Magnetocardiogram signal feature wave detection and evaluation method based on adaptive threshold estimation

Abstract

The present application relates to a magnetocardiogram feature wave detection and evaluation method based on adaptive threshold estimation, which automatically realizes the detection, correction and abnormal estimation of P wave, QRS wave group and T wave of multi-channel magnetocardiogram. The method comprises: using a sensor to collect multi-channel resting-state magnetocardiogram, removing high-frequency interference and baseline drift; detecting and correcting R wave peak value based on adaptive signal threshold, buffer threshold and noise threshold, and positioning Q wave and S wave according to the detection result; detecting and correcting T wave and P wave position based on an adaptive single threshold method; extracting features and evaluating quality of the magnetocardiogram based on the feature wave detection result, and determining abnormal feature waves and abnormal cardiac cycles. The method fully utilizes the advantages of magnetocardiogram itself, realizes the synchronous detection of multi-channel magnetocardiogram feature waves, improves the accuracy and automation degree of magnetocardiogram feature wave detection, and has important guiding significance for clinical diagnosis and data analysis of magnetocardiogram.

Description

Technical Field

[0001] This invention belongs to the field of biomedical signal processing, and in particular relates to a method for detecting and evaluating characteristic waves of magnetic cardiomyocyte signals based on adaptive threshold estimation. Background Technology

[0002] Magnetocardiography (MCG) is a technique that utilizes extremely weak magnetic field measurements to measure the spatial magnetic field distribution of the heart caused by its electrophysiological activity. MCG signals contain rich physiological and pathological information and have become a promising non-invasive diagnostic tool with significant medical application value.

[0003] Similar to electrocardiograms (ECGs), the characteristic waveforms of standard magnetic resonance imaging (MRI) signals include the P wave, QRS complex, and T wave. The P wave reflects the depolarization process of the left and right atria; the QRS complex reflects the depolarization process of the ventricles, with the Q wave being the first downward-sloping wave before the R wave and the S wave being the first downward-sloping wave after the R wave; the T wave reflects the repolarization process of the ventricles. The temporal and amplitude characteristics of these MRI signals are crucial for clinical diagnosis and form the basis for subsequent feature extraction and the establishment of disease diagnostic classification models. Therefore, establishing an automated, standardized, and highly accurate characteristic waveform detection process is of great significance for the diagnosis of diseases using MRI signals.

[0004] Currently, there are relatively few methods for feature wave detection in multi-channel magnetocardiogram (MCC) signals, mainly including deep learning methods and traditional signal processing methods. Deep learning-based methods heavily rely on building a pre-constructed MCC signal dataset with labeled feature wave locations, training a neural network based on these features to detect the feature waves. However, these methods require a large amount of high-quality MCC data, and the manual labeling of each feature wave location beforehand is costly and susceptible to interference, resulting in poor model stability and robustness. Feature wave detection based on traditional signal processing techniques mainly includes methods based on empirical mode decomposition, wavelet transform, and signal envelope. These methods are effective in identifying R-wave and T-wave peaks and have been widely used in electrocardiography (ECG) with good results. However, due to differences between MCC and ECG signals in acquisition methods, sensor array arrangement, and feature wave morphology, the accuracy of feature wave detection using traditional preprocessing methods for MCC signals is poor. Therefore, it is necessary to develop an automatic method for MCC signal feature wave detection and evaluation, taking into account the characteristics of MCC signals and combining their advantages in disease diagnosis. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for detecting and evaluating characteristic waves of magnetic resonance imaging (MRI) signals based on adaptive threshold estimation. By combining the characteristics of multi-channel MRI signals, it automatically detects, corrects, and estimates anomalies in P waves, QRS complexes, and T waves, providing significant guidance for clinical diagnosis and data analysis of MRI signals.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation, specifically including the following steps: S10: Use sensors to acquire multi-channel magnetocardiogram signals and remove high-frequency interference and baseline drift; S20: R-wave peak detection and correction are performed based on an adaptive multi-threshold method, and Q-wave and S-wave are located based on the detection results. The adaptive multi-threshold method includes signal threshold, buffer threshold and noise threshold. S30: T-wave and P-wave position detection and correction based on adaptive single threshold method; S40: Based on the characteristic wave detection results, perform feature extraction and quality assessment on the magnetic heart signal to identify abnormal characteristic waves and abnormal cardiac cycles.

