A magnetocardiogram-based physiological abnormality parameter detection method and system

Abstract

The application discloses a physiological abnormality parameter detection method and system based on a magnetocardiogram, and comprises the following steps: collecting multi-channel magnetocardiogram data, recording an environmental reference magnetic field, and preprocessing to obtain a space-time magnetocardiogram tensor and a signal-to-noise ratio evaluation sequence; performing multivariate variation mode decomposition on the space-time magnetocardiogram tensor and introducing a space correlation constraint to obtain a multi-mode tensor component set, constructing a heart source mode subset, and weighting reconstruction to obtain a denoising magnetocardiogram tensor; inversely solving an equivalent current dipole trajectory based on the denoising magnetocardiogram tensor and calculating an anatomic conduction consistency index, and adjusting a bandwidth constraint parameter in a closed loop; performing improved semi-classical signal analysis on the denoising magnetocardiogram tensor and the equivalent current dipole trajectory to obtain a semi-classical eigenvalue sequence and a semi-classical spectral feature set, and constructing a phase space manifold to output a detection result and an abnormality severity evaluation result. The application realizes detection and evaluation of physiological abnormality parameters in a low signal-to-noise ratio environment.

Description

Technical Field

[0001] This invention relates to the field of medical signal processing technology, and in particular to a method and system for detecting abnormal physiological parameters based on magnetocardiography. Background Technology

[0002] Magnetocardiography (MCG) technology utilizes the magnetic field information generated by the heart's electrical activity to achieve contactless measurement. It features rich spatial distribution information and low dependence on surface electrodes, and has been researched and experimentally applied in areas such as arrhythmia screening, myocardial ischemia assessment, and conduction abnormality analysis. Current engineering implementations often employ multi-channel magnetic sensing arrays to acquire MCG signals, simultaneously acquiring an environmental reference magnetic field to suppress external magnetic disturbances. In the signal processing chain, common approaches include temporal baseline correction and drift cancellation, adaptive noise cancellation based on the reference channel, spatial filtering and beamforming, statistical separation and subspace projection, as well as non-stationary decomposition methods such as wavelet decomposition, empirical mode decomposition, and variational mode decomposition to obtain MCG timing data with higher signal-to-noise ratios.

[0003] Regarding anomaly parameter extraction, existing solutions primarily extract features such as peak and trough amplitudes, time intervals, frequency band energy, phase synchronization, and nonlinear complexity from the denoised magnetocardiogram (MCC) time series. These features are then combined with threshold discrimination, template matching, or data-driven classification models to output the anomaly type and severity. Some solutions employ inversion to solve for the equivalent current dipole trajectory to locate cardiac activity and analyze the conduction path. However, MCC signals naturally have weak amplitudes and are easily affected by magnetic drift from electrical equipment, transportation vehicles, and the environment. Spatial correlations and spatial gradient differences exist between multiple channels. If decomposition and reconstruction are performed using only a single-channel or weakly spatially constrained approach, over-filtering, aliasing of cardiac components, or amplification of residual environmental noise can easily occur, leading to unstable anomaly features.

[0004] Individual anatomical differences and changes in cardiac rhythm cause feature distribution to drift significantly over time and across populations. Fixed parameter decomposition and filtering are difficult to ensure robustness in different scenarios. Data-driven models are sensitive to the size and distribution consistency of labeled samples, limiting their generalization and interpretability. Schemes based on dipole inversion, if lacking consistency verification with the physical path of the cardiac conduction system and closed-loop parameter adjustment mechanisms, are difficult to distinguish between real cardiac signals and algorithmic reconstruction artifacts, thus affecting the reliability and accuracy of engineered output.

[0005] Therefore, how to provide a method and system for detecting abnormal physiological parameters based on magnetocardiography is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method and system for detecting abnormal physiological parameters based on magnetocardiography. This invention achieves high-accuracy and robust detection and evaluation under low signal-to-noise ratio and strong interference by combining multivariate variational mode decomposition and anatomical closed-loop constraints with semiclassical spectral topology discrimination.

[0007] A method for detecting abnormal physiological parameters based on magnetocardiography according to an embodiment of the present invention includes:

[0008] Multi-channel magnetocardiogram data of the subjects were collected, the spatial coordinate parameters of the corresponding channels and the environmental reference magnetic field were recorded, and preprocessing was performed to obtain the spatial-temporal magnetocardiogram tensor and signal-to-noise ratio evaluation sequence.

[0009] Multivariate variational mode decomposition of the space-time magnetocardiogram tensor yields a set of multimodal tensor components;

[0010] A subset of cardiac modalities is constructed based on the set of multimodal tensor components and the spatial coordinate parameters of the channels;

[0011] A weighted reconstruction of a subset of cardiac modalities is performed, and the reconstruction boundary is constrained to obtain a denoised spatial-temporal cardiac magnetotensor.

[0012] The equivalent current dipole trajectory is derived by inverse calculation of the denoised space-time magnetic cardiomyocyte tensor, and the anatomical conduction consistency index is calculated to obtain the target denoised space-time magnetic cardiomyocyte tensor.

[0013] By processing the target noise reduction space-time cardiomagnetic tensor and equivalent current dipole trajectory using an improved semiclassical signal analysis algorithm, a semiclassical eigenvalue sequence and a semiclassical spectral feature set are obtained.

[0014] A high-dimensional phase space manifold is constructed based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set. The phase space topological feature parameters are calculated, and the results of physiological abnormality parameter detection and abnormality severity assessment are output according to the structural changes of the phase space topological feature parameters.

[0015] Optionally, the preprocessing specifically includes:

[0016] The multi-channel magnetocardiogram (MCC) data was timestamped and aligned with the environmental reference magnetic field. Cross-channel phase consistency calibration and alignment resampling were then performed on the multi-channel MCC data to obtain a time-aligned multi-channel MCC sequence.

[0017] The drift baseline is determined based on the environmental reference magnetic field, and the baseline drift is estimated and drift is canceled on the time-aligned multi-channel magnetocardiogram sequence to obtain the drift-free multi-channel magnetocardiogram sequence.

[0018] Based on the spatial coordinate parameters of each channel, a spatial smoothing constraint is established, and under the spatial smoothing constraint, the drift-free multi-channel magnetocardiogram sequence is canceled by the environmental reference magnetic field to obtain a multi-channel magnetocardiogram sequence free of environmental interference.

[0019] The channel amplitudes of the multi-channel magnetocardiogram (MCC) sequence were made uniform to remove environmental interference, and the initial spatial-temporal MCC tensor corresponding to the channel dimension and time dimension was established based on the spatial coordinate parameters of each channel.

[0020] Based on the initial space-time magnetic cardiomyocyte tensor, the channel energy stability characterization and residual volatility characterization are calculated within a preset sliding window, and the signal-to-noise ratio evaluation sequence is calculated based on the channel energy stability characterization and residual volatility characterization.

[0021] Based on the signal-to-noise ratio (SNR) evaluation sequence, low SNR segments in the initial spatial-temporal magnetic-cardiogram (MCC) tensor are identified. Under cross-channel consistency constraints, tensor completion and consistency correction are performed on the low SNR segments to form the spatial-temporal MCC tensor.

[0022] Optionally, the multivariate variational mode decomposition specifically includes:

[0023] A Hilbert transform is performed along the time dimension of the spacetime magnetocardiogram tensor, followed by energy normalization, to obtain the normalized complex analytic spacetime magnetocardiogram tensor.

[0024] The distance matrix between channels is calculated based on the spatial coordinate parameters of each channel, and a spatial correlation constraint matrix is ​​constructed.

[0025] An initial set of modal component values ​​is established based on a preset number of modes, and the preset bandwidth constraint parameters are adaptively adjusted based on the signal-to-noise ratio evaluation sequence to obtain adaptive bandwidth constraint parameters.

[0026] An augmented optimization objective function is constructed based on the normalized complex analytic space-time magnetocardiogram tensor, the initial set of modal components, the spatial correlation constraint matrix, and the adaptive bandwidth constraint parameters.

[0027] The augmented optimization objective function is iteratively updated in alternating directions in the frequency domain, and cross-channel spatial consistency constraints are applied to the modal components based on the spatial correlation constraint matrix in each update to obtain the set of frequency domain modal components.

[0028] By performing an inverse transformation and consistency constraints on the frequency domain modal component set, a multimodal tensor component set is obtained.

