ECG data matrix is ​​formed. Each row in the matrix represents the time series data of one channel, and each column represents the multi-channel sampled values ​​at the same time.

[0023] A sliding window cross-correlation calculation was performed on any two channels in the multi-channel ECG data. The sliding window size was set to 2 seconds, i.e., 1000 sampling points, with a window overlap rate of 50%. For example, for the data sequences of channel i and channel j, the mean of both data sequences was first removed, and then the cross-correlation function was calculated by moving the data sequence point by point. The moving step size was set to 1 sampling point, and the moving range was set to ±200 milliseconds, i.e., ±100 sampling points. During the cross-correlation function calculation, the two data sequences were moved relative to each other, and the sum of their products at different moving positions was calculated to form the cross-correlation function curve. The peak value was extracted from the cross-correlation function as the cross-correlation coefficient, with a value ranging from -1 to 1. The accuracy of the peak position was improved using a three-point parabolic interpolation method. The time position corresponding to the peak value of the cross-correlation function was recorded, and the time difference between this position and the zero-delay point was calculated to obtain the phase difference between the two channels, in milliseconds.

[0024] Fill the corresponding positions in the spatial correlation strength matrix with the cross-correlation coefficients between all channel pairs, setting the diagonal elements to 1. For N channels, N×(N-1) / 2 cross-correlation coefficients need to be calculated. Fill the corresponding positions in the time propagation delay matrix with the phase differences between all channel pairs, setting the diagonal elements to 0. The two matrices are then combined element-weighted to form the spatiotemporal correlation matrix, with weighting coefficients of 0.6 and 0.4, respectively. For an 8-channel system, this results in an 8×8 spatiotemporal correlation matrix.

[0025] Eigenvalue decomposition is performed on the spatiotemporal correlation matrix, and the Jacobi iteration method is used to solve for eigenvalues ​​and eigenvectors. The iteration termination condition is set as an error less than 0.00001 or the number of iterations exceeding 100. The obtained eigenvalues ​​are sorted in descending order of numerical value, and the first three eigenvalues ​​are defined as dominant eigenvalues. The energy of the dominant eigenvalues ​​is calculated by summing the squares of the three dominant eigenvalues; the energy of all eigenvalues ​​is also calculated by summing the squares of all eigenvalues. The ratio of the two sums is the dominant energy proportion.

[0026] The sum of the energy of the last four eigenvalues ​​is calculated, and the squares of the four smallest eigenvalues ​​are added together to obtain the interference energy. The ratio of the interference energy to the sum of the eigenvalue energies is the interference energy ratio. The signal quality index calculation formula is: Signal Quality Index = Dominant Energy Ratio × 70 + (1 - Interference Energy Ratio) × 30, where the signal quality index ranges from 0 to 100. For each channel, the signal quality index of all channel pairs including the current channel is calculated, and the average value is taken as the signal quality index corresponding to the current channel.

[0027] In this embodiment, by constructing an inter-channel correlation mechanism based on sliding window cross-correlation analysis, spatial correlation strength and temporal propagation delay are modeled in a unified manner, achieving a comprehensive characterization of the intrinsic consistency and propagation characteristics of multi-channel ECG signals. This enables more effective differentiation between real ECG activity and random noise or local interference, improving the ability to identify abnormal channels and interference signals. By performing eigenvalue decomposition on the spatiotemporal correlation matrix and constructing dominant energy ratio and interference energy ratio indices based on energy distribution, quantitative separation of the dominant signal structure and interference components is achieved. Signal quality is evaluated from the overall energy structure level, enhancing the objectivity and robustness of the evaluation results. By weighted fusion of dominant energy ratio and interference energy ratio to form a comprehensive signal quality index, the precision and automation level of ECG signal quality assessment are improved, providing a more reliable data foundation for subsequent anomaly detection, heart rhythm analysis, or remote monitoring.

[0028] In one alternative implementation, Using the channel with the highest signal quality index as the reference channel, amplitude scaling and phase shifting are performed to obtain complete ECG data, including: The channel with the highest signal quality index is selected as the reference channel. The electrode position coordinates corresponding to the reference channel are obtained as the reference position coordinates. Each channel is traversed and the channel with a signal quality index lower than a preset threshold is identified as the channel to be reconstructed. The electrode position coordinates corresponding to the channel to be reconstructed are obtained as the position coordinates to be reconstructed. The Euclidean distance between the reference position coordinates and the position coordinates to be reconstructed is calculated to obtain the electrode spacing. An attenuation function is established based on the electrode spacing, and the attenuation amplitude of the ECG signal in the reference channel is calculated to obtain the attenuation amplitude. The attenuation amplitude is used as the initial reconstruction amplitude of the channel to be reconstructed. The spatial angle between the reference position coordinates and the position coordinates to be reconstructed is calculated to obtain the electrode position angle. The propagation delay of the cardiac electric field is calculated by weighting the electrode position angle to obtain the propagation delay. The ECG signal of the reference channel is shifted along the time axis according to the propagation delay to obtain a phase shift signal. The preliminary reconstructed amplitude is combined with the phase shift signal to obtain a reconstructed ECG signal. The reconstructed ECG signal is used to replace the signal segments in the channel to be reconstructed that have a signal quality index lower than a preset threshold, and the signal segments in the channel to be reconstructed that have a signal quality index higher than the preset threshold are retained to obtain complete ECG data.

[0029] The quality index (QI) of the acquired multi-channel ECG data is calculated. For an 8-channel ECG acquisition system, the QI ranges from 0 to 100, and the channel with the highest QI is selected as the reference channel. For example, if the QIs of the eight channels are 85, 92, 78, 65, 82, 76, 59, and 73, the second channel with a QI of 92 is selected as the reference channel. The electrode position coordinates corresponding to the reference channel are obtained as reference position coordinates. A three-dimensional Cartesian coordinate system is established with the midpoint of the line connecting the sternums as the origin, and the unit is centimeters. For example, the reference channel electrode position coordinates are (2.5, 3.8, 0.6).

[0030] Each channel is traversed, and channels with a signal quality index below a preset threshold are identified as channels to be reconstructed. The preset threshold can be set between 60 and 75 depending on the actual application scenario; in this embodiment, it is set to 70. Based on the previous example, the signal quality index of the fourth channel is 65, and the signal quality index of the seventh channel is 59, both below the preset threshold of 70. Therefore, these two channels are marked as channels to be reconstructed. The electrode position coordinates corresponding to the channels to be reconstructed are obtained as the reconstruction position coordinates. For example, the electrode position coordinates of the fourth channel are (-1.2, 4.3, 0.8), and the electrode position coordinates of the seventh channel are (3.6, -2.1, 1.2).

[0031] The electrode spacing is obtained by calculating the Euclidean distance between the reference position coordinates and the position coordinates to be reconstructed. The Euclidean distance is calculated as the straight-line distance between two points in space. Taking the fourth channel as an example, the electrode spacing between the reference channel and the fourth channel is calculated as the distance between the reference position coordinates (2.5, 3.8, 0.6) and the position coordinates to be reconstructed (-1.2, 4.3, 0.8), and the result is 3.78 cm. For the seventh channel, the calculated electrode spacing is 6.42 cm.

[0032] An attenuation function was established based on the electrode spacing, and the attenuation amplitude was calculated by attenuating the ECG signal amplitude of the reference channel. The attenuation function adopted an exponential attenuation model, where the amplitude attenuation coefficient decreases as the electrode spacing increases. The attenuation coefficient is equal to a power function with the base of the natural logarithm and the exponent being the negative electrode spacing multiplied by the attenuation constant. The attenuation constant was set to 0.2 per centimeter based on the conductivity characteristics of human tissue. For the fourth channel with an electrode spacing of 3.78 cm, the attenuation coefficient was calculated to be 0.47; for the seventh channel with an electrode spacing of 6.42 cm, the attenuation coefficient was calculated to be 0.28.