[0007] Optionally, step S10 includes: Multi-channel resting-state magnetocardiogram (MCC) signals were acquired using an optically pumped magnetometer. High-frequency interference and baseline drift in the MCC signals were removed using a wavelet thresholding denoising method to obtain the denoised MCC signals.

[0008] Optionally, step S20 includes: S21. Principal component analysis (PCA) is used to reduce the dimension of the denoised magnetic-cardiogram (MCC) signal. The first m principal components are selected based on the cumulative variance contribution rate threshold of 90%, and the data is reconstructed according to the weight of the cumulative variance contribution rate to obtain the dimensionality-reduced MCC signal. The dimensionality-reduced MCC signal is then subjected to 8-level wavelet decomposition, and the signal is reconstructed using the 5th and 6th level wavelet approximation coefficients for R-wave peak detection. S22, the reconstructed signal used for R-wave peak detection is dot-producted with itself, and processed using a moving average filter with a window size of 150 to obtain the enhanced R-wave magnetocardiogram (MCC) signal. An adaptive multi-threshold algorithm is used to detect and correct the effective R-wave peaks in the dimensionality-reduced MCC signal. The adaptive multi-threshold algorithm includes a signal threshold, a buffer threshold, and a noise threshold. RR interval estimation is introduced as supplementary information to constrain the detection results. The RR interval estimation is an estimate of the interval between effective R-wave peak points in the MCC signal. The signal threshold serves as the final criterion for determining effective R-wave peak points. The buffer threshold acts as a buffer region for the signal threshold, further judging local peak points when the input signal exhibits baseline drift, abnormal pacemakers, large fluctuations in R-wave peak points, and large T-wave amplitudes. The noise threshold serves as a criterion for judging noise signals, reducing the impact of baseline drift on the signal threshold and buffer threshold. The RR interval estimation serves as one of the criteria for judging effective R-wave peak points, providing supplementary information for judging various local peak points, and simultaneously performing error checking and correction of the detection results. S23. Calculate the global average field power (GMFP) curve of the denoised magnetic field signal. Determine the search regions for Q and S waves based on the minimum points of the GMFP curve. Determine the positions of Q and S waves based on the maximum points of the dimensionality-reduced magnetic field signal.

[0009] Optionally, step S30 includes: S31, Remove the QRS complex based on the calculation results of step S23, and perform baseline correction on the denoised magnetocardiogram signal based on the position coordinates of the edge points of the removal window to obtain the characteristic curves of the T wave and P wave respectively. S32, calculate the GMFP curves of the T-wave and P-wave characteristic curves; process the T-wave GMFP curves with moving average filters with window sizes of 30 and 110 respectively; process the P-wave GMFP curves with moving average filters with window sizes of 30 and 50 respectively, to obtain the T-wave and P-wave GMFP curves after characteristic wave enhancement. S33, an adaptive single-threshold algorithm is used to locate the T-wave and P-wave GMFP curves after the feature wave enhancement. The adaptive single threshold is determined jointly by the T-wave and P-wave GMFP curves and the buffer threshold described in step S22.

[0010] Optionally, step S40 includes: Based on the location of each characteristic wave in the denoised magnetocardiogram, corresponding isomagnetic maps and pseudo current density maps are constructed. Time-domain signal feature sets, isomagnetic map feature sets, and pseudo current density map feature sets are constructed respectively, and feature scores for each characteristic wave are calculated. Z-scores and local outlier factors are used to identify outlier values ​​in the feature scores, thereby determining possible abnormal characteristic waves. The mean absolute error and mean correlation coefficient are calculated based on the effective R-wave peak detection results as evaluation indicators. If the abnormal characteristic wave position is contained within the abnormal cardiac cycle, it indicates that the abnormal characteristic wave is affected by the abnormal cardiac cycle and no correction is needed. If the cardiac cycle corresponding to the abnormal characteristic wave position is normal, it indicates that the characteristic wave detection result is abnormal and the result is discarded.