[0029] Optionally, the subset of cardiac modalities specifically includes:

[0030] Energy normalization is performed on the multimodal tensor component set, and the corresponding modal time series are extracted in the time dimension to obtain the modal space energy distribution vector set and the modal time series set;

[0031] Based on the spatial coordinate parameters of each channel, a set of candidate dipole positions is established in the preset thoracic cavity neighborhood, and a set of candidate dipole spatial response templates is generated based on the set of candidate dipole positions and the spatial coordinate parameters of each channel.

[0032] Channel-level matching calculations are performed on the modal space energy distribution vector set and the candidate dipole space response template set to obtain the spatial dipole consistency index sequence, and the corresponding optimal candidate dipole position identifier is determined to obtain the spatial dipole consistency result set.

[0033] For each modal time series in the modal time series set, an amplitude envelope sequence and a main period interval sequence are constructed. The envelope stability characterization quantity and the period stability characterization quantity are calculated within a preset sliding window to obtain the temporal cardiac origin consistency index sequence.

[0034] The screening thresholds for the spatial dipole consistency index sequence and the temporal cardiac source consistency index sequence are adjusted based on the signal-to-noise ratio evaluation sequence to obtain an adaptive screening threshold set.

[0035] Under the constraint of the adaptive screening threshold set, the multimodal tensor components that simultaneously satisfy the screening conditions of the spatial dipole consistency result set and the temporal cardiac source consistency index sequence are selected as the cardiac source mode subset.

[0036] Optionally, the weighted reconstruction of the subset of cardiac modalities and the application of continuity constraints to the reconstruction boundary specifically involve:

[0037] Energy normalization is performed on each multimodal tensor component in the cardiac modality subset, and credibility is calibrated based on the signal-to-noise ratio evaluation sequence to obtain a set of modal credibility weight sequences.

[0038] A spatial consistency constraint weight matrix is ​​established based on the spatial correlation constraint matrix, and cross-channel phase consistency correction is performed on each multimodal tensor component in the cardiac modality subset to obtain a set of spatial consistency corrected multimodal tensor components.

[0039] The set of multimodal tensor components for spatial consistency correction is jointly weighted based on the set of modal confidence weight sequences and the weight matrix of spatial consistency constraints to obtain the set of multimodal tensor components with joint weights.

[0040] The initial reconstructed spatial-temporal magnetocardiogram tensor is obtained by weighted superposition and reconstruction of the joint weighted multimodal tensor component set.

[0041] Within a preset sliding window, the amplitude envelope continuity constraint and phase continuity constraint are applied to the adjacent window boundaries of the initial reconstructed spatial-temporal magnetic-oscelematic tensor, and the windows are spliced ​​to obtain the boundary-continuous reconstructed spatial-temporal magnetic-oscelematic tensor.

[0042] Cross-channel consistency correction is performed on the boundary continuously reconstructed spatial-temporal magnetic-cardiogram tensor under the constraint of the spatial consistency constraint weight matrix to obtain the denoised spatial-temporal magnetic-cardiogram tensor.

[0043] Optionally, the step of inversely calculating the equivalent current dipole trajectory based on the denoised space-time magnetic cardiomyocyte tensor and calculating the anatomical conduction consistency index specifically involves:

[0044] The equivalent current dipole forward magnetic field response relationship is established based on the noise-reduced space-time magnetic tensor and the spatial coordinate parameters of each channel, and the serialized inversion solution is performed along the time dimension to obtain the initial equivalent current dipole trajectory.

[0045] In the preset cardiac anatomical coordinate system, the conduction center line of the His bundle-Purkinje fiber is represented as a set of directed physical paths with bifurcations, and the conduction direction vector and path length parameters are labeled to obtain the His bundle-Purkinje fiber physical path diagram.

[0046] Based on the dipole position at each moment of the initial equivalent current dipole trajectory, the path projection matching is performed with the His bundle-Purkinje fiber physical path map to obtain the path projection point sequence.

[0047] A path travel sequence is generated along the path projection point sequence and the path length parameter conduction direction, and a displacement direction sequence is generated based on the initial equivalent current dipole trajectory.

[0048] Calculate the anatomical conduction consistency index based on the path travel sequence, displacement direction sequence, and conduction direction vector;

[0049] When the anatomical conduction consistency index does not meet the preset threshold, the preset bandwidth constraint parameters are adjusted and the multivariate variational mode decomposition, the cardiac source mode subset and weighted reconstruction are re-executed to obtain the target noise-reduced spatial-temporal cardiac magnetic tensor and the equivalent current dipole trajectory that pass through the anatomical constraint.

[0050] Optionally, the improved semi-classical signal analysis algorithm is specifically as follows:

[0051] Based on the equivalent current dipole trajectory and the spatial coordinate parameters of each channel, the target noise reduction spatial-temporal magnetic-cardiogram tensor is weighted and aggregated across channels to obtain the heart source aggregated magnetic-cardiogram time series.

[0052] The amplitude envelope sequence and local curvature sequence were calculated based on the heart-source aggregated magnetocardiogram time series, and normalized and calibrated to obtain the envelope potential construction sequence and curvature potential construction sequence.

[0053] Construct an adaptive semiclassical potential function sequence based on the envelope potential and curvature potential sequences;

[0054] Construct a dynamically coupled analog of Planck's constant;

[0055] A semiclassical spectral operator is constructed based on an adaptive semiclassical potential function sequence and a dynamically coupled Planck constant analog quantity. The semiclassical spectral operator is then discretized in the time dimension to form a discrete representation of the spectral operator.

[0056] The discrete representation of the spectral operator is decomposed to obtain a semiclassical eigenvalue sequence, and the semiclassical spectral feature set is calculated based on the semiclassical eigenvalue sequence and the adaptive semiclassical potential function sequence.

[0057] Optionally, the step of outputting physiological abnormality parameter detection results and abnormality severity assessment results based on structural changes in phase space topological feature parameters specifically includes:

[0058] Based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set, a phase space state vector sequence is constructed under a preset delay dimension and a preset embedding dimension, and then scale normalization and coordinate consistency are performed.

[0059] A high-dimensional phase space manifold is constructed based on the phase space state vector sequence, and local neighborhood graph connectivity and geodesic distance consistency correction are performed to obtain a measurable high-dimensional phase space manifold.

[0060] Based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set, measurable high-dimensional phase space manifolds are segmented in stages to obtain depolarized phase space submanifolds and complex polarized phase space submanifolds.

[0061] The set of topological characteristic parameters of phase space is calculated based on depolarized phase space submanifolds and complex polarized phase space submanifolds, respectively.

[0062] The topological structure change is calculated based on the set of phase space topological characteristic parameters, and the electrophysiological instability index is calculated based on the topological structure change.

[0063] The results of physiological abnormality parameter detection and abnormality severity assessment are output based on the electrophysiological instability index.

[0064] Optionally, a system for detecting abnormal physiological parameters based on magnetocardiography includes:

[0065] The magnetocardiogram (MCC) acquisition unit is used to acquire multi-channel MCC data of the subject and record spatial coordinate parameters and environmental reference magnetic field.

[0066] The preprocessing unit preprocesses the multi-channel magnetocardiogram data and the environmental reference magnetic field, and outputs time-aligned magnetocardiogram tensor and signal-to-noise ratio evaluation sequences.

[0067] The mode decomposition unit performs multivariate variational mode decomposition on the space-time magnetocardiogram tensor, outputs a set of multimodal tensor components, and calculates the correlation constraint matrix;

[0068] The cardiac modality subset construction unit constructs cardiac modality subsets based on modal components and spatial coordinate parameters, and performs energy normalization, confidence calibration, and consistency correction.

[0069] The weighted reconstruction unit performs weighted reconstruction on a subset of cardiac modalities, applies boundary continuity constraints, and outputs a denoised space-time cardiac magnetic tensor.

[0070] The equivalent current dipole inversion unit, based on the denoised space-time magnetic cardiotensor and spatial coordinate parameters, inversely calculates the equivalent current dipole trajectory and the anatomical conduction consistency index.

[0071] The bandwidth constraint adjustment unit adjusts the bandwidth constraint parameters and triggers multivariate variational mode decomposition and cardiac modality subset reconstruction when the anatomical conduction consistency index does not meet the threshold.