[0033] The amplitude of the ECG signal from the reference channel is multiplied by the corresponding attenuation coefficient to obtain the preliminary reconstructed amplitude of the channel to be reconstructed. The amplitude of each sampling point in the reference channel is then multiplied by the attenuation coefficient. Taking the reference channel signal acquired at a sampling rate of 500 Hz as an example, if the peak value of the R wave in the reference channel is 1.2 mV, then the peak value of the reconstructed R wave in the fourth channel is 0.56 mV, and the peak value of the reconstructed R wave in the seventh channel is 0.34 mV.

[0034] The electrode position angle is obtained by calculating the spatial angle between the reference position coordinates and the position to be reconstructed. The spatial angle is calculated by the angle between vectors. Taking the midpoint of the line connecting the sternum and the position to be reconstructed as the origin, the reference position and the position to be reconstructed form two vectors, and the angle between the two vectors is calculated. Taking the fourth channel as an example, the electrode position angle between the reference channel and the fourth channel is calculated to be 76.5 degrees; for the seventh channel, the calculated electrode position angle is 135.2 degrees.

[0035] The propagation delay is calculated by weighting the propagation delay of the cardiac electric field based on the electrode position angle. The propagation speed of the cardiac electric field varies in different directions, generally faster along the direction of the myocardial fibers and slower perpendicular to the direction of the myocardial fibers. The propagation delay is related to both the electrode position angle and the electrode spacing. In this embodiment, the propagation delay is calculated by dividing the electrode spacing by the conduction speed and then multiplying by an angle correction factor. The conduction speed is set to 0.5 meters per second, and the angle correction factor is 1 plus the sine value of the angle deviating from the reference direction divided by 2. The reference direction is set to 0 degrees, corresponding to the main direction of the cardiac conduction system.

[0036] For the fourth channel, the propagation delay is calculated to be 75.6 milliseconds; for the seventh channel, the propagation delay is calculated to be 154.1 milliseconds. The ECG signal of the reference channel is shifted along the time axis based on the propagation delay to obtain a phase shift signal. The signal of the reference channel is then shifted along the time axis by the number of sampling points corresponding to the propagation delay. Taking a 500 Hz sampling rate as an example, the time shift for the fourth channel is 38 sampling points, and the time shift for the seventh channel is 77 sampling points.

[0037] The reconstructed ECG signal is obtained by combining the initial reconstructed amplitude and the phase shift signal. The combination method is direct replacement, meaning the amplitude of the phase shift signal is replaced with the value of the initial reconstructed amplitude. To ensure a smooth transition of the reconstructed signal, median filtering is performed after amplitude replacement, with a filtering window size of 5 sampling points. The reconstructed signal retains the morphological characteristics of the reference channel while taking into account amplitude attenuation and phase delay caused by spatial differences.

[0038] The channel to be reconstructed is segmented. Segments with a signal quality index below a preset threshold are replaced with reconstructed ECG signals, while segments with a signal quality index above the preset threshold are retained. Segmentation uses a sliding window method with a window size of 2 seconds and a step size of 0.5 seconds. The local quality index is calculated for each window. If the local quality index is below the threshold, the original signal within that window is replaced with the reconstructed signal; if the local quality index is above the threshold, the original signal is retained. To avoid abrupt changes between signal segments, a linear fusion transition region is used at the boundaries of the replacement regions. The transition region length is set to 0.2 seconds, or 100 sampling points.

[0039] In this embodiment, by introducing the Euclidean distance between electrodes to construct an amplitude attenuation function and combining it with the spatial angle of the electrodes to calculate the propagation delay of the cardiac electric field, synchronous modeling of the amplitude attenuation characteristics and phase propagation characteristics of the ECG signal during spatial propagation is achieved. This fully considers the spatial propagation law of the cardiac electric field, ensuring that the reconstructed signal conforms to the physiological electrical propagation characteristics in both amplitude and time phase, thereby significantly improving the authenticity and physical consistency of the reconstructed signal. By performing amplitude attenuation processing on the reference signal and combining it with time axis offset to obtain a phase offset signal, and then combining them to generate the reconstructed ECG signal, coordinated reconstruction of amplitude and phase is achieved. This enhances the ability to maintain the QRS complex and other key waveform morphologies, reduces the probability of waveform distortion during reconstruction, and improves the integrity of waveform morphology. By only replacing signal segments with a signal quality index lower than a preset threshold while retaining original signal segments with good quality, segmented and adaptive reconstruction is achieved, avoiding over-processing of effective data, reducing the risk of information loss, and improving data utilization and reconstruction efficiency.

[0040] In one alternative implementation, The complete ECG data is subjected to wavelet packet decomposition and the ECG waveform is reconstructed. Based on the reconstructed ECG waveform, the peak position of the ventricular depolarization wave is detected and the continuous interval sequence is calculated. The boundary of the atrial depolarization wave is identified and the amplitude integral is calculated to obtain the atrial parameters, including: The complete ECG data is decomposed using wavelet packets, and the dominant frequency band is identified by calculating the energy distribution of each frequency band. Based on the decomposition coefficients corresponding to the dominant frequency band, wavelet packets are reconstructed to obtain the reconstructed ECG waveform. Time-frequency joint analysis is then performed to obtain the instantaneous frequency sequence and instantaneous amplitude sequence, and a time-frequency feature matrix is ​​constructed. The time-frequency feature matrix is ​​then decomposed using singular value decomposition to obtain the dominant singular values. Based on the dominant singular values, feature enhancement is performed on the reconstructed ECG waveform to obtain the enhanced ECG waveform. Morphological gradient operations are performed on the enhanced ECG waveform to obtain a waveform gradient sequence, and extreme point detection is used to determine the candidate peak positions. Local maximum verification is performed on the amplitude of the enhanced ECG waveform corresponding to the candidate peak positions, and the temporal width feature and frequency energy distribution feature of the ventricular depolarization wave are extracted and filtered to obtain the peak position of the ventricular depolarization wave. The peak positions of the ventricular depolarization wave are arranged in chronological order and differential operations are performed to obtain an instantaneous interval value sequence. Outlier detection and correction are performed on the instantaneous interval value sequence to obtain a continuous interval sequence. In the enhanced ECG waveform, a reverse time window is constructed with the peak position of the ventricular depolarization wave as the anchor point, and the zero point search of the second derivative is performed to obtain the starting boundary of the atrial depolarization wave. The waveform curvature change rate analysis is performed within the reverse time window to obtain the ending boundary of the atrial depolarization wave. The atrial depolarization wave amplitude integral is calculated by combining the starting boundary of the atrial depolarization wave as the atrial parameter.

[0041] Wavelet packet decomposition was performed on complete electrocardiogram (ECG) data, using bioorthogonal wavelets as the basis functions. A decomposition layer of 5 was set, resulting in wavelet packet decomposition coefficients for 32 frequency bands. The energy distribution of each frequency band's coefficients was calculated using the sum of squares of the wavelet packet coefficients. Taking an ECG signal with a sampling rate of 500 Hz as an example, after 5-layer wavelet packet decomposition, the frequency bands ranged from 0 to 250 Hz, with each sub-band having a width of 7.8125 Hz. Calculations of the energy distribution characteristics revealed that the energy of normal ECG signals is mainly concentrated in the range of 0.5 to 45 Hz, corresponding to sub-bands 1 to 6. Energy ranking shows that the third sub-band (15.625 to 23.4375 Hz) has the highest energy share, reaching 42% of the total energy; the second sub-band (7.8125 to 15.625 Hz) is second, accounting for 27%; and the fourth sub-band (23.4375 to 31.25 Hz) accounts for 15%. These three bands are identified as the dominant bands, and their combined energy exceeds 80% of the total energy.