[0011] The beneficial effects of this invention are: This invention implements a method for detecting and evaluating characteristic waves of magnetic resonance imaging (MRI) signals based on adaptive threshold estimation. By combining adaptive threshold estimation and feature scoring, it achieves the detection, correction, and anomaly estimation of P waves, QRS complexes, and T waves in MRI signals. This invention fully utilizes the inherent advantages of MRI signals, enabling simultaneous detection of multi-channel characteristic waves, improving the accuracy and automation of MRI characteristic wave detection, and providing important guidance for clinical diagnosis and data analysis of MRI signals. Attached Figure Description

[0012] Figure 1 This is a flowchart of a method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation according to the present invention. Figure 2 The present invention provides the characteristic wave detection results of a method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation. Figure 3 The R-wave detection result is shown in the present invention, which describes a method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation. Figure 4 The Q-wave detection result is shown in the present invention, which describes a method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation. Figure 5 The result of S-wave detection is shown in the present invention, which describes a method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation. Figure 6 The result of T-wave detection is shown in the present invention, which describes a method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation. Figure 7 The result is the P-wave detection result of the magnetic field signal characteristic wave detection and evaluation method based on adaptive threshold estimation of the present invention. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0014] The present invention provides a method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation. The overall process is as follows: Figure 1 As shown, the localization results of this method on 36-channel magnetocardiogram signals are as follows: Figure 2 As shown, Figure 3 The results are from the R-wave detection. Figure 4 This is the result of Q-wave detection. Figure 5 The results are for S-wave detection. Figure 6 The results are from the T-wave detection. Figure 7 This is the result of the P-wave detection. Specifically, the implementation process is as follows: S10: Multi-channel magnetocardiogram (MCC) signals were acquired using an optically pumped magnetometer with 36 acquisition channels and a sampling rate of 1000 Hz. The optically pumped magnetometer is an extremely weak magnetic field measurement sensor based on spin-exchange without relaxation effect. The MCC signal was decomposed into wavelets using the Mallat pyramid algorithm with the wavelet basis function "Daubechies" selected. The decomposition level was 8. An unbiased risk estimation threshold was selected and a soft thresholding function was used for thresholding. The denoised MCC signal was obtained after wavelet reconstruction.