[0072] The semiclassical signal analysis unit performs cross-channel weighted aggregation based on the equivalent current dipole trajectory and spatial coordinate parameters to generate a cardiac source aggregated time series, and performs envelope and curvature calculations.

[0073] The phase space manifold and topology analysis unit constructs a high-dimensional phase space manifold based on the eigenvalue sequence and spectral feature set, calculates topological feature parameters, and outputs physiological abnormality detection results and abnormality severity assessment.

[0074] The beneficial effects of this invention are:

[0075] (1) This invention collects multi-channel magnetocardiogram data and records the environmental reference magnetic field simultaneously. It combines timestamp alignment, cross-channel phase consistency calibration, drift cancellation, environmental reference magnetic field cancellation under spatial smoothing constraints, and tensor completion repair and consistency correction of low signal-to-noise ratio segments to form a spatial-temporal magnetocardiogram tensor and signal-to-noise ratio evaluation sequence. This achieves robust preprocessing of low signal-to-noise ratio magnetocardiogram signals, effectively improves the signal purity and cross-channel consistency of subsequent decomposition and reconstruction, and enhances the detection stability under strong environmental magnetic interference scenarios.

[0076] (2) This invention introduces multi-channel spatial correlation constraints in multivariate variational mode decomposition and adaptively adjusts the bandwidth constraint parameters according to the signal-to-noise ratio evaluation sequence. It constructs a subset of cardiac source modes by combining spatial dipole consistency and temporal cardiac source consistency, and achieves joint weighted reconstruction and reconstruction boundary continuity constraints through confidence calibration and spatial consistency constraint weight matrix. This results in a denoised spatial-temporal cardiac magnetic tensor, which realizes the directional preservation of cardiac source components and effective suppression of noise components. It significantly improves the robustness and repeatability of mode decomposition and reconstruction results under individual differences and rhythm changes.

[0077] (3) The improved semiclassical signal analysis algorithm in this invention constructs an adaptive semiclassical potential function based on the amplitude envelope sequence and the local curvature sequence, and constructs a dynamically coupled Planck constant analog quantity to form a discrete representation of the semiclassical spectral operator and performs eigenvalue decomposition. Compared with the semiclassical analysis method where the Planck constant analog quantity is fixed and the potential function is weakly correlated with the physiological process, it realizes the adaptive characterization of the changes in cardiac electrophysiological state by the semiclassical spectral features. At the same time, this invention constructs a high-dimensional phase space manifold based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set and calculates the phase space topological feature parameters. It uses the change in topological structure to form an electrophysiological instability index, thereby improving the abnormality discrimination from a single numerical deviation to a characterization of the structural evolution of the depolarization and repolarization process, which significantly improves the engineering interpretability of the abnormality detection results and the consistency of severity assessment. Attached Figure Description

[0078] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0079] Figure 1 This is a flowchart of a method for detecting abnormal physiological parameters based on magnetocardiography proposed in this invention;

[0080] Figure 2 This is a data flow diagram of an improved semi-classical signal analysis algorithm for a method of detecting abnormal physiological parameters based on magnetocardiography proposed in this invention.

[0081] Figure 3 This is a structural block diagram of a physiological abnormality parameter detection system based on magnetocardiography proposed in this invention. Detailed Implementation

[0082] Example 1: Reference Figures 1-3 A method for detecting abnormal physiological parameters based on magnetocardiography, comprising:

[0083] Multi-channel magnetocardiogram data of the subjects were collected, the spatial coordinate parameters of the corresponding channels and the environmental reference magnetic field were recorded, and preprocessing was performed to obtain the spatial-temporal magnetocardiogram tensor and signal-to-noise ratio evaluation sequence.

[0084] Multivariate variational mode decomposition is performed on the space-time magnetocardiogram tensor. Under the constraint of shared center frequency of modal components, multi-channel spatial correlation constraint is introduced and preset mode number and bandwidth constraint parameters are used to obtain the set of multimodal tensor components.

[0085] A subset of cardiac modalities is constructed based on the set of multimodal tensor components and the spatial coordinate parameters of the channels;

[0086] A weighted reconstruction of a subset of cardiac modalities is performed, and the reconstruction boundary is constrained to obtain a denoised spatial-temporal cardiac magnetotensor.

[0087] The equivalent current dipole trajectory is inversely derived from the denoised space-time magnetic tensor, and the anatomical conduction consistency index is calculated. When the anatomical conduction consistency index does not meet the preset threshold, the bandwidth constraint parameters are adjusted and the multivariate variational mode decomposition, cardiac source mode and weighted reconstruction are re-executed to obtain the target denoised space-time magnetic tensor that passes the anatomical constraint.

[0088] An improved semiclassical signal analysis algorithm is used to process the target denoising space-time cardiomagnetic tensor and the equivalent current dipole trajectory. An adaptive semiclassical potential function sequence is constructed based on the amplitude envelope and local curvature of the target denoising space-time cardiomagnetic tensor. The Planck constant analog quantity is set as the function mapping of the signal-to-noise ratio evaluation sequence and a dynamic coupling relationship is established with the conduction velocity characterization quantity and magnetic field attenuation characterization quantity corresponding to the equivalent current dipole trajectory to obtain the semiclassical eigenvalue sequence and the semiclassical spectral feature set.

[0089] A high-dimensional phase space manifold is constructed based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set. The phase space topological feature parameters are calculated, and the results of physiological abnormality parameter detection and abnormality severity assessment are output according to the structural changes of the phase space topological feature parameters.

[0090] In this embodiment, the preprocessing specifically includes:

[0091] The multi-channel magnetocardiogram (MCC) data was timestamped and aligned with the environmental reference magnetic field. Cross-channel phase consistency calibration and alignment resampling were then performed on the multi-channel MCC data to obtain a time-aligned multi-channel MCC sequence.

[0092] The drift baseline is determined based on the environmental reference magnetic field, and the baseline drift is estimated and drift is canceled on the time-aligned multi-channel magnetocardiogram sequence to obtain the drift-free multi-channel magnetocardiogram sequence.

[0093] Based on the spatial coordinate parameters of each channel, a spatial smoothing constraint is established, and under the spatial smoothing constraint, the drift-free multi-channel magnetocardiogram sequence is canceled by the environmental reference magnetic field to obtain a multi-channel magnetocardiogram sequence free of environmental interference.

[0094] Among them, spatial smoothing constraint refers to: determining the spatial proximity relationship between channels based on the spatial coordinate parameters of each channel, and establishing smoothing conditions to constrain the spatial continuity between channels, so that the changes in the spatial dimension of multi-channel magnetocardiogram data at the same time point meet the continuity requirement matching the spatial distance between channels, avoiding abrupt changes in space that are disproportionate to the distance between channels; when performing environmental reference magnetic field cancellation on the drift-free multi-channel magnetocardiogram sequence, the smoothing condition is applied as a constraint to the cancellation process of each channel, so that the canceled multi-channel sequence maintains a consistent smooth structure in the spatial dimension, thereby suppressing the spatially discontinuous environmental interference components and maintaining the structural consistency of cardiac activity in the spatial dimension;

[0095] The channel amplitudes of the multi-channel magnetocardiogram (MCC) sequence were made uniform to remove environmental interference, and the initial spatial-temporal MCC tensor corresponding to the channel dimension and time dimension was established based on the spatial coordinate parameters of each channel.

[0096] Based on the initial space-time magnetic cardiomyocyte tensor, the channel energy stability characterization and residual volatility characterization are calculated within a preset sliding window, and the signal-to-noise ratio evaluation sequence is calculated based on the channel energy stability characterization and residual volatility characterization.

[0097] In this embodiment 1, the initial spatial-temporal magnetic tensor is windowed along the time axis within a preset sliding window. For each channel, the amplitude of each sampling point in the channel amplitude sequence within each window is squared and accumulated to obtain the window energy value. The window energy values ​​of the same channel in consecutive windows constitute the window energy sequence. The energy change amplitude of adjacent windows is calculated for the window energy sequence, and the energy change amplitude is normalized according to the sliding mean of the window energy sequence within several adjacent windows to obtain the channel energy stability characterization quantity. Within the same sliding window, the sliding mean trend of the channel amplitude sequence of each channel is calculated. The sequence is processed by subtracting the channel amplitude sequence from the moving average trend sequence to obtain the residual sequence. The root mean square value of the residual sequence is calculated and normalized according to the corresponding window energy value to obtain the residual volatility characterization. The channel energy stability characterization and residual volatility characterization of each channel within the same window are respectively converged across channels to form window-level energy stability index and window-level residual volatility index. The window-level energy stability index and window-level residual volatility index are then fused according to preset weights and converted into window-level signal-to-noise ratio evaluation value through preset mapping rules. The values ​​are then arranged in the sliding window order to form the signal-to-noise ratio evaluation sequence.