[0042] Wavelet packet reconstruction was performed based on the decomposition coefficients corresponding to the dominant frequency band. Wavelet packet coefficients for the 2nd, 3rd, and 4th sub-bands were retained, while coefficients for other bands were set to zero. The reconstructed ECG waveform was obtained through inverse wavelet packet transform. The reconstruction process employed an inverse bioorthogonal wavelet transform algorithm, preserving the main frequency components of the original signal while effectively suppressing high-frequency noise and low-frequency baseline drift. Time-frequency joint analysis was performed on the reconstructed ECG waveform using the Hilbert-Huang transform method with a window length of 0.2 seconds and an overlap rate of 50%. Instantaneous frequency sequences (in Hertz) and instantaneous amplitude sequences (in millivolts) were obtained through the transform calculations. The two sequences were combined to form a time-frequency feature matrix, where rows represent time points and columns represent the corresponding instantaneous frequency and instantaneous amplitude.

[0043] Singular value decomposition (SVD) is performed on the time-frequency feature matrix to obtain a series of singular values ​​and their corresponding left and right singular vectors. The singular values ​​are arranged in descending order of magnitude. The first singular value, typically accounting for more than 85% of the total energy, is defined as the dominant singular value. The first two singular values ​​are selected as the dominant singular values ​​for subsequent processing. Feature enhancement is performed on the reconstructed ECG waveform based on the dominant singular values. The left and right singular vectors corresponding to the dominant singular values ​​are combined to form enhancement coefficients, which are then multiplied by the original reconstructed waveform. The enhancement process preserves the morphological characteristics of the original signal while improving the signal-to-noise ratio of ventricular and atrial depolarization waves.

[0044] Morphological gradient calculations were performed on the enhanced ECG waveform using a morphological operator with a structuring element length of 5 sampling points to calculate the waveform gradient sequence. The gradient calculation consisted of two parts: forward differencing and backward differencing. Forward differencing reflected the upward trend of the waveform, while backward differencing reflected the downward trend. These two parts were combined to form a complete gradient sequence, measured in millivolts per sampling point. Candidate peak positions were determined by extreme point detection, with the detection criterion being the zero-crossing point where the gradient sequence transitioned from positive to negative. To improve detection stability, a gradient threshold of 0.05 millivolts per sampling point was set; zero-crossing points below this threshold were filtered out.

[0045] Local maxima verification was performed on the amplitude of the enhanced ECG waveform corresponding to the candidate peak positions. The true local maxima were found within a range of 10 sampling points before and after the candidate position, serving as the corrected candidate positions. Time-domain width features were extracted for each candidate position, calculating the time interval between the peak and the point where the waveform amplitude decreased to 50% of its maximum value. The time-domain width of a normal ventricular depolarization wave is typically between 60 and 120 milliseconds. Simultaneously, frequency-domain energy distribution features were extracted. Fast Fourier Transform was performed on the signal segments of the candidate peak and the 50 sampling points before and after it, calculating the proportion of energy in the 5-25 Hz frequency band to the total energy. The frequency-domain energy proportion of a normal ventricular depolarization wave is typically higher than 70%. Combining the time-domain width and frequency-domain energy features, candidate peaks were screened to ultimately determine the peak position of the ventricular depolarization wave.

[0046] The peak positions of ventricular depolarization waves were arranged chronologically and differentially analyzed to obtain an instantaneous interval sequence in milliseconds. Outlier detection was performed on this sequence using the Thompson's τ-test with a significance level of 0.01. Detected outliers may represent arrhythmic events or false positives, requiring further correction. The correction method was based on the local interval mean. For intervals marked as abnormal, if their deviation from the local mean exceeded 40% and adjacent intervals were normal, they were considered false positives. If consecutive abnormal intervals occurred and conformed to specific patterns such as 2:1 or 3:1, they were marked as potential arrhythmic events. Through outlier detection and correction, a continuous interval sequence was obtained, providing a foundation for subsequent rhythm analysis.

[0047] In the enhanced ECG waveform, a reverse time window was constructed using the peak position of the ventricular depolarization wave as the anchor point, with a window length of 250 milliseconds. A search for the zero point of the second derivative was performed within the reverse time window. The second derivative of the ECG waveform reflects changes in waveform curvature, and zero-crossing points often correspond to inflection points in the waveform. The starting boundary of the atrial depolarization wave typically corresponds to the first significant second-derivative zero-crossing point within the reverse time window, and the search was performed retrospectively from the peak of the ventricular depolarization wave. A minimum amplitude threshold of 0.03 mV was set; zero-crossing points below this threshold were ignored.

[0048] Within a defined reverse time window, waveform curvature change rate analysis is performed, calculating the ratio of the first to second derivatives of the waveform to obtain a curvature change rate sequence. The termination boundary of the atrial depolarization wave corresponds to the maximum point of the curvature change rate sequence. For healthy adults, the duration of the atrial depolarization wave is typically between 60 and 100 milliseconds; if the detected duration exceeds this range, it is corrected based on physiological knowledge. Combining the atrial depolarization wave start and termination boundaries, the amplitude integral of the enhanced ECG waveform within this interval is calculated, i.e., the area between the waveform and the baseline, in millivolt-milliseconds.

[0049] In this embodiment, by performing morphological gradient calculation and extreme value verification on the enhanced electrocardiogram waveform, the peak position of the ventricular depolarization wave is accurately located, effectively reducing the probability of false detection and false negative detection, and improving the recognition stability and robustness under low amplitude or noise interference conditions. After anomaly detection and correction of the instantaneous interval value sequence, a more continuous and smooth interval sequence can be obtained, improving the accuracy of heart rate variability analysis and rhythm assessment. By constructing a reverse time window with the peak value of the ventricular depolarization wave as the anchor point, and combining the second derivative zero-point search and curvature change rate analysis to determine the start and end boundaries, adaptive positioning of the start and end positions of the atrial depolarization wave is achieved, enhancing the detection capability of low amplitude P waves.

[0050] In one alternative implementation, Ventricular parameters are obtained by measuring the time span from ventricular depolarization to repolarization. The time-domain statistics and frequency-domain power indices corresponding to the continuous interval sequence are calculated, and a physiological state feature vector is constructed by combining atrial and ventricular parameters, including: In the pre-acquired enhanced electrocardiogram waveform, the waveform amplitude baseline regression detection is performed starting from the peak position of the ventricular depolarization wave, and the ventricular repolarization endpoint is identified. The time difference between the peak position of the ventricular depolarization wave and the ventricular repolarization endpoint is calculated to obtain the ventricular parameters. The central tendency measure and dispersion measure are calculated for the continuous interval series to obtain the interval central location parameter and the interval dispersion parameter. The adjacent element change rate measure is calculated for the continuous interval series to obtain the interval short-range fluctuation parameter. The interval central location parameter and the interval dispersion parameter are combined and combined with the interval short-range fluctuation parameter to calculate the time domain statistics. The continuous interval sequence is transformed in the frequency domain to obtain a frequency domain representation sequence. Energy concentration is calculated in the low-frequency component range and the high-frequency component range to obtain a low-frequency energy concentration index and a high-frequency energy concentration index. Based on the relative relationship between the low-frequency energy concentration index and the high-frequency energy concentration index, an autonomic nervous system regulation balance index is calculated and the frequency domain power index is obtained. The time-domain statistics and the frequency-domain power index are obtained and concatenated with the atrial parameters and the ventricular parameters in a preset physiological correlation order to obtain the original feature sequence. The original feature sequence is then subjected to numerical scaling to obtain the physiological state feature vector.