[0015] S20: Effective R-wave peak detection and correction are performed based on an adaptive multi-threshold method. The Q-wave and S-wave are located based on the detection results. The specific process is as follows: S21. Principal component analysis (PCA) was used to reduce the dimensionality of the denoised 36-channel magnetocardiogram (MCC) signal. The covariance matrix C of the MCC signal was calculated, and its eigenvalue matrix S and eigenvector matrix E were also calculated. The eigenvalues ​​in the eigenvalue matrix S were arranged in descending order. , where the eigenvalues The eigenvectors represent the principal component variances, and the eigenvectors represent the principal component directions. The cumulative variance contribution rate threshold is set to 90%. , Where M is the number of calculated eigenvalues, and N represents the number of eigenvalues ​​whose cumulative variance contribution rate is greater than 90% among the first N principal components. The first N principal components are used to generate the dimensionality-reduced magnetocardiogram signal. , in, This represents the reduced-dimensional magnetocardiogram signal. Represents eigenvalues The corresponding feature vectors are used to perform discrete wavelet transform on the dimension-reduced magnetocardiogram signal. The wavelet basis function is selected as "Daubechies", the decomposition level is 8, and the signal is reconstructed using the 5th and 6th level wavelet approximation coefficients for QRS group detection. S22, an adaptive multi-threshold algorithm is used to detect and correct the effective R-wave peak value of the reconstructed signal. The specific process is as follows: (1) Perform max-min normalization on the reconstructed signal: , Where x is the input signal and y is the normalized signal, the normalized magnetocardiogram signal is multiplied by itself, and then processed by a moving average filter with a window size of 150 to obtain the magnetocardiogram signal with enhanced R wave. (2) Threshold initialization is performed based on the first 10% of the length of the enhanced R-wave magnetic signal. If the time represented by the length of the initial data is greater than 10 seconds, then the data points of the first 10 seconds are used as the starting data. The adaptive threshold includes the signal threshold. Buffer threshold and noise threshold The signal threshold is 90% of the maximum value in the initial data, the buffer threshold is 70% of the average value of the initial data, and the noise threshold is 10% of the average value of the initial data. The RR interval estimation is initialized to 600. The RR interval estimation is an estimate of the effective R-wave peak interval of the magnetic resonance imaging (MRI) signal. The acceptable upper limit of the RR interval is 1.6 times the RR interval estimation, and the acceptable lower limit of the RR interval is 0.8 times the RR interval estimation. The acceptable upper and lower limits of the RR interval are used for adaptive correction of R-wave peak detection. The update strategy for the RR interval estimation is as follows: , in, , which represents the gain rate. This indicates the RR interval estimate. This represents the RR interval calculated based on the current effective R-wave peak value. This indicates the upper limit of acceptable RR intervals. This indicates the lower limit of the acceptable RR interval; The update strategy for the signal threshold is as follows: , The update strategy for the buffer threshold is as follows: , The update strategy for the noise threshold is as follows: , in, This represents the maximum value within the time window of a local peak point. This represents the average value within the time window of the local peak point. This indicates the location of a local peak point, and the time window ranges from 150 sampling points before the local peak point to 150 data points after it. (3) In the initial stage of the search, the signal threshold needs to be corrected according to the initial data to eliminate the interference of abnormal peak values; the initial stage algorithm uses 2 seconds as the local peak point time window and 1 second as the step size, and traverses the maximum value within each local peak point time window. The signal threshold is then corrected using the following strategy: , (4) Based on the enhanced R wave magnetic signal, use buffer thresholds to determine all local peak locations, which may include: effective R wave peak points, abnormal pacemakers, T waves or P waves. When a local peak point is detected, determine whether the local peak point is an effective R wave peak point based on the three types of thresholds and RR interval estimation, and execute an update strategy based on the determination result. (5) If the RR interval between the local peak point and the previous valid R-wave peak point is greater than the upper limit of the acceptable RR interval, it is determined that there may be a missed detection. In this case, the maximum value within a specific search window between the local peak point and the previous valid R-wave peak point is taken as the missed selection point. The specific search window is from 100 sampling points forward from the previous valid R-wave peak point to 100 sampling points backward from the local peak point. The 100 sampling points forward and backward represent the search refractory period, which is used to eliminate the problem that the missed detection point cannot be identified due to the large amplitude of the previous valid R-wave peak point or the local peak point. After determining the missed detection situation, the signal threshold and buffer threshold are first adjusted. The adjustment method is as follows: , , If the amplitude of the missed point is less than the noise threshold, it is determined to be a noise signal, and the noise threshold is updated accordingly. If the amplitude of the missed point is greater than the signal threshold, the RR interval between the missed point and the previous valid R-wave peak point is determined. If it is within the acceptable RR interval range, it is determined to be a valid R-wave peak point, and the signal threshold and RR interval estimate are updated. If it is greater than the upper limit of the acceptable RR interval, it means that there is another missed point before this