[0098] Based on the signal-to-noise ratio (SNR) evaluation sequence, low SNR segments in the initial spatial-temporal magnetic-cardiogram (MCC) tensor are identified. Under cross-channel consistency constraints, tensor completion and consistency correction are performed on the low SNR segments to form the spatial-temporal MCC tensor.

[0099] In this embodiment 1, within the low signal-to-noise ratio (SNR) segment determined by the SNR evaluation sequence, the corresponding channel-dimensional and time-dimensional tensor segments in the initial spatial-temporal magnetocardiogram (MTC) tensor are marked as low-confidence data segments, and these low-confidence data segments are considered as segments to be repaired while retaining their position indices in the tensor. Subsequently, using the non-low-confidence data segments in the initial spatial-temporal MTC tensor as reference segments, the amplitude coordination relationship between channels in the reference segment at the same time dimension is extracted to form a channel-related structure representation corresponding to the cross-channel consistency constraint, and this channel-related structure representation is used as a constraint condition for the segment to be repaired. In the tensor completion and repair stage, based on the time-neighborhood segments on both sides of the segment to be repaired and the channel-related structure representation of the reference segment, the segment to be repaired is iterated. In the completion process, each iteration first generates candidate completion fragments for the segment to be repaired based on the continuity of the temporal neighborhood. Then, the candidate completion fragments are projected onto the channel consistency structure defined by the channel-related structure representation in the channel dimension, thereby suppressing isolated mutations in a single channel and maintaining multi-channel coordinated changes. In the consistency correction stage, the consistency of the related structure in the channel dimension between the completed segment to be repaired and the reference segment is compared. Amplitude and phase corrections are performed on channel components that deviate from the channel consistency structure. At the boundary of the segment to be repaired, a continuous transition method is used to eliminate splicing discontinuities. Finally, the repaired segment and the adjacent high-confidence segment simultaneously satisfy cross-channel consistency constraints in both the channel-dimensional related structure and the temporal continuity, thereby forming a spatial-temporal magnetic tensor.

[0100] In this embodiment, the multivariate variational mode decomposition specifically refers to:

[0101] A Hilbert transform is performed along the time dimension of the spacetime magnetocardiogram tensor, followed by energy normalization, to obtain the normalized complex analytic spacetime magnetocardiogram tensor.

[0102] The distance matrix between channels is calculated based on the spatial coordinate parameters of each channel, and a spatial correlation constraint matrix is ​​constructed.

[0103] In this embodiment 1, when constructing the spatial correlation constraint matrix, the spatial coordinate parameters of each channel are used as the only input. First, the distance matrix between channels is calculated using the spatial coordinate parameters of any two channels. Then, the distance values ​​in the distance matrix between channels are converted into correlation coefficients through a monotonically decaying mapping from distance to correlation, such that the smaller the spatial distance, the larger the correlation coefficient, and the larger the spatial distance, the smaller the correlation coefficient. Subsequently, the correlation coefficients are normalized row by row so that the sum of the correlation coefficients in each row is 1, thereby organizing the correlation coefficients into a spatial correlation constraint matrix that satisfies numerical stability. In subsequent iterative updates, the spatial correlation constraint matrix is ​​used to apply cross-channel spatial consistency constraints to the modal components corresponding to each channel, so that adjacent channels maintain spatial continuity and consistency during the decomposition process.

[0104] An initial set of modal component values ​​is established based on a preset number of modes, and the preset bandwidth constraint parameters are adaptively adjusted based on the signal-to-noise ratio evaluation sequence to obtain adaptive bandwidth constraint parameters.

[0105] In this embodiment 1, the signal-to-noise ratio (SNR) evaluation sequence value within the sliding window is used as the basis for adjusting the bandwidth constraint parameter. First, the SNR segment threshold and the corresponding bandwidth adjustment direction are set. When the SNR evaluation sequence falls into the low SNR segment, the preset bandwidth constraint parameter is increased to make the multivariate variational mode decomposition apply stronger bandwidth convergence constraints to the modal components, making the frequency band of the modal components narrower and suppressing broadband random noise from entering the modal component set. When the SNR evaluation sequence falls into the high SNR segment, the preset bandwidth constraint parameter is decreased to moderately relax the bandwidth convergence constraint, retain more detailed components related to cardiac activity, and reduce the risk of over-smoothing. For samples in the medium SNR segment, the bandwidth constraint parameter is continuously adjusted according to the deviation of the SNR value from the threshold, forming an adaptive bandwidth constraint parameter that monotonically responds to changes in SNR.

[0106] An augmented optimization objective function is constructed based on the normalized complex analytic space-time magnetocardiogram tensor, the initial set of modal components, the spatial correlation constraint matrix, and the adaptive bandwidth constraint parameters.

[0107] In this embodiment 1, when constructing the augmented optimization objective function, a function containing multiple optimization objectives is defined by combining the normalized complex analytic space-time magnetocardiogram tensor, the initial set of modal components, the spatial correlation constraint matrix, and the adaptive bandwidth constraint parameters. The augmented objective function mainly consists of two parts: on the one hand, the objective function should include a bandwidth constraint term for the frequency distribution of modal components, forcing the frequency band range of the modal components to be within the preset bandwidth range to avoid the influence of broadband random noise; on the other hand, the objective function should also include a cross-channel spatial consistency constraint term, which constrains the modal components of adjacent channels based on the spatial correlation constraint matrix to ensure that the modal component values ​​of spatially similar channels are similar. The augmented optimization objective function balances these optimization objectives by introducing dual variables, ensuring the balanced convergence of modal components in the frequency domain and spatial domain. Furthermore, the alternating direction method is used to achieve alternating optimization of frequency constraints and spatial consistency constraints in each iteration, ensuring that the optimal solution can effectively separate the signal and noise.

[0108] The augmented optimization objective function is iteratively updated in alternating directions in the frequency domain, and cross-channel spatial consistency constraints are applied to the modal components based on the spatial correlation constraint matrix in each update to obtain the set of frequency domain modal components.

[0109] In this embodiment 1, in the frequency domain, the frequency component corresponding to each modal component is mapped to the spatiotemporal point of each channel. The modal component values ​​between adjacent channels should be constrained according to their spatial distance and correlation coefficient. The modal components of adjacent channels should maintain consistency at the same frequency. The modal component values ​​of spatially close channels should be as similar as possible, while channels with large spatial distances can have large differences. This constraint is optimized through an iterative process to ensure that spatially similar channels share similar modal components during the modal decomposition process, eliminating signal errors and noise caused by different spatial locations.

[0110] By performing an inverse transformation and real-value consistency constraints on the frequency domain modal component set, a multimodal tensor component set is obtained.

[0111] In this embodiment, the subset of cardiac modalities specifically refers to:

[0112] Energy normalization is performed on the multimodal tensor component set, and the corresponding modal time series are extracted in the time dimension to obtain the modal space energy distribution vector set and the modal time series set;

[0113] Based on the spatial coordinate parameters of each channel, a set of candidate dipole positions is established in the preset thoracic cavity neighborhood, and a set of candidate dipole spatial response templates is generated based on the set of candidate dipole positions and the spatial coordinate parameters of each channel.

[0114] In this embodiment 1, the spatial coordinate parameters of each channel are read to form a set of sensor coordinate points, and the least squares plane fitting is used to obtain the fitting plane and normal vector; the set of sensor coordinate points is projected onto the fitting plane to obtain a set of planar coordinate points, and the convex hull of the set of planar coordinate points is calculated to obtain a candidate planar coverage area; the distance from each point in the set of planar coordinate points to its nearest neighbor is calculated, and the median distance is taken as the grid step size; the maximum distance between any two points in the candidate planar coverage area is calculated as the upper limit of depth, and the depth direction is established along the positive and negative directions of the normal vector, and the depth range is limited by the upper limit of depth; a set of planar grid points is generated in the candidate planar coverage area according to the grid step size, and a set of depth layers is generated in the depth range according to the grid step size; the set of planar grid points and the set of depth layers are combined by Cartesian combination to obtain a set of candidate dipole positions;

[0115] Channel-level matching calculations are performed on the modal space energy distribution vector set and the candidate dipole space response template set to obtain the spatial dipole consistency index sequence, and the corresponding optimal candidate dipole position identifier is determined to obtain the spatial dipole consistency result set.