[0051] In the pre-acquired enhanced ECG waveform, baseline regression detection of waveform amplitude was performed starting from the peak position of the ventricular depolarization wave. The baseline regression detection employed a tangent slope change method, searching for the amplitude decrease interval along the time axis from the peak position of the ventricular depolarization wave, and calculating the first-order difference value between adjacent sampling points. For ECG signals from young children, due to the rapid repolarization of myocardial cells, a detection threshold of 0.008 mV was set. When the absolute value of five consecutive difference values ​​was less than the threshold, the waveform was considered to be stable; the search continued, and when the amplitude change rate of eight consecutive sampling points was less than 0.4%, and the amplitude was close to the isoelectric line, that point was determined as the ventricular repolarization endpoint. In practical applications, the enhanced ECG waveform sampling rate was 500 Hz. The peak position of the ECG in a 3-year-old child corresponded to an amplitude of 0.95 mV, and the detected ventricular repolarization endpoint was located at the 165th sampling point after the peak, corresponding to an amplitude of 0.05 mV.

[0052] The ventricular parameters are obtained by calculating the time difference between the peak position of the ventricular depolarization wave and the endpoint of ventricular repolarization. The time difference is calculated by dividing the number of sampling points between the two points by the sampling rate, in milliseconds. In the example above, the time difference is 165 divided by 500 multiplied by 1000, which is 330 milliseconds. This ventricular parameter reflects the duration of the ventricular repolarization process in young children and is an important indicator for assessing myocardial repolarization characteristics. The normal range for ventricular parameters in young children is typically between 280 and 350 milliseconds. If it exceeds 370 milliseconds, it may indicate abnormal ventricular repolarization in young children.

[0053] Central tendency was calculated for continuous interval sequences using a weighted average method, assigning higher weights to more recent interval values. The weighting coefficients employed an exponential decay function, with the most recent interval having a weight of 1, and each subsequent interval decreasing in weight to 0.9 times the previous one. For a sequence containing 100 interval values, the weighted average yielded a central location parameter of 515 milliseconds for 2-year-old children, significantly lower than that of adults, corresponding to a heart rate of approximately 117 beats per minute. Simultaneously, dispersion was calculated using a weighted standard deviation method, with weighting consistent with the central tendency calculation. The results showed an interval dispersion parameter of 18 milliseconds, representing the range of fluctuation in the children's heart rate intervals.

[0054] The short-range fluctuation parameter of the interval was obtained by calculating the rate of change of adjacent elements in the continuous interval sequence. The rate of change was calculated by dividing the absolute difference between adjacent intervals by the value of the previous interval to obtain the relative rate of change; the root mean square of all relative rates of change was then taken to obtain the short-range fluctuation parameter. For the aforementioned interval sequence of young children, the calculated short-range fluctuation parameter was 0.052, indicating that the short-term fluctuation of heart rate variability in young children is relatively large. The interval centrality parameter and the interval dispersion parameter were combined by calculating their product, which yielded 92.7. This product was then weighted and summed with the short-range fluctuation parameter at a weight ratio of 6:4, resulting in a time-domain statistic of 77.3. This statistic comprehensively reflects the overall level, stability, and short-term variability characteristics of the young children's heart rate.

[0055] Frequency domain transformation was performed on the continuous interval sequence using the Fast Fourier Transform (FFT) method. Due to the high baseline heart rate of young children, cubic spline interpolation was performed on the interval sequence before transformation to convert it into an evenly sampled time series with a sampling frequency of 6 Hz. A Hanning window function was applied to the interpolated time series to reduce spectral leakage. A 1024-point FFT was then performed to obtain the frequency domain representation sequence. Energy concentration was calculated within the low-frequency component range (0.04 to 0.18 Hz) of the young children. The sum of the power spectral densities within this range was divided by the sum of the total power spectral densities to obtain the low-frequency energy concentration index, which was calculated to be 0.32.

[0056] Similarly, energy concentration was calculated within the high-frequency component range (0.18 to 0.5 Hz) for young children, yielding a high-frequency energy concentration index of 0.58, with an upper limit of 0.5 Hz for high frequencies. Based on the relative relationship between the low-frequency and high-frequency energy concentration indices, their ratio was calculated to obtain an autonomic nervous system regulation balance index of 0.552. This index reflects the balance between sympathetic and parasympathetic nervous system regulation in young children; a value below 1 indicates a relative dominance of the parasympathetic nervous system, consistent with the developmental characteristics of the parasympathetic nervous system in young children. The total power in the frequency domain was calculated as the sum of the power spectral densities within the range of 0.04 to 0.5 Hz, yielding a result of 2750 milliseconds squared.

[0057] The time-domain statistic 77.3 and frequency-domain power index 2750 were obtained and concatenated with the atrial parameter (integral amplitude of atrial depolarization wave) 3.8 mV·ms and the ventricular parameter 330 ms according to a preset physiological correlation order. The concatenation order was: ventricular parameter, atrial parameter, time-domain statistic, frequency-domain power index, and autonomic nervous system regulation balance index, resulting in the original feature sequence [330, 3.8, 77.3, 2750, 0.552]. Since the units and orders of magnitude of each element are different, numerical scaling is required.

[0058] The original feature sequence was numerically standardized using a mean-variance standardization method specific to preschool children. Standardization was performed based on the mean and standard deviation of each indicator within the normal reference range obtained from a large sample of preschool children. The reference mean for preschool ventricular parameters was 320 ms, with a standard deviation of 25 ms; the reference mean for preschool atrial parameters was 4.0 mV·ms, with a standard deviation of 0.9 mV·ms; the reference mean for preschool time-domain statistics was 75, with a standard deviation of 12; the reference mean for preschool frequency-domain power indicators was 3000 ms², with a standard deviation of 600 ms²; and the reference mean for preschool autonomic nervous system regulation balance indicators was 0.6, with a standard deviation of 0.15. Each element in the original feature sequence was subtracted from its corresponding reference mean and then divided by its corresponding standard deviation to obtain the standardized preschool physiological state feature vector [0.4, -0.22, 0.19, -0.42, -0.32].

[0059] In this embodiment, by comprehensively measuring the central tendency and dispersion of continuous interval sequences and constructing time-domain statistics by combining the rate of change of adjacent intervals, it is possible to simultaneously reflect the overall rhythm level, fluctuation amplitude, and short-range dynamic change characteristics, thereby improving the sensitivity and completeness of the representation of subtle changes in heart rate variability and enhancing the ability to identify rhythm abnormalities and changes in autonomic regulation. By performing frequency domain transformation on the interval sequences and calculating the low-frequency and high-frequency energy concentration indices respectively, a quantitative expression of the relative relationship between sympathetic and parasympathetic regulation is realized, enhancing the comprehensiveness and interpretability of the assessment of autonomic nervous system function. By structurally splicing time-domain statistics, frequency-domain power indices, and atrial and ventricular parameters according to the physiological correlation order to form a unified original feature sequence, a systematic integration of multidimensional physiological information is realized, improving the ability to express the correlation between different physiological dimensions.

[0060] In one alternative implementation, The system queries the current child's corresponding heart rate baseline reference interval and calculates the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, it calculates a calibrated heart rate baseline and corrects the physiological state feature vector to obtain the calibrated feature vector, which includes: Obtain the current child's age information in months and determine the current child's corresponding developmental stage classification. Based on the developmental stage classification, query the corresponding heart rate baseline reference interval from the preset physiological database. Extract the interval concentration location parameter from the physiological state feature vector and convert it into a heart rate value. Calculate the deviation distance between the heart rate value and the heart rate baseline reference interval and select the smallest non-negative value as the deviation amount. Calculate the width of the heart rate baseline reference interval and determine whether the deviation is greater than the width. If the deviation exceeds the width of the heart rate baseline reference interval, obtain the historical heart rate sequence corresponding to the current child, calculate the individualized heart rate benchmark value corresponding to the current child, and use it as the calibration heart rate baseline. The baseline offset between the calibrated heart rate baseline and the center value of the heart rate baseline reference interval is calculated. Correlation analysis is performed on each feature component in the physiological state feature vector to obtain the feature coupling relationship matrix and determine the correlation strength coefficient of each feature component with the interval concentration position parameter. The decomposed offset is obtained based on the correlation strength coefficient and the baseline offset. Based on the decomposed offset, each feature component in the physiological state feature vector is coordinatedly adjusted to obtain the calibrated feature components. The calibrated feature vector is obtained by combining the interval concentration position parameter.