missed point. This missed point is taken as a valid R-wave peak point, and the missed detection judgment steps are repeated. If it is less than the lower limit of the acceptable RR interval, it is determined to be a valid R-wave peak candidate point, and further judgment is required based on the subsequent feature extraction steps. If the amplitude of the missed detection point is between the signal threshold and the buffer threshold, it is necessary to determine whether it is a T-wave, P-wave, or R-wave. If the maximum slope value within the time window of the missed detection point is less than half of the maximum slope value within the time window of the previous valid R-wave peak point, it is determined to be a T-wave or P-wave, and the buffer threshold is updated. The time window is from 100 sampling points before the missed detection point to 100 sampling points after the missed detection point. Otherwise, continue to determine the RR interval between the missed detection point and the previous valid R-wave peak point. If the RR interval is within the acceptable RR interval range, it is determined to be a valid R-wave peak point, and the signal threshold and RR interval estimate are updated. If it is greater than the upper limit of the acceptable RR interval, it indicates that there is another missed detection point before this missed detection point. This missed detection point is taken as the valid R-wave peak point, and the judgment steps for missed detection are repeated. If it is less than the lower limit of the acceptable RR interval, it can be determined to be an abnormal pacemaker. If the amplitude of the missed point is between the buffer threshold and the noise threshold, update the buffer threshold and the noise threshold; If the RR interval between the local peak point and the previous valid R-wave peak point is less than the lower limit of the acceptable RR interval, then if the local peak point is less than the buffer threshold, the noise threshold and the buffer threshold are updated. If the local peak point is greater than the buffer threshold, then if the maximum slope within the time window of the local peak point is less than half of the maximum slope within the time window of the previous valid R-wave peak point, it is determined to be a T-wave or P-wave, and the buffer threshold is updated. Otherwise, it is determined to be a candidate point for a valid R-wave peak, and further judgment is required based on subsequent feature extraction steps. The time window is from the 80 sampling points before to the 80 sampling points after the local peak point. (6) If the local peak point is within the acceptable RR interval range, the category of the local peak point is determined according to the threshold. If the local peak point is greater than the signal threshold, it is determined to be a valid R-wave peak point and the signal threshold is updated. If the local peak point is between the signal threshold and the buffer threshold, if the maximum slope value within the time window of the local peak point is less than half of the maximum slope value within the time window of the previous valid R-wave peak point, it is determined to be a P-wave or T-wave and the buffer threshold is updated. The time window is from the 100 sampling points before the local peak point to the 100 sampling points after the local peak point. Otherwise, it is determined to be a valid R-wave peak point and the signal threshold and the buffer threshold are updated. If the local peak point is between the buffer threshold and the noise threshold, the buffer threshold and the noise threshold are updated. If the local peak point is less than the noise threshold, it is determined to be a noise signal and the noise threshold is updated. (7) After each threshold update, the correspondence between the three types of thresholds must be maintained: , For example, if after the threshold is updated, Then the values ​​of the signal threshold and the buffer threshold are swapped; In this invention, the accuracy of detecting valid R-wave peak points in magnetic resonance imaging (MRI) signals is ensured by introducing three types of adaptive thresholds and RR interval estimation. On one hand, the three types of adaptive thresholds include a signal threshold, a buffer threshold, and a noise threshold, which can be adjusted in real time based on the detection results. The signal threshold serves as the final criterion for determining valid R-wave peak points. The buffer threshold acts as a buffer region for the signal threshold, further judging local peak points when the input signal exhibits baseline drift, abnormal pacemakers, large fluctuations in R-wave peak points, or large T-wave amplitudes, thus improving the algorithm's robustness. The noise threshold serves as the criterion for judging noise signals, reducing the impact of baseline drift on the signal threshold and buffer threshold, ensuring the stability of signal threshold and buffer threshold updates, and allowing direct judgment of noise signals, thereby improving the algorithm's detection efficiency. On the other hand, the RR interval estimation serves as another important criterion for judging valid R-wave peak points, providing supplementary information for judging various local peak points, and enabling error checking and correction of the detection results, thus improving the algorithm's accuracy and flexibility. S23, Calculate the global mean field power (GMFP) curve of the denoised magnetocardiogram signal: , Where t represents the sampling point location, and C represents the number of channels. This represents the average value of all channels at position t. Based on the effective R-wave peak point position determined in S22, the first minimum value in the GMFP curve is searched forward as the starting point of the Q-wave search region, and the first minimum value in the GMFP curve is searched backward as the starting point of the S-wave search region. Then, based on the dimension-reduced magnetocardiogram signal, 20 data points are searched forward from the starting point of the Q-wave search region to find the maximum value in the Q-wave search region as the Q-wave position; 20 data points are searched backward from the starting point of the S-wave search region to find the maximum value in the S-wave search region as the S-wave position.