[0116] In this embodiment 1, each modal space energy distribution vector in the modal space energy distribution vector set and each candidate dipole space response template in the candidate dipole space response template set are aligned in the order of consistent channel indices to form a channel alignment vector pair; Euclidean norm normalization is performed on the modal space energy distribution vector and the candidate dipole space response template in the channel alignment vector pair to obtain normalized modal space energy distribution vector and normalized candidate dipole space response template; channel-level product summation is performed on each pair of normalized modal space energy distribution vector and normalized candidate dipole space response template to obtain a matching score, and all matching scores are arranged according to the candidate dipole position identifier to form a space dipole consistency index sequence; maximum value search is performed on the space dipole consistency index sequence and the candidate dipole position identifier corresponding to the maximum value is output to form a space dipole consistency result set;

[0117] For each modal time series in the modal time series set, an amplitude envelope sequence and a main period interval sequence are constructed. The envelope stability characterization quantity and the period stability characterization quantity are calculated within a preset sliding window to obtain the temporal cardiac origin consistency index sequence.

[0118] In this embodiment 1, each modal time series in the modal time series set is read in the order of sampling time to form a sequence sample array; the sequence sample array is subjected to Hilbert transform along the time axis to obtain orthogonal component sequences, and the square root of the sum of squares of the sequence sample and the orthogonal component samples is calculated for each sampling time to form an amplitude envelope sequence; the set of local maxima that satisfy the condition that the previous sampling point is less than the current sampling point and the next sampling point is less than the current sampling point is searched in the amplitude envelope sequence to obtain the peak index sequence; the peak index difference between adjacent peaks is calculated in the time order of the peak index sequence, and the index difference is converted into a time interval according to the timestamp interval of the multi-channel magnetocardiogram data to form the main period interval sequence;

[0119] The screening thresholds for the spatial dipole consistency index sequence and the temporal cardiac source consistency index sequence are adjusted based on the signal-to-noise ratio evaluation sequence to obtain an adaptive screening threshold set.

[0120] Under the constraint of the adaptive screening threshold set, the multimodal tensor components that simultaneously satisfy the screening conditions of the spatial dipole consistency result set and the temporal cardiac source consistency index sequence are selected as the cardiac source mode subset.

[0121] In this embodiment, the weighted reconstruction of the subset of cardiac modalities and the continuity constraint on the reconstruction boundary specifically involve:

[0122] Energy normalization is performed on each multimodal tensor component in the cardiac modality subset, and credibility is calibrated based on the signal-to-noise ratio evaluation sequence to obtain a set of modal credibility weight sequences.

[0123] A spatial consistency constraint weight matrix is ​​established based on the spatial correlation constraint matrix, and cross-channel phase consistency correction is performed on each multimodal tensor component in the cardiac modality subset to obtain a set of spatial consistency corrected multimodal tensor components.

[0124] In this embodiment 1, the spatial correlation constraint matrix is ​​processed element-wise by absolute value to obtain a non-negative spatial correlation matrix. A diagonal gain matrix with diagonal elements of value 1 is superimposed on the diagonal position of the non-negative spatial correlation matrix to obtain an enhanced non-negative spatial correlation matrix. The row sum of the enhanced non-negative spatial correlation matrix is ​​calculated row by row to obtain a row normalization coefficient sequence. The row normalization coefficient sequence is used to normalize the enhanced non-negative spatial correlation matrix row by row to obtain a spatial consistency constraint weight matrix with a sum of 1 for each row.

[0125] The set of multimodal tensor components for spatial consistency correction is jointly weighted based on the set of modal confidence weight sequences and the weight matrix of spatial consistency constraints to obtain the set of multimodal tensor components with joint weights.

[0126] The initial reconstructed spatial-temporal magnetocardiogram tensor is obtained by weighted superposition and reconstruction of the joint weighted multimodal tensor component set.

[0127] Within a preset sliding window, the amplitude envelope continuity constraint and phase continuity constraint are applied to the adjacent window boundaries of the initial reconstructed spatial-temporal magnetic-oscelematic tensor, and the windows are spliced ​​to obtain the boundary-continuous reconstructed spatial-temporal magnetic-oscelematic tensor.

[0128] In this embodiment 1, overlapping intervals of adjacent sliding windows are taken along the time dimension to form boundary sample pairs; Hilbert transform is performed on the boundary sample pairs along the time dimension to obtain complex analytic boundary sequences; amplitude envelope boundary sequences and expanded phase boundary sequences are calculated from the complex analytic boundary sequences; endpoint differences of amplitude envelope boundary sequences and endpoint differences of expanded phase boundary sequences are calculated to obtain amplitude endpoint differences and phase endpoint differences; linear interpolation is performed on amplitude endpoint differences and phase endpoint differences according to time position within the overlapping interval to obtain amplitude compensation and phase compensation amounts for each sampling point in the overlapping interval; amplitude compensation and phase compensation amounts are used to synchronously compensate the amplitude envelope and expanded phase of the corresponding sampling points in the next sliding window to obtain continuous amplitude envelope and continuous expanded phase; the complex analytic signal is reconstructed based on the continuous amplitude envelope and continuous expanded phase, and the real part is taken to complete window splicing to obtain the boundary continuous reconstructed spatial-temporal magnetic tensor;

[0129] Cross-channel consistency correction is performed on the boundary continuously reconstructed spatial-temporal magnetic tensor under the constraint of the spatial consistency constraint weight matrix to obtain the denoised spatial-temporal magnetic tensor.

[0130] In this embodiment, the step of inversely calculating the equivalent current dipole trajectory based on the denoised space-time magnetic cardiomyocyte tensor and calculating the anatomical conduction consistency index specifically involves:

[0131] The equivalent current dipole forward magnetic field response relationship is established based on the noise-reduced space-time magnetic tensor and the spatial coordinate parameters of each channel, and the serialized inversion solution is performed along the time dimension to obtain the initial equivalent current dipole trajectory.

[0132] In this embodiment 1, a set of sensing channel position vectors is established based on the spatial coordinate parameters of each channel, and a one-to-one correspondence is established with the channel dimensions of the denoised spatial-temporal magnetic field tensor. The multi-channel magnetic field observation vector corresponding to each time sampling point is extracted from the denoised spatial-temporal magnetic field tensor. Under the equivalent current dipole magnetic dipole physical model, the multi-channel magnetic field observation vector is expressed as a mapping relationship between the dipole position vector and the dipole moment vector, and a forward magnetic field response relationship expression is formed. For each time sampling point, under the constraint of the forward magnetic field response relationship expression, the dipole position vector and dipole moment vector that minimize the residual between the forward calculated magnetic field and the multi-channel magnetic field observation vector are solved, and the dipole state at the corresponding time is output. The dipole position vectors output from each time sampling point are spliced ​​together in time order to form the initial equivalent current dipole trajectory.

[0133] In the preset cardiac anatomical coordinate system, the conduction center line of the His bundle-Purkinje fiber is represented as a set of directed physical paths with bifurcations, and the conduction direction vector and path length parameters are labeled to obtain the His bundle-Purkinje fiber physical path diagram.

[0134] Based on the dipole position at each moment of the initial equivalent current dipole trajectory, the path projection matching is performed with the His bundle-Purkinje fiber physical path map to obtain the path projection point sequence.

[0135] A path travel sequence is generated along the path projection point sequence and the path length parameter conduction direction, and a displacement direction sequence is generated based on the initial equivalent current dipole trajectory.