[0061] The system obtains the current child's age in months and determines their corresponding developmental stage. The age in months is calculated by the difference between the child's birth date and the current date, accurate to the month. The developmental stage classification adopts the World Health Organization's recommended standards for child development, divided into six stages: neonatal period (0-1 month), early infancy (1-6 months), middle infancy (6-12 months), late infancy (12-18 months), early childhood (18-36 months), and late childhood (36-72 months). For example, a 24-month-old child would be classified as early childhood.

[0062] Based on developmental stage classification, the corresponding baseline heart rate reference range is retrieved from a pre-defined physiological database. The physiological database stores normal ranges for physiological parameters at each developmental stage, including heart rate, respiration, and blood pressure. The baseline heart rate reference range for early childhood is 90-130 beats / minute, corresponding to an interval of 600-462 milliseconds. The interval setpoint parameter is extracted from the physiological state feature vector and converted into a heart rate value. The conversion method is 60000 divided by the interval setpoint parameter, with units of beats / minute. For example, from the physiological state feature vector [0.4, -0.22, 0.19, -0.42, -0.32] obtained in the previous example, the interval setpoint parameter of 515 milliseconds before standardization is extracted and converted into a heart rate value of 116.5 beats / minute.

[0063] Calculate the deviation between the heart rate value and the baseline reference range, and select the smallest non-negative value as the deviation. The deviation is calculated as the absolute value of the difference between the heart rate value and the upper and lower boundaries of the reference range. When the heart rate value falls within the reference range, the deviation is 0. For a heart rate of 116.5 beats / minute, relative to the early childhood reference range of 90-130 beats / minute, the lower boundary deviation is 116.5 - 90 = 26.5, and the upper boundary deviation is 130 - 116.5 = 13.5. The smallest non-negative value of 13.5 is selected as the deviation.

[0064] The width of the heart rate baseline reference interval is calculated, and it is determined whether the deviation exceeds this width. The interval width is the upper boundary minus the lower boundary, i.e., 130 minus 90 equals 40. The deviation of 13.5 is checked against the interval width of 40; the result is no. If the deviation exceeds the width of the heart rate baseline reference interval, the historical heart rate sequence corresponding to the current child is retrieved, and the individualized heart rate baseline value corresponding to the current child is calculated and used as the calibration heart rate baseline. In this example, the deviation does not exceed the interval width, but for the sake of illustrating the complete process, it is assumed that the child's heart rate in another measurement scenario is 150 beats / minute, with a deviation of 20 beats / minute, exceeding half of the interval width (40 / 2=20), thus triggering the individualized calibration process.

[0065] Obtain the child's historical heart rate records for the past 30 days, totaling 20 data points. The values ​​are 115, 118, 120, 117, 122, 125, 121, 119, 123, 116, 120, 124, 119, 121, 118, 123, 125, 120, 122, and 117. The average value is calculated to be 120 beats / minute. This value is used as the child's individualized heart rate baseline, i.e., the calibrated heart rate baseline. Calculate the baseline offset between the calibrated heart rate baseline and the center value of the reference interval. The center value of the reference interval is (90+130) / 2 = 110 beats / minute. The baseline offset is the calibrated heart rate baseline minus the center value of the reference interval, i.e., 120 - 110 = 10 beats / minute.

[0066] Correlation analysis was performed on the feature components of the physiological state feature vector to obtain the coupling relationship matrix between features and to determine the correlation strength coefficient of each feature component with respect to the interval central location parameter. The correlation analysis used the Pearson correlation coefficient calculation method, based on a physiological data sample set of 500 children. The results showed that the correlation coefficients between ventricular parameters and the interval central location parameter were 0.65, atrial parameters 0.32, time-domain statistics 0.78, frequency-domain power index 0.56, and autonomic nervous system regulatory balance index 0.41. After normalization, the correlation strength coefficients of each feature component with respect to the interval central location parameter were 0.24, 0.12, 0.29, 0.21, and 0.14, respectively, with a sum of 1.

[0067] The decomposed offset is obtained by solving for the correlation strength coefficient and the baseline offset. The decomposed offset is calculated by multiplying the baseline offset by the correlation strength coefficient of each feature component. The interval offset corresponding to a baseline offset of 10 times / minute is -50 milliseconds. Multiplying this by the correlation strength coefficient yields a decomposed offset of -12 milliseconds for ventricular parameters, -6 milliseconds for atrial parameters, -14.5 milliseconds for time-domain statistics, -10.5 milliseconds for frequency-domain power indices, and -7 milliseconds for autonomic nervous system regulatory balance indices.

[0068] The calibrated feature components are obtained by co-adjusting each feature component in the physiological state feature vector based on the decomposition offset. The co-adjustment employs a component inverse mapping method, converting the decomposition offset in the interval domain into an adjustment amount in the standardized feature space. The conversion formula is the decomposition offset divided by the standard deviation of the corresponding feature component, then multiplied by the weighting coefficient of the feature component. The standard deviation of ventricular parameters is 25 milliseconds, atrial parameters are 0.9 mV·milliseconds, time-domain statistics are 12, frequency-domain power indices are 600 ms², and autonomic nervous system regulatory balance indices are 0.15. The weighting coefficients are set according to the developmental stage; in early childhood, the weights of each feature component are 0.3, 0.15, 0.25, 0.2, and 0.1, respectively.

[0069] The calculated adjustment values ​​for each feature component are: ventricular parameter -0.144, atrial parameter -1.0, time-domain statistic -0.302, frequency-domain power index -0.035, and autonomic nervous system regulation balance index -0.467. Subtracting the corresponding adjustment values ​​from each component in the original physiological state feature vector [0.4, -0.22, 0.19, -0.42, -0.32] yields the calibrated feature components [0.544, 0.78, 0.492, -0.385, 0.147]. The calibrated feature vector is obtained by combining this with the interval central location parameter. The interval central location parameter has been adjusted in the offset decomposition calculation; the new value is 515 - (-50) = 565 milliseconds, corresponding to a heart rate of 106.2 beats / minute, which is closer to the center value of the reference interval. The adjusted interval central location parameters are combined with the calibrated feature components to finally obtain the calibrated feature vector [0.544, 0.78, 0.492, -0.385, 0.147, 565].

[0070] In this embodiment, by calculating the deviation between the current heart rate value and the baseline reference interval, and combining the interval width for reasonableness judgment, when the deviation exceeds the group baseline range, historical heart rate sequences are introduced to construct an individualized heart rate benchmark value. This realizes a two-layer calibration mechanism that combines group reference and individual characteristics, improving the adaptability to children with long-term high or low heart rates, enhancing the inclusiveness of individual differences and the stability of judgment. By calculating the baseline offset and performing correlation analysis on each feature component in the physiological state feature vector, a feature coupling relationship matrix is ​​constructed, realizing multi-feature collaborative calibration. This allows the influence of heart rate offset on other physiological characteristics to be adjusted synchronously, avoiding the problem of imbalance in the overall feature structure caused by local correction, and improving the internal consistency and physiological rationality of the feature system. By decomposing the offset based on the correlation strength coefficient and coordinating the adjustment of each feature component, a weighted and directional fine correction mechanism is realized, improving the pertinence and scientific nature of the calibration process, and making the calibrated feature vector closer to the real physiological state.

[0071] Figure 2 This is a flowchart illustrating the individualized calibration process for heart rate monitoring using an intelligent analysis method for electrocardiogram signals in young children based on conductive fabric, as described in an embodiment of the present invention.