[0016] S30: Position detection and correction of T-wave and P-wave based on an adaptive single threshold method, the specific process is as follows: S31, Remove the QRS complex according to the result of step S20, and perform baseline correction on the denoised magnetic-cardiogram signal according to the position coordinates of the edge points of the removal window to obtain the characteristic curves of T wave and P wave respectively. For example, assuming two valid R-wave peaks are R1 and R2, and their corresponding Q-waves and S-waves are Q1, S1 and Q2, S2 respectively, the QRS complex removal window used for T-wave detection is from *a* sampling points before the Q-wave to *b* sampling points after the S-wave in the current QRS complex. The denoised magnetocardiogram signal within the removal window is zeroed out, and the remaining data area is the candidate region for the T-wave. Each T-wave candidate region window is represented as from *b* sampling points after S1 to *a* sampling points before Q2. The formula for establishing the correction line between the two points is: , Where t represents the sampling point location and y represents the sampling point amplitude. and The coordinates of the boundary points of the T-wave candidate region window are represented; after determining the correction line, the correction value of each sampling point in the T-wave candidate region is calculated according to the bilinear interpolation method, and the correction value is used to perform baseline correction on the denoised magnetocardiogram signal in the candidate region; The QRS complex removal window used for P-wave detection is from c sampling points before the Q-wave to d sampling points after the S-wave. Therefore, each P-wave candidate region window represents from d sampling points after S1 to c sampling points before Q2. The formula for establishing the correction line between these two points is: , a, b, c, and d are initialized to 120, 20, 20, and 300 respectively, and then fine-tuned based on the current effective R-wave peak value. The adjustment strategy is as follows: , The RR interval of the current candidate window. The estimated RR interval as described in S22; S32, calculate the GMFP curves of the T-wave and P-wave characteristic curves. The T-wave GMFP curve is processed by moving average filters with window sizes of 30 and 110 respectively; the P-wave GMFP curve is processed by moving average filters with window sizes of 30 and 50 respectively, to obtain the T-wave and P-wave GMFP curves after characteristic wave enhancement. S33. An adaptive single-threshold algorithm is used to locate the T-wave and P-wave GMFP curves after feature wave enhancement. Taking T-wave localization as an example, the specific process is as follows: (1) Initialize the threshold based on the first 10% of the length of the T-wave GMFP curve after feature wave enhancement. If the time represented by the length of the initial data is greater than 10 seconds, then the data points of the first 10 seconds are used as the starting data. Initial T-wave detection threshold Maximum value of the initial data 90% and the buffer threshold in step S22 Weighted combination: , (2) Traverse all valid R-wave peak points detected in step S22, and search for the local T-wave maxima and maxima of the GMFP curve in the T-wave candidate region after each valid R-wave peak point. According to the noise threshold in step S22 Exclude all smaller than The local T-wave maximum point; If the remaining local T-wave maximum point is unique and in the same location as the maximum point, it can be determined as a valid T-wave peak point; if the remaining local T-wave maximum point is not unique, the T-wave detection threshold is used to determine it. If there exists a local T-wave maximum greater than If so, it is determined to be a valid T-wave peak point; If there are multiple local T-wave maxima greater than Then update The update strategy is as follows: , in, This is the point with the largest amplitude among multiple local T-wave maxima. Then, the updated value is applied repeatedly to the remaining local T-wave maxima. Make judgments until the number of local T-wave maxima is equal to 1. If there is no local T-wave maximum greater than Then update The update strategy is as follows: , Then, following the steps described above, apply the updated method to the remaining local T-wave maxima. Make judgments until the number of local T-wave maxima is equal to 1. (3) If during iterative updates During the process, In this situation, at this time and Values ​​swapped. Updated simultaneously. Then repeat step (2) to re-evaluate the local T-wave maximum point; (4) Based on the detected effective T-wave peak points and effective P-wave peak points, further determine the start and end points of the T-wave and P-wave. The search window for the start and end points of the T-wave is e sampling points forward to f sampling points backward from the effective T-wave peak point; the search window for the start and end points of the P-wave is g sampling points forward to h sampling points backward from the effective P-wave peak point. The e, f, g, and h are initialized to 100, 100, 60, and 60, respectively. Taking the T wave as an example, the specific process is as follows: Based on the T-wave GMFP curve, search e sampling points forward from the effective T-wave peak point, and determine the first minimum point as the T-wave start point; search f sampling points backward from the effective T-wave peak point, and determine the first minimum point as the T-wave end point. If no minimum point is found within the search window, increase the number of sampling points before and after the search window by 10 until a minimum point meeting the composite conditions is found. If the search window increases forward beyond the first effective R-wave peak point before the effective T-wave peak point, or increases backward beyond the first effective R-wave peak point after the effective T-wave peak point, stop increasing the number of sampling points. At this point, start searching forward and backward simultaneously from the effective T-wave peak point until the amplitude drops to 10% of the amplitude of the T-wave peak point. Use this position as the T-wave start point and end point.