[0136] In this embodiment 1, adjacent path projection points are read according to the time index along the path projection point sequence, and the path segment to which the adjacent path projection points belong is located in the His bundle-Purkinje fiber physical path map. The conduction direction vector and path length parameters of the corresponding path segment are read. When the path segments to which adjacent path projection points belong are the same, the projection distance of the position difference vector of the adjacent path projection points in the direction of the conduction direction vector is calculated, and the conduction direction vector is normalized to obtain the path travel direction vector, forming the corresponding elements of the path travel sequence. When the path segments to which adjacent path projection points belong are inconsistent, the His bundle-Purkinje fiber physical path map is used to determine the path travel direction vector. The path connection relationship of the wild fiber physical path map determines the subsequent path segments after the path projection point sequence is crossed, and the conduction direction vector of the subsequent path segment is read. The conduction direction vector of the subsequent path segment is normalized to obtain the path travel direction vector and added to the path travel sequence. The positions of adjacent dipoles are read according to the time index along the initial equivalent current dipole trajectory. The position difference vector of adjacent dipoles is calculated and normalized to obtain the displacement direction vector, forming a displacement direction sequence. The path travel sequence and the displacement direction sequence are aligned according to the time index to form a direction vector sequence pair of the same length, which is used for the calculation of the anatomical conduction consistency index.

[0137] The anatomical conduction consistency index is calculated based on the path travel sequence, displacement direction sequence and conduction direction vector. The anatomical conduction consistency index is used to characterize the degree of path conformity of the initial equivalent current dipole trajectory along the His bundle-Purkinje fiber physical path map.

[0138] In this embodiment 1, the path travel direction vector in the path travel sequence and the displacement direction vector in the displacement direction sequence are mapped one-to-one along the time index, and the transmission direction vector of the path segment to which the path projection point belongs is synchronously mapped; a direction consistency score is calculated for the path travel direction vector and the displacement direction vector along the time index, the direction consistency score is obtained by the inner product of two unit direction vectors, and the score is used as the direction consistency score at the time point; a transmission consistency score is calculated for the displacement direction vector and the transmission direction vector along the time index, the transmission consistency score is obtained by the inner product of two unit direction vectors, and the score is used as the direction consistency score at the time point. The above score is used as the conduction consistency score at time points; the projection distance of the displacement vector along the conduction direction vector is calculated along the time index and normalized by the ratio of this distance to the path length parameter of the corresponding path segment to obtain the path travel amplitude score at time points, and the path travel amplitude score at time points corresponding to negative projection distances is set to 0; for each time index, the direction consistency score, conduction consistency score, and path travel amplitude score at time points are multiplied and fused to obtain the anatomical conduction consistency score at time points; the arithmetic mean of all anatomical conduction consistency scores at all time points is calculated to obtain the anatomical conduction consistency index;

[0139] When the anatomical conduction consistency index does not meet the preset threshold, the preset bandwidth constraint parameters are adjusted and the multivariate variational mode decomposition, the cardiac source mode subset and weighted reconstruction are re-executed to obtain the target noise-reduced spatial-temporal cardiac magnetic tensor and the equivalent current dipole trajectory that pass through the anatomical constraint.

[0140] In this embodiment, the improved semi-classical signal analysis algorithm is specifically as follows:

[0141] Based on the equivalent current dipole trajectory and the spatial coordinate parameters of each channel, the target noise reduction spatial-temporal magnetic-cardiogram tensor is weighted and aggregated across channels to obtain the heart source aggregated magnetic-cardiogram time series.

[0142] The amplitude envelope sequence and local curvature sequence were calculated based on the heart-source aggregated magnetocardiogram time series, and normalized and calibrated to obtain the envelope potential construction sequence and curvature potential construction sequence.

[0143] In this embodiment 1, a Hilbert transform is performed on the aggregated magnetocardiogram time series from the heart source along the time dimension to obtain a complex analytic time series, and the magnitude envelope sequence is obtained by taking the modulus of the complex analytic time series. The difference between adjacent sampling points of the aggregated magnetocardiogram time series from the heart source is calculated to obtain a first-order difference sequence, and the difference between adjacent points of the first-order difference sequence is calculated to obtain a second-order difference sequence. The absolute value of the second-order difference sequence is taken to obtain a curvature intensity sequence, and the curvature intensity sequence is divided by the square of the first-order difference sequence plus 1 raised to the power of 3 / 2 to obtain a local curvature sequence. The mean and standard deviation of the amplitude envelope sequence are calculated, and zero-mean unit variance normalization is performed on the amplitude envelope sequence to obtain an envelope potential construction sequence. The mean and standard deviation of the local curvature sequence are calculated, and zero-mean unit variance normalization is performed on the local curvature sequence to obtain a curvature potential construction sequence.

[0144] Construct an adaptive semiclassical potential function sequence based on the envelope potential and curvature potential sequences;

[0145] In this embodiment 1, the envelope potential construction sequence and the curvature potential construction sequence are corresponding to the sampling point numbers, and squared respectively to obtain the envelope potential energy sequence and the curvature potential energy sequence; for each sampling point, the sum of the energy of the envelope potential energy sequence and the curvature potential energy sequence is calculated, and when the sum of the energy is 0, both the envelope weight and the curvature weight are set to 0.5. When the energy is not zero, the envelope weight is set to the envelope potential energy sequence divided by the sum of the energy and the curvature weight is set to the curvature potential energy sequence divided by the sum of the energy, to obtain the adaptive weight sequence; according to the sampling point number, the envelope weight multiplied by the envelope potential energy sequence and the curvature weight multiplied by the curvature potential energy sequence are added to obtain the adaptive semiclassical potential function sequence;

[0146] A dynamically coupled Planck constant analog quantity is constructed, in which the conduction velocity characterization quantity is calculated based on the equivalent current dipole trajectory, and the magnetic field attenuation characterization quantity is calculated based on the target noise reduction space-time magnetic tensor and the spatial coordinate parameters of each channel. The Planck constant analog quantity is set as a function mapping of the signal-to-noise ratio evaluation sequence, and the Planck constant analog quantity is dynamically coupled and corrected based on the conduction velocity characterization quantity and the magnetic field attenuation characterization quantity.

[0147] In this embodiment 1, the displacement length sequence is obtained by calculating the difference in the equivalent current dipole positions of adjacent moments according to the time index, and the time interval sequence is obtained by calculating the difference in the timestamps of adjacent moments according to the time index. The displacement length sequence is divided by the time interval sequence to obtain the conduction velocity characterization sequence. The magnetic field amplitude sequence of each channel is extracted from the target noise reduction spatial-temporal magnetic tensor according to the time index. The channel-to-dipole distance sequence is calculated according to the time index based on the spatial coordinate parameters of each channel and the position of the equivalent current dipole. The natural logarithm of the magnetic field amplitude sequence and the channel-to-dipole distance sequence is taken and linearly fitted to obtain the magnetic field attenuation characterization sequence. The minimum and maximum values ​​of the signal-to-noise ratio evaluation sequence are calculated and linearly normalized to obtain the normalized signal-to-noise ratio sequence. The complement of the normalized signal-to-noise ratio sequence at value 1 is calculated to form the generalized signal-to-noise ratio. The Planck constant analog quantity sequence is obtained by performing linear normalization on the conduction velocity characterization quantity sequence and the magnetic field attenuation characterization quantity sequence to obtain a normalized velocity sequence and a normalized attenuation sequence, respectively. The sum of the squares of the normalized velocity sequence and the normalized attenuation sequence is calculated according to the time index to obtain a sum of squares sequence. When the sum of squares sequence is 0, both the velocity weight and the attenuation weight are set to 0.5. When the sum of squares sequence is non-zero, the velocity weight is set to the square of the normalized velocity sequence divided by the sum of squares sequence, and the attenuation weight is set to the square of the normalized attenuation sequence divided by the sum of squares sequence. The coupling factor sequence is obtained by multiplying the velocity weight by the sum of the normalized velocity sequence and the attenuation weight by the normalized attenuation sequence according to the time index. The dynamic coupled Planck constant analog quantity is obtained by multiplying the coupling factor sequence calculated according to the time index and the Planck constant analog quantity sequence point by point.

[0148] A semiclassical spectral operator is constructed based on an adaptive semiclassical potential function sequence and a dynamically coupled Planck constant analog quantity. The semiclassical spectral operator is then discretized in the time dimension to form a discrete representation of the spectral operator.

[0149] The discrete representation of the spectral operator is decomposed to obtain a semiclassical eigenvalue sequence, and the semiclassical spectral feature set is calculated based on the semiclassical eigenvalue sequence and the adaptive semiclassical potential function sequence.