[0072] In one alternative implementation, Calculating the adjacent differences of the continuous interval sequence and determining interval stability, extracting the power ratio from the calibration feature vector to determine the autonomic nervous state, and combining the interval stability to determine the heart rhythm type and generate analytical results include: The interval change sequence is obtained by calculating the point-to-point adjacent difference of the continuous interval sequence, and the interval fluctuation amplitude sequence is obtained by extracting the absolute value. The variance of the interval fluctuation amplitude sequence within a preset sliding window is calculated to obtain the local fluctuation intensity sequence, and the interval stability is obtained by performing global statistical analysis. The low-frequency energy concentration index and the high-frequency energy concentration index are extracted from the calibration feature vector and their ratio is calculated to obtain the autonomic nervous system regulation balance index as the power ratio. The power ratio is then mapped to the sympathetic nervous system sensitive area and the parasympathetic nervous system sensitive area to obtain the quantification value of the sympathetic nervous system dominance and the quantification value of the parasympathetic nervous system dominance, and the autonomic nervous system state is obtained by solving the problem. The temporal regularity of the interval stability is quantified to obtain the rhythmic characteristics of the heart rhythm; the dynamic regulation of the autonomic nervous state is quantified to obtain the regulatory characteristics of the heart rhythm; the rhythmic characteristics of the heart rhythm are subjected to a periodicity test to obtain the periodicity test result; the regulatory characteristics of the heart rhythm are subjected to a consistency test to obtain the consistency test result; based on the periodicity test result and the consistency test result, a co-discrimination is performed to obtain the heart rhythm type; the heart rhythm type is data encapsulated and combined with the interval stability and the autonomic nervous state fields to obtain the parsing result.

[0073] The interval variation sequence is obtained by calculating the point-to-point difference between consecutive interval sequences. Taking a consecutive interval sequence collected from a 2-year-old child as an example, the consecutive interval sequence contains 120 cardiac interval values ​​in milliseconds, some of which are: [512, 520, 505, 525, 518, 509, 530, 522, 508, 516]. The interval variation sequence is obtained by calculating the difference between adjacent interval values, with corresponding values ​​of: [8, -15, 20, -7, -9, 21, -8, -14, 8]. The absolute value of the interval variation sequence is extracted to obtain the interval fluctuation amplitude sequence, with corresponding values ​​of: [8, 15, 20, 7, 9, 21, 8, 14, 8].

[0074] The variance of the interval fluctuation amplitude sequence within a preset sliding window is calculated to obtain the local fluctuation intensity sequence. The sliding window size is set to 10 heartbeats, with a step size of 1 heartbeat. For the ECG data of young children, the window size corresponds to a time span of approximately 5 to 6 seconds. The variance of the interval fluctuation amplitude is calculated within each sliding window to obtain the local fluctuation intensity. For example, the variance of the interval fluctuation amplitude in the first window is 26.4, the variance in the second window is 28.7, and so on. Global statistical analysis is performed on the local fluctuation intensity sequence, including calculating the mean, median, and interquartile range to obtain the interval stability index. For the above data of young children, the calculated local fluctuation intensity is 24.8, the median is 22.5, and the interquartile range is 12.3. The weighted average of the mean and median is used as the core index of interval stability, with a weight ratio of 6:4, and the calculated result is 23.9.

[0075] The low-frequency energy concentration index and high-frequency energy concentration index are extracted from the calibration feature vector, and their ratio is calculated to obtain the autonomic nervous system regulation balance index as the power ratio. Taking the feature vector of a 3-year-old child as an example, the low-frequency energy concentration index is 0.35, the high-frequency energy concentration index is 0.55, and the calculated power ratio is 0.636. The power ratio is mapped to the sympathetic nervous system sensitivity interval using a piecewise linear mapping function. When the power ratio is less than 0.5, the sympathetic nervous system dominance quantification value is 0; when the power ratio is between 0.5 and 1.2, it is linearly mapped to the range of 0 to 1; when the power ratio is greater than 1.2, the quantification value is 1. For a power ratio of 0.636, the sympathetic nervous system dominance quantification value is calculated to be 0.195.

[0076] The power ratio was mapped to the parasympathetic sensitivity zone. When the power ratio was greater than 1.5, the quantified value of parasympathetic dominance was 0; when the power ratio was between 0.6 and 1.5, it was linearly mapped to the range of 0 to 1; and when the power ratio was less than 0.6, the quantified value was 1. For a power ratio of 0.636, the quantified value of parasympathetic dominance was calculated to be 0.958. The quantified values ​​of sympathetic and parasympathetic dominance together constitute a quantitative description of the autonomic nervous system state of the child. The calculated difference was -0.763, with a negative value indicating parasympathetic dominance and the absolute value reflecting the degree of dominance. Mapping the aforementioned difference to a five-level scale yielded an autonomic nervous system state of "parasympathetic dominance".

[0077] The temporal regularity of interval stability was quantified to obtain the rhythmic characteristics of the heart rhythm. The complexity and regularity of the interval sequence were evaluated using entropy calculation. The original interval sequence was normalized and quantified into eight levels, and the corresponding sample entropy was calculated. For the aforementioned preschool data, the calculated sample entropy was 1.72. The sample entropy was converted into a regularity index, calculated by subtracting the sample entropy from 2, resulting in 0.28. The closer this value is to 1, the more regular the heart rate changes; the closer it is to 0, the more random the heart rate changes. Based on this, the rhythmic characteristics of the heart rhythm were determined to be "low regularity".

[0078] The regulatory dynamics of the autonomic nervous system were quantified to obtain the rhythm regulation characteristics. Regulatory dynamics are measured by the degree of change in the autonomic nervous system state over time. The raw electrocardiogram (ECG) recordings were divided into 5-minute segments, and the autonomic nervous system state was calculated for each segment, resulting in a state sequence. The frequency and amplitude of state changes in the sequence were calculated, and a regulatory dynamics index was obtained. For the analysis of a 20-minute ECG recording of a 4-year-old child, the calculated regulatory dynamics index was 0.65, indicating that the autonomic nervous system regulation has a moderate degree of dynamic change capability. Based on this, the rhythm regulation characteristic was determined to be "moderate regulation".

[0079] A periodicity test was performed on the rhythmic characteristics of the heart rhythm, yielding periodicity test results. Autocorrelation analysis was used to calculate the autocorrelation coefficients at different time lags to identify significant periodic patterns. For the aforementioned data on young children, the maximum autocorrelation coefficient occurred at a time lag of 12 heartbeats, with a coefficient value of 0.32, which is below the significance threshold of 0.5. Therefore, the periodicity test result was "no significant periodicity." A consistency test was performed on the regulatory characteristics of the heart rhythm, yielding consistency test results. The consistency test was conducted by comparing the stability of the regulatory characteristics across different time periods. The coefficient of variation of the regulatory index for each time period was calculated, and the result was 0.21, below the consistency threshold of 0.3, indicating good consistency of the regulatory characteristics. Therefore, the consistency test result was "good consistency."

[0080] Heart rhythm types are determined through a collaborative discrimination based on the results of periodicity and consistency tests. A discrimination rule table is established, and the heart rhythm type is determined according to the combination of periodicity and consistency. When the periodicity is "no significant periodicity" and the consistency is "good consistency," the heart rhythm type is determined to be "stable aperiodic." The heart rhythm type data is encapsulated, and the encapsulation format includes three fields: type code, type name, and type description. The type code is "SNP," the type name is "stable aperiodic," and the type description is "heart rate changes have no obvious periodic pattern but have stable regulatory characteristics."