[0017] S40: Based on the characteristic wave detection results, feature extraction and quality assessment are performed on the magnetic cardiac signal to identify abnormal characteristic waves and abnormal cardiac cycles. The specific process is as follows: (1) Construct corresponding isomagnetic maps and pseudo current density maps based on the positions of each characteristic wave in the denoised magnetocardiogram signal, and construct time-domain signal feature sets respectively. Isomagnetic feature set and pseudo current density map feature set The time-domain signal feature set includes the RT interval, RS interval, QR interval, PR interval, T-wave duration, and P-wave duration; the isomagnetic feature set includes the isomagnetic dipole angle and isomagnetic dipole distance at the peak points of the P-wave, Q-wave, R-wave, S-wave, and T-wave, as well as the isomagnetic dipole angle change from the start point to the end point of the T-wave; the pseudo-current density map feature set includes the maximum current vector angle, maximum current vector amplitude, total current vector angle, and total current vector amplitude at the peak points of the P-wave, Q-wave, R-wave, S-wave, and T-wave, as well as the average maximum current vector from the start point to the end point of the T-wave and the average maximum current vector from the start point to the end point of the P-wave. Weighted values ​​for the feature sets are calculated for each characteristic wave. , Where i represents the category of each characteristic wave in each cardiac cycle, N represents the number of time-domain features calculated for the corresponding characteristic wave i, M represents the number of isomagnetic features calculated for the corresponding characteristic wave i, and P represents the number of pseudo-current density map features calculated for the corresponding characteristic wave i. To detect the feature scores of all feature waves contained in feature wave i; The feature waves include effective peak points and candidate points. When the acquired multi-channel magnetic resonance imaging signal has Q complete cardiac cycles, ideally the number of all feature waves is Q. If the number of feature waves is less than Q, the initial parameters of the algorithm of this invention need to be fine-tuned; if it is greater than Q, redundant feature wave points need to be deleted in combination with feature scoring. (2) Use the Z-score and Local Outlier Factor (LOF) combined index to identify outliers in the feature set and correct the detection results. The final detection results must ensure that the number of all feature waves is consistent or the difference in number does not exceed 1. For the feature score of feature wave i The Z-score is calculated as follows: , for The mean and standard deviation. When When this characteristic wave is considered as a possible outlier in characteristic wave i, where b represents the b-th characteristic wave in characteristic wave i; the local outlier factor is calculated as follows: , express neighborhood Other local density and The average of the ratios of the locally reachable densities is used to calculate the local outlier factor for all feature scores of feature wave i. If such outliers exist... If this characteristic wave is considered as a possible outlier in characteristic wave i; When both the Z-score and the local outlier factor index are used to identify a possible outlier, the position of the feature wave corresponding to the outlier is first fine-tuned. The fine-tuning window is from 5 sampling points forward to 5 sampling points backward from the position of the feature wave. The fine-tuned feature score is then updated. If the outlier disappears at this time, the detection result of this feature wave is retained. If the feature score corresponding to this feature wave is still an outlier, the outlier result is retained for further judgment. (3) Based on the location of the abnormal value corresponding to the characteristic wave, on the one hand, it is necessary to determine whether it is caused by the characteristic wave detection algorithm, and on the other hand, it is necessary to determine whether it is caused by the abnormal cardiac cycle of the collector. Therefore, the average absolute error and average correlation coefficient of all cardiac cycles are calculated based on the effective R wave peak detection results as evaluation indicators. The abnormality of the cardiac cycle is judged based on the evaluation indicators in step (2). If the abnormal characteristic wave location selected in step (2) is included in the abnormal cardiac cycle, it means that the abnormal characteristic wave is affected by the abnormal cardiac cycle and does not need to be corrected. If the cardiac cycle corresponding to the abnormal characteristic wave location is normal, it means that the characteristic wave detection result is abnormal and the result is removed.