[0150] In this embodiment 1, the discrete representation of the spectral operator is written as a real symmetric matrix and arranged in rows and columns according to time index; the real symmetric matrix is ​​subjected to QR iteration to perform eigenvalue decomposition to calculate all eigenvalues ​​and sorted in ascending order, and the sorted eigenvalues ​​are sequentially formed into a semiclassical eigenvalue sequence; the mean, standard deviation, and square mean energy of the adaptive semiclassical potential function sequence are calculated to form a potential function energy feature vector; the number of eigenvalues, the sum of eigenvalues, the mean, the variance, the mean of the interval between adjacent eigenvalues, and the variance of the interval between adjacent eigenvalues ​​are calculated for the semiclassical eigenvalue sequence to form a spectral structure feature vector; the potential function energy feature vector and the spectral structure feature vector are concatenated in a fixed order and zero-mean unit variance normalization is performed to obtain a semiclassical spectral feature set.

[0151] In this embodiment, the step of outputting physiological abnormality parameter detection results and abnormality severity assessment results based on structural changes in phase space topological characteristic parameters specifically includes:

[0152] Based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set, a phase space state vector sequence is constructed under a preset delay dimension and a preset embedding dimension, and then scale normalization and coordinate consistency are performed.

[0153] In this embodiment 1, along the time index The semiclassical eigenvalue sequence in The values ​​at time n and the feature values ​​of the semiclassical spectral feature set at time n are concatenated in a fixed order to form the feature vector. Defining the delay dimension as 1 sampling interval and the embedding dimension as 3, the time index used for concatenation is obtained by... , , Composition; Time Index Each value not less than 3 ,Will , , The phase space state vector is formed by piecing together data from most recent to oldest time. Indexed by time All in ascending order Form a phase space state vector sequence;

[0154] A high-dimensional phase space manifold is constructed based on the phase space state vector sequence, and local neighborhood graph connectivity and geodesic distance consistency correction are performed to obtain a measurable high-dimensional phase space manifold.

[0155] In this embodiment 1, the phase space state vector sequence is represented as an array arranged by time index. Each phase space state vector to For each Calculation and All Euclidean distance ,exist Select under the conditions The smallest set of 10 indices And in and Within each Establish undirected edges between them to form a local neighborhood graph; connect each undirected edge The boundary weight is set as The local neighborhood graph is decomposed into connected components, and the minimum inter-component ratio is calculated when the number of connected components is greater than 1. Add undirected edges to the corresponding node pairs until the number of connected components is 1, thus obtaining a connected weighted neighborhood graph; for each node in the connected weighted neighborhood graph... Perform single-source shortest path calculation to obtain the shortest path distance. And will be by The resulting distance matrix is ​​defined as the geodesic distance matrix; the set of nodes and edges of the connected weighted neighborhood graph, together with the geodesic distance matrix, are represented as a measurable high-dimensional phase space manifold; the distance between any pair of nodes in the geodesic distance matrix is ​​defined as the distance of the corresponding shortest path on the connected weighted neighborhood graph, and geodesic distance consistency correction is completed;

[0156] Based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set, measurable high-dimensional phase space manifolds are segmented in stages to obtain depolarized phase space submanifolds and complex polarized phase space submanifolds.

[0157] The set of topological characteristic parameters of phase space is calculated based on depolarized phase space submanifolds and complex polarized phase space submanifolds respectively. The set of topological characteristic parameters of phase space consists of Lyapunov exponent parameters, Hausdorff dimension parameters and topological durability parameters.

[0158] In this embodiment 1, the depolarized phase space submanifold and the complex polarized phase space submanifold are represented as point sets arranged by time index, and the corresponding geodesic distance matrix is ​​read. Within the same point set, for each time index, the outlier with the smallest geodesic distance is selected as the neighbor and the initial geodesic distance is recorded. The time step is defined as the integer obtained by dividing the point set size by 10, and if it is less than 1, it is set to 1. The point pairs are synchronously moved forward by the time step along the time index to obtain the subsequent geodesic distance. The subsequent geodesic distance is divided by the initial geodesic distance to obtain the growth factor. The natural logarithm of the growth factor is taken and then divided by the time step to obtain the Lyapunov exponent value of a single point. The average of all time indices is calculated to obtain the Lyapunov exponent parameter. Within the same point set, the maximum geodesic distance and the minimum non-zero geodesic distance are calculated, and the coverage radius is halved each time starting from the maximum geodesic distance until the coverage radius is no greater than the minimum non-zero geodesic distance. For each coverage radius, uncovered points are selected as coverage centers in ascending order of time index. Points with a geodesic distance not greater than the coverage radius are marked as covered, and the number of coverage centers is counted to obtain a coverage number sequence. The natural logarithm sequence of the inverse of the coverage radius is fitted with the natural logarithm sequence of the coverage number using least squares linear fitting, and the fitting slope is used as the Hausdorff dimension parameter. On the same point set, a Vietoris-Rips simple complex sequence is constructed according to the coverage radius sequence, and the set of persistent intervals of 0-dimensional homology and 1-dimensional homology is calculated. For each persistent interval, the termination scale is subtracted from the generation scale to obtain the lifetime, and the summation of all lifetimes is used to obtain the topological persistence parameter. The Lyapunov exponent parameter, Hausdorff dimension parameter, and topological persistence parameter are combined in a fixed order to form a set of topological feature parameters for depolarized phase space and a set of topological feature parameters for complex polarized phase space.

[0159] The topological structure change is calculated based on the set of phase space topological feature parameters, and the electrophysiological instability index is calculated based on the topological structure change. The electrophysiological instability index is used to characterize the electrophysiological instability of the depolarization and repolarization processes of cardiomyocytes.

[0160] The results of physiological abnormality parameter detection and abnormality severity assessment are output based on the electrophysiological instability index.

[0161] In this embodiment 1, electrophysiological instability indices are calculated for each time period obtained by segmentation and an electrophysiological instability index sequence is formed according to the time index. Within the electrophysiological instability index sequence, the time index corresponding to the maximum electrophysiological instability index is determined and the abnormal time period identifier is output as the physiological abnormality parameter detection result. Within the electrophysiological instability index sequence, the minimum and maximum electrophysiological instability indices are determined, and the maximum value minus the minimum value is used as the scale range. The minimum electrophysiological instability index is subtracted from the maximum electrophysiological instability index, and the difference is divided by the scale range to obtain the severity value. When the severity value is between 0 and 0.20, the abnormal severity assessment result is output as level 1; when the severity value is between 0.20 and 0.40, the abnormal severity assessment result is output as level 2; when the severity value is between 0.40 and 0.60, the abnormal severity assessment result is output as level 3; when the severity value is between 0.60 and 0.80, the abnormal severity assessment result is output as level 4; and when the severity value is between 0.80 and 1, the abnormal severity assessment result is output as level 5.

[0162] In this embodiment, a system for detecting abnormal physiological parameters based on magnetocardiography includes:

[0163] The magnetocardiogram (MCC) acquisition unit is used to acquire multi-channel MCC data of the subject and record spatial coordinate parameters and environmental reference magnetic field.

[0164] The preprocessing unit performs time alignment, phase consistency calibration, drift benchmark determination, baseline drift estimation and cancellation on the multi-channel magnetocardiogram data and the environmental reference magnetic field, and outputs the time-aligned magnetocardiogram tensor and signal-to-noise ratio evaluation sequence.

[0165] The mode decomposition unit performs multivariate variational mode decomposition on the space-time magnetocardiogram tensor, outputs a set of multimodal tensor components, and calculates the correlation constraint matrix;

[0166] The cardiac modality subset construction unit constructs cardiac modality subsets based on modal components and spatial coordinate parameters, and performs energy normalization, confidence calibration, and consistency correction.

[0167] The weighted reconstruction unit performs weighted reconstruction on a subset of cardiac modalities, applies boundary continuity constraints, and outputs a denoised space-time cardiac magnetic tensor.

[0168] The equivalent current dipole inversion unit, based on the denoised space-time magnetic cardiotensor and spatial coordinate parameters, inversely calculates the equivalent current dipole trajectory and the anatomical conduction consistency index.

[0169] The bandwidth constraint adjustment unit adjusts the bandwidth constraint parameters and triggers multivariate variational mode decomposition and cardiac modality subset reconstruction when the anatomical conduction consistency index does not meet the threshold.