[0081] In this embodiment, by calculating the ratio of low-frequency to high-frequency energy concentration indices and mapping them to the sensitive regions of the sympathetic and parasympathetic nervous systems, a quantitative decomposition of the autonomic nervous system's regulatory state is achieved. This makes the expression of the autonomic nervous system's state more intuitive and interpretable, improving the accuracy of identifying imbalances in the nervous system's regulation. By quantifying the temporal regularity of interval stability, the rhythmic characteristics of the heart rhythm are obtained, and by quantifying the dynamic regulation of the autonomic nervous system's state, the regulatory characteristics of the heart rhythm are obtained. This achieves a dual-dimensional characterization from both the rhythmic structure and regulatory mechanism levels, improving the analytical depth of complex heart rhythm manifestations. By performing periodicity and consistency tests separately and then co-judging based on the results of the two tests, misclassification problems caused by a single discrimination criterion are avoided, improving the accuracy and robustness of heart rhythm type identification.

[0082] A second aspect of this invention provides an intelligent analysis system for electrocardiogram (ECG) signals in young children based on conductive fabric, comprising: The ECG acquisition unit is used to acquire multi-channel ECG data of children through flexible conductive fabric, calculate the cross-correlation coefficient and phase difference between different channels to construct a spatiotemporal correlation matrix, perform eigenvalue decomposition on the spatiotemporal correlation matrix and calculate the signal quality index of each channel, and use the channel with the highest signal quality index as the reference channel to perform amplitude scaling and phase shifting to obtain complete ECG data. The feature extraction unit is used to perform wavelet packet decomposition on the complete electrocardiogram data and reconstruct the electrocardiogram waveform. Based on the reconstructed electrocardiogram waveform, it detects the peak position of the ventricular depolarization wave and calculates the continuous interval sequence. It identifies the boundary of the atrial depolarization wave and calculates the amplitude integral to obtain the atrial parameters. It measures the time span from ventricular depolarization to repolarization to obtain the ventricular parameters. It calculates the time domain statistics and frequency domain power index corresponding to the continuous interval sequence and constructs a physiological state feature vector by combining the atrial and ventricular parameters. The heart rate calibration unit is used to query the heart rate baseline reference interval corresponding to the current child and calculate the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, the calibration heart rate baseline is calculated and the physiological state feature vector is corrected to obtain the calibration feature vector. The heart rhythm analysis unit is used to calculate the adjacent differences of the continuous interval sequence and determine the interval stability, extract the power ratio in the calibration feature vector to determine the autonomic nervous state, and combine the interval stability to determine the heart rhythm type and generate analysis results.

[0083] A third aspect of the present invention provides an electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0084] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0085] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent analysis of electrocardiogram signals in young children based on conductive fabric, characterized in that, include: Multi-channel electrocardiogram (ECG) data of young children are collected by flexible conductive fabric, and the cross-correlation coefficients and phase differences between different channels are calculated to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix is ​​decomposed by eigenvalue and the signal quality index of each channel is calculated. The channel with the highest signal quality index is used as the reference channel for amplitude scaling and phase shifting to obtain complete ECG data. The complete electrocardiogram (ECG) data is decomposed by wavelet packets and the ECG waveform is reconstructed. Based on the reconstructed ECG waveform, the peak position of the ventricular depolarization wave is detected and the continuous interval sequence is calculated. The boundary of the atrial depolarization wave is identified and the amplitude integral is calculated to obtain the atrial parameters. The time span from ventricular depolarization to repolarization is measured to obtain the ventricular parameters. The time-domain statistics and frequency-domain power index corresponding to the continuous interval sequence are calculated and the physiological state feature vector is constructed by combining the atrial and ventricular parameters. Query the current child's corresponding heart rate baseline reference interval and calculate the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, calculate the calibrated heart rate baseline and correct the physiological state feature vector to obtain the calibrated feature vector. The adjacent differences of the continuous interval sequence are calculated and the interval stability is determined. The power ratio in the calibration feature vector is extracted to determine the autonomic nervous state. The heart rhythm type is determined by combining the interval stability and the analysis results are generated.

2. The method according to claim 1, characterized in that, Multi-channel electrocardiogram (ECG) data of young children were collected using flexible conductive fabric, and the cross-correlation coefficients and phase differences between different channels were calculated to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix was then subjected to eigenvalue decomposition, and the signal quality index of each channel was calculated, including: Raw electrocardiogram (ECG) signals are collected by contacting the skin of the infant's chest with a flexible conductive fabric electrode unit. The raw ECG signals are then converted from analog to digital to obtain digital ECG signals. The digital ECG signals are stored according to channel numbers and time-aligned to obtain multi-channel ECG data. A sliding window cross-correlation calculation is performed between any two channels in the multi-channel electrocardiogram data to obtain a cross-correlation function, and the peak value of the cross-correlation function is extracted as the cross-correlation coefficient. The phase difference is obtained by calculating the time delay of the peak position of the cross-correlation function. The cross-correlation coefficient is used as the correlation strength in the spatial dimension, and the phase difference is used as the propagation delay in the time dimension to construct a spatiotemporal correlation matrix. The spatiotemporal correlation matrix is ​​decomposed into eigenvalues ​​to obtain an eigenvalue sequence. The eigenvalue sequence is sorted in descending order of numerical value and the first few eigenvalues ​​are extracted as dominant eigenvalues. The ratio between the total energy of the dominant eigenvalues ​​and the total energy of the entire eigenvalue sequence is calculated to obtain the dominant energy percentage. The interference energy is obtained by calculating the sum of the energy of the last few eigenvalues ​​in the eigenvalue sequence. The interference energy ratio is obtained by calculating the ratio between the interference energy and the sum of the energy of the dominant eigenvalues. The signal quality index of each channel is obtained by weighting and combining the dominant energy ratio and the interference energy ratio.

3. The method according to claim 1, characterized in that, Using the channel with the highest signal quality index as the reference channel, amplitude scaling and phase shifting are performed to obtain complete ECG data, including: The channel with the highest signal quality index is selected as the reference channel. The electrode position coordinates corresponding to the reference channel are obtained as the reference position coordinates. Each channel is traversed and the channel with a signal quality index lower than a preset threshold is identified as the channel to be reconstructed. The electrode position coordinates corresponding to the channel to be reconstructed are obtained as the position coordinates to be reconstructed. The Euclidean distance between the reference position coordinates and the position coordinates to be reconstructed is calculated to obtain the electrode spacing. An attenuation function is established based on the electrode spacing, and the attenuation amplitude of the ECG signal in the reference channel is calculated to obtain the attenuation amplitude. The attenuation amplitude is used as the initial reconstruction amplitude of the channel to be reconstructed. The spatial angle between the reference position coordinates and the position coordinates to be reconstructed is calculated to obtain the electrode position angle. The propagation delay of the cardiac electric field is calculated by weighting the electrode position angle to obtain the propagation delay. The ECG signal of the reference channel is shifted along the time axis according to the propagation delay to obtain a phase shift signal. The preliminary reconstructed amplitude is combined with the phase shift signal to obtain a reconstructed ECG signal. The reconstructed ECG signal is used to replace the signal segments in the channel to be reconstructed that have a signal quality index lower than a preset threshold, and the signal segments in the channel to be reconstructed that have a signal quality index higher than the preset threshold are retained to obtain complete ECG data.