[0018] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting and evaluating characteristic waves of magnetic cardiomyocyte signals based on adaptive threshold estimation, characterized in that, Based on the characteristics of multi-channel magnetic resonance imaging (MRI) signals, the system automatically detects, corrects, and estimates anomalies in P waves, QRS complexes, and T waves, specifically including the following steps: S10: Use sensors to acquire multi-channel magnetocardiogram signals and remove high-frequency interference and baseline drift; S20: R-wave peak detection and correction are performed based on an adaptive multi-threshold method, and Q-wave and S-wave are located based on the detection results. The adaptive multi-threshold method includes signal threshold, buffer threshold and noise threshold. S30: T-wave and P-wave position detection and correction based on adaptive single threshold method; S40: Based on the characteristic wave detection results, perform feature extraction and quality assessment on the magnetic heart signal to identify abnormal characteristic waves and abnormal cardiac cycles.

2. The method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation according to claim 1, characterized in that, Step S10 includes: Multi-channel resting-state magnetocardiogram (MCC) signals were acquired using an optically pumped magnetometer. High-frequency interference and baseline drift in the MCC signals were removed using a wavelet thresholding denoising method to obtain the denoised MCC signals.

3. The method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation according to claim 2, characterized in that, Step S20 includes: S21. Principal component analysis (PCA) is used to reduce the dimensionality of the denoised magnetic-cardiogram (MCC) signal. The first m principal components are selected based on the cumulative variance contribution rate threshold of 90%, and the data is reconstructed according to the weight of the cumulative variance contribution rate to obtain the dimensionality-reduced MCC signal. The dimensionality-reduced MCC signal is then subjected to 8-level wavelet decomposition, and the signal is reconstructed using the 5th and 6th level wavelet approximation coefficients for R-wave peak detection. S22, the reconstructed signal used for R-wave peak detection is dot-producted with itself, and processed using a moving average filter with a window size of 150 to obtain the enhanced R-wave magnetocardiogram (MCC) signal. An adaptive multi-threshold algorithm is used to detect and correct the effective R-wave peaks in the dimensionality-reduced MCC signal. The adaptive multi-threshold algorithm includes a signal threshold, a buffer threshold, and a noise threshold. RR interval estimation is introduced as supplementary information to constrain the detection results. The RR interval estimation is an estimate of the interval between effective R-wave peak points in the MCC signal. The signal threshold serves as the final criterion for determining effective R-wave peak points. The buffer threshold serves as a buffer region for the signal threshold, further judging local peak points when the input signal exhibits baseline drift, abnormal pacemakers, large fluctuations in R-wave peak points, or large T-wave amplitudes. The noise threshold serves as a criterion for judging noise signals, reducing the impact of baseline drift on the signal threshold and buffer threshold. The RR interval estimation serves as one of the criteria for judging effective R-wave peak points, providing supplementary information for judging various local peak points, and simultaneously performing error checking and correction of the detection results. S23. Calculate the global average field power (GMFP) curve of the denoised magnetic field signal. Determine the search regions for Q and S waves based on the minimum points of the GMFP curve. Determine the positions of Q and S waves based on the maximum points of the dimensionality-reduced magnetic field signal.

4. The method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation according to claim 3, characterized in that, Step S30 includes: S31, Remove the QRS complex based on the calculation results of step S23, and perform baseline correction on the denoised magnetocardiogram signal based on the position coordinates of the edge points of the removal window to obtain the characteristic curves of the T wave and P wave respectively. S32, calculate the GMFP curves of the T-wave and P-wave characteristic curves. The T-wave GMFP curve is processed by moving average filters with window sizes of 30 and 110, and the P-wave GMFP curve is processed by moving average filters with window sizes of 30 and 50, to obtain the T-wave and P-wave GMFP curves after characteristic wave enhancement. S33, use an adaptive single threshold algorithm to locate the T-wave and P-wave GMFP curves after the feature wave enhancement. The adaptive single threshold is determined by the T-wave and P-wave GMFP curves and the buffer threshold mentioned in step S22.

5. The method for detecting and evaluating characteristic waves of magnetic field signals based on adaptive threshold estimation according to claim 4, characterized in that, Step S40 includes: Based on the location of each characteristic wave in the denoised magnetocardiogram, corresponding isomagnetic maps and pseudo current density maps are constructed. Time-domain signal feature sets, isomagnetic map feature sets, and pseudo current density map feature sets are constructed respectively, and feature scores for each characteristic wave are calculated. Z-scores and local outlier factors are used to identify outlier values ​​in the feature scores, thereby determining possible abnormal characteristic waves. The mean absolute error and mean correlation coefficient are calculated based on the effective R-wave peak detection results as evaluation indicators. If the abnormal characteristic wave position is contained within the abnormal cardiac cycle, it indicates that the abnormal characteristic wave is affected by the abnormal cardiac cycle and no correction is needed. If the cardiac cycle corresponding to the abnormal characteristic wave position is normal, it indicates that the characteristic wave detection result is abnormal and the result is discarded.

Images