4. The method according to claim 1, characterized in that, The complete ECG data is subjected to wavelet packet decomposition and the ECG waveform is reconstructed. Based on the reconstructed ECG waveform, the peak position of the ventricular depolarization wave is detected and the continuous interval sequence is calculated. The boundary of the atrial depolarization wave is identified and the amplitude integral is calculated to obtain the atrial parameters, including: The complete ECG data is decomposed using wavelet packets, and the dominant frequency band is identified by calculating the energy distribution of each frequency band. Based on the decomposition coefficients corresponding to the dominant frequency band, wavelet packets are reconstructed to obtain the reconstructed ECG waveform. Time-frequency joint analysis is then performed to obtain the instantaneous frequency sequence and instantaneous amplitude sequence, and a time-frequency feature matrix is ​​constructed. The time-frequency feature matrix is ​​then decomposed using singular value decomposition to obtain the dominant singular values. Based on the dominant singular values, feature enhancement is performed on the reconstructed ECG waveform to obtain the enhanced ECG waveform. Morphological gradient operations are performed on the enhanced ECG waveform to obtain a waveform gradient sequence, and extreme point detection is used to determine the candidate peak positions. Local maximum verification is performed on the amplitude of the enhanced ECG waveform corresponding to the candidate peak positions, and the temporal width feature and frequency energy distribution feature of the ventricular depolarization wave are extracted and filtered to obtain the peak position of the ventricular depolarization wave. The peak positions of the ventricular depolarization wave are arranged in chronological order and differential operations are performed to obtain an instantaneous interval value sequence. Outlier detection and correction are performed on the instantaneous interval value sequence to obtain a continuous interval sequence. In the enhanced ECG waveform, a reverse time window is constructed with the peak position of the ventricular depolarization wave as the anchor point, and the zero point search of the second derivative is performed to obtain the starting boundary of the atrial depolarization wave. The waveform curvature change rate analysis is performed within the reverse time window to obtain the ending boundary of the atrial depolarization wave. The atrial depolarization wave amplitude integral is calculated by combining the starting boundary of the atrial depolarization wave as the atrial parameter.

5. The method according to claim 1, characterized in that, Ventricular parameters are obtained by measuring the time span from ventricular depolarization to repolarization. The time-domain statistics and frequency-domain power indices corresponding to the continuous interval sequence are calculated, and a physiological state feature vector is constructed by combining atrial and ventricular parameters, including: In the pre-acquired enhanced electrocardiogram waveform, the waveform amplitude baseline regression detection is performed starting from the peak position of the ventricular depolarization wave, and the ventricular repolarization endpoint is identified. The time difference between the peak position of the ventricular depolarization wave and the ventricular repolarization endpoint is calculated to obtain the ventricular parameters. The central tendency measure and dispersion measure are calculated for the continuous interval series to obtain the interval central location parameter and the interval dispersion parameter. The adjacent element change rate measure is calculated for the continuous interval series to obtain the interval short-range fluctuation parameter. The interval central location parameter and the interval dispersion parameter are combined and combined with the interval short-range fluctuation parameter to calculate the time domain statistics. The continuous interval sequence is transformed in the frequency domain to obtain a frequency domain representation sequence. Energy concentration is calculated in the low-frequency component range and the high-frequency component range to obtain a low-frequency energy concentration index and a high-frequency energy concentration index. Based on the relative relationship between the low-frequency energy concentration index and the high-frequency energy concentration index, an autonomic nervous system regulation balance index is calculated and the frequency domain power index is obtained. The time-domain statistics and the frequency-domain power index are obtained and concatenated with the atrial parameters and the ventricular parameters in a preset physiological correlation order to obtain the original feature sequence. The original feature sequence is then subjected to numerical scaling to obtain the physiological state feature vector.

6. The method according to claim 1, characterized in that, The system queries the current child's corresponding heart rate baseline reference interval and calculates the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, it calculates a calibrated heart rate baseline and corrects the physiological state feature vector to obtain the calibrated feature vector, which includes: Obtain the current child's age information in months and determine the current child's corresponding developmental stage classification. Based on the developmental stage classification, query the corresponding heart rate baseline reference interval from the preset physiological database. Extract the interval concentration location parameter from the physiological state feature vector and convert it into a heart rate value. Calculate the deviation distance between the heart rate value and the heart rate baseline reference interval and select the smallest non-negative value as the deviation amount. Calculate the width of the heart rate baseline reference interval and determine whether the deviation is greater than the width. If the deviation exceeds the width of the heart rate baseline reference interval, obtain the historical heart rate sequence corresponding to the current child, calculate the individualized heart rate benchmark value corresponding to the current child, and use it as the calibration heart rate baseline. The baseline offset between the calibrated heart rate baseline and the center value of the heart rate baseline reference interval is calculated. Correlation analysis is performed on each feature component in the physiological state feature vector to obtain the feature coupling relationship matrix and determine the correlation strength coefficient of each feature component with the interval concentration position parameter. The decomposed offset is obtained based on the correlation strength coefficient and the baseline offset. Based on the decomposed offset, each feature component in the physiological state feature vector is coordinatedly adjusted to obtain the calibrated feature components. The calibrated feature vector is obtained by combining the interval concentration position parameter.

7. The method according to claim 1, characterized in that, Calculating the adjacent differences of the continuous interval sequence and determining interval stability, extracting the power ratio from the calibration feature vector to determine the autonomic nervous state, and combining the interval stability to determine the heart rhythm type and generate analytical results include: The interval change sequence is obtained by calculating the point-to-point adjacent difference of the continuous interval sequence, and the interval fluctuation amplitude sequence is obtained by extracting the absolute value. The variance of the interval fluctuation amplitude sequence within a preset sliding window is calculated to obtain the local fluctuation intensity sequence, and the interval stability is obtained by performing global statistical analysis. The low-frequency energy concentration index and the high-frequency energy concentration index are extracted from the calibration feature vector and their ratio is calculated to obtain the autonomic nervous system regulation balance index as the power ratio. The power ratio is then mapped to the sympathetic nervous system sensitive area and the parasympathetic nervous system sensitive area to obtain the quantification value of the sympathetic nervous system dominance and the quantification value of the parasympathetic nervous system dominance, and the autonomic nervous system state is obtained by solving the problem. The temporal regularity of the interval stability is quantified to obtain the rhythmic characteristics of the heart rhythm; the dynamic regulation of the autonomic nervous state is quantified to obtain the regulatory characteristics of the heart rhythm; the rhythmic characteristics of the heart rhythm are subjected to a periodicity test to obtain the periodicity test result; the regulatory characteristics of the heart rhythm are subjected to a consistency test to obtain the consistency test result; based on the periodicity test result and the consistency test result, a co-discrimination is performed to obtain the heart rhythm type; the heart rhythm type is data encapsulated and combined with the interval stability and the autonomic nervous state fields to obtain the parsing result.

8. A smart analysis system for electrocardiogram signals of infants based on conductive fabric, used to implement the method of any one of claims 1-7, characterized in that, include: The ECG acquisition unit is used to acquire multi-channel ECG data of children through flexible conductive fabric, calculate the cross-correlation coefficient and phase difference between different channels to construct a spatiotemporal correlation matrix, perform eigenvalue decomposition on the spatiotemporal correlation matrix and calculate the signal quality index of each channel, and use the channel with the highest signal quality index as the reference channel to perform amplitude scaling and phase shifting to obtain complete ECG data. The feature extraction unit is used to perform wavelet packet decomposition on the complete electrocardiogram data and reconstruct the electrocardiogram waveform. Based on the reconstructed electrocardiogram waveform, it detects the peak position of the ventricular depolarization wave and calculates the continuous interval sequence. It identifies the boundary of the atrial depolarization wave and calculates the amplitude integral to obtain the atrial parameters. It measures the time span from ventricular depolarization to repolarization to obtain the ventricular parameters. It calculates the time domain statistics and frequency domain power index corresponding to the continuous interval sequence and constructs a physiological state feature vector by combining the atrial and ventricular parameters. The heart rate calibration unit is used to query the heart rate baseline reference interval corresponding to the current child and calculate the deviation corresponding to the physiological state feature vector. If the deviation exceeds the width of the heart rate baseline reference interval, the calibration heart rate baseline is calculated and the physiological state feature vector is corrected to obtain the calibration feature vector. The heart rhythm analysis unit is used to calculate the adjacent differences of the continuous interval sequence and determine the interval stability, extract the power ratio in the calibration feature vector to determine the autonomic nervous state, and combine the interval stability to determine the heart rhythm type and generate analysis results.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images