[0170] The semiclassical signal analysis unit performs cross-channel weighted aggregation based on the equivalent current dipole trajectory and spatial coordinate parameters to generate a cardiac source aggregated time series, and performs envelope and curvature calculations.

[0171] The phase space manifold and topology analysis unit constructs a high-dimensional phase space manifold based on the eigenvalue sequence and spectral feature set, calculates topological feature parameters, and outputs physiological abnormality detection results and abnormality severity assessment.

[0172] Example 2:

[0173] To verify the feasibility of this invention in practice, the method and system for detecting abnormal physiological parameters based on magnetocardiography (MCG) disclosed in this invention were applied to a screening and follow-up scenario for MCG testing. Subjects underwent resting data acquisition within a shielded chamber, and the system output the detection results and an assessment of the severity of the abnormalities. In this scenario, the environmental reference magnetic field exhibits drift and fluctuations, and the spatial gradient differences in multi-channel MCG data, coupled with individual differences, can easily lead to low signal-to-noise ratios, residual environmental interference, and unstable abnormal features. This invention improves output stability through spatial-temporal MCG tensor modeling, spatial correlation constraints from multivariate variational mode decomposition, closed-loop verification of anatomical conduction consistency, and phase space topology discrimination.

[0174] In this embodiment, the number of channels in the multi-channel magnetocardiogram (MCG) data is set to 64, the sampling frequency is set to 1000, and the duration of a single acquisition is set to 180. During preprocessing, the multi-channel MCG data is timestamped and aligned with the environmental reference magnetic field. Cross-channel phase consistency calibration and alignment resampling are performed on the multi-channel MCG data, with the resampling target frequency set to 1000. A drift baseline is determined based on the environmental reference magnetic field, and baseline drift estimation and drift cancellation are performed within a sliding window of length 8. Spatial smoothing constraints are established based on spatial coordinate parameters, and environmental reference magnetic field cancellation is performed under the spatial smoothing constraints using the environmental reference magnetic field. Channel amplitude uniformity is performed on the environmental interference-free multi-channel MCG sequence, and an initial spatial-temporal MCG tensor is established. The system calculates the signal-to-noise ratio (SNR) evaluation sequence within a sliding window of length 2 and step size 0.5. The segments with an SNR evaluation sequence below 0.45 are identified as low SNR segments. Tensor completion repair and consistency correction are performed under cross-channel consistency constraints. The number of iterations for tensor completion repair is set to 30 to form a spatial-temporal magnetic-cardiogram tensor.

[0175] During multivariate variational mode decomposition, a Hilbert transform is performed along the time dimension of the spatial-temporal magnetic-cardiogram tensor, followed by energy normalization, to obtain a normalized complex analytic spatial-temporal magnetic-cardiogram tensor. The inter-channel distance matrix is ​​calculated based on spatial coordinate parameters, and a spatial correlation constraint matrix is ​​constructed, with a distance attenuation coefficient set to 0.2. The preset number of modes is set to 6, and the initial value of the preset bandwidth constraint parameter is set to 1200, adaptively adjusted based on the signal-to-noise ratio (SNR) evaluation sequence. Alternating direction iterative updates are used in the frequency domain, with a maximum number of iterations set to 120 and a convergence threshold set to 0.00001, resulting in a set of multimodal tensor components. When constructing the cardiac source mode subset, 240 candidate dipole positions are established in a preset thoracic cavity neighborhood, and a spatial response template set is generated. A spatial dipole consistency index sequence is obtained based on channel-level matching. Combining the amplitude envelope sequence and the main period interval sequence of the modal time series, a temporal cardiac source consistency index sequence is obtained. The screening threshold is adjusted based on the SNR evaluation sequence to output the cardiac source mode subset. During weighted reconstruction, a set of modal confidence weight sequences is obtained by confidence calibration based on the signal-to-noise ratio evaluation sequence. A spatial consistency constraint weight matrix is ​​established based on the spatial correlation constraint matrix and cross-channel phase consistency correction is performed to obtain the denoised spatial-temporal magnetic cardiometry tensor. At the reconstruction boundary, a sliding window with a length of 2 and an overlap length of 1.5 is used to perform amplitude envelope continuity constraints and phase continuity constraints, and the window is stitched together.

[0176] During the equivalent current dipole inversion, the forward magnetic field response relationship of the equivalent current dipole is established based on the denoised spatial-temporal cardiac magnetic tensor and spatial coordinate parameters, and the inversion is performed sequentially along the time dimension. The regularization weight is set to 0.08, and the time step is set to 0.002 to obtain the initial equivalent current dipole trajectory. The system establishes a His bundle-Purkinje fiber physical path map in a preset cardiac anatomical coordinate system, with the number of path segments set to 36, and performs path projection matching to obtain a path projection point sequence. Based on the path projection point sequence, a path travel sequence is generated, and based on the equivalent current dipole trajectory, a displacement direction sequence is generated. The anatomical conduction consistency index is calculated, and the threshold is set to 0.82. For samples that do not meet the threshold, the bandwidth constraint parameters are adjusted by a ratio of 15, and multivariate variational mode decomposition, construction of cardiac source mode subsets, and weighted reconstruction are performed again. The maximum number of loops is set to 3, thereby obtaining the target denoised spatial-temporal cardiac magnetic tensor and equivalent current dipole trajectory that pass through the anatomical constraints.

[0177] In the improved semiclassical signal analysis algorithm, the target noise reduction spatial-temporal magnetic-cardiogram tensor is weighted and aggregated across channels based on the equivalent current dipole trajectory and spatial coordinate parameters to obtain the aggregated magnetic-cardiogram time series. The amplitude envelope sequence and local curvature sequence are calculated and normalized to construct an adaptive semiclassical potential function. The Planck constant analog quantity is set as a function mapping of the signal-to-noise ratio evaluation sequence and a dynamic coupling relationship is established with the conduction velocity characterization quantity and the magnetic field attenuation characterization quantity to construct a dynamically coupled Planck constant analog quantity. Based on the adaptive semiclassical potential function and the dynamically coupled Planck constant analog quantity, a semiclassical spectral operator is constructed and discretized to form a discrete representation of the spectral operator. The discretization length is set to 512, resulting in a semiclassical eigenvalue sequence and a semiclassical spectral feature set. Subsequently, a phase space state vector sequence is constructed based on a preset delay dimension of 6 and a preset embedding dimension of 8, and a measurable high-dimensional phase space manifold is obtained with a neighborhood size of 12. The manifold is then segmented into stages to obtain depolarized and repolarized phase space submanifolds. The set of phase space topological feature parameters, consisting of Lyapunov exponent parameters, Hausdorff dimension parameters, and topological durability parameters, is calculated to form the topological structure change quantity and calculate the electrophysiological instability index. The detection result is output with an electrophysiological instability index threshold of 0.65 and mapped to obtain the abnormality severity assessment result.

[0178] In this embodiment, 50 samples were selected, and the reference values ​​obtained from synchronous control detection were used as the measured values. The detection accuracy was 0.92, the sensitivity was 0.90, the specificity was 0.94, the mean absolute error of the anomaly severity assessment results was 3.6, and the mean square error was 18.4. In the control procedure, the closed-loop adjustment of the anatomical conduction consistency index was removed, and the bandwidth constraint parameter was fixed at 1200. Simultaneously, the Planck constant analog quantity was fixed at 0.70. The detection accuracy was 0.83, the sensitivity was 0.79, the specificity was 0.86, the mean absolute error was 8.9, and the mean square error was 76.5, indicating that the present invention has a more stable ability to output anomaly parameters under low signal-to-noise ratio and strong interference conditions.

[0179] Table 1 Comparison of Measured and Predicted Values ​​of Physiological Abnormal Parameters

[0180] Sample number Measured values ​​of anatomical conduction consistency index Anatomical conduction consistency index predicted value Measured values ​​of conduction velocity characterization Predicted values ​​of conduction velocity characterization Measured values ​​of magnetic field attenuation characterization Predicted values ​​of magnetic field attenuation characterization Measured values ​​of electrophysiological instability index Predicted values ​​of electrophysiological instability index Measured values ​​of abnormal severity assessment results Predicted values ​​of abnormal severity assessment results 001 0.88 0.87 0.46 0.45 1.34 1.37 0.72 0.70 74.0 71.0 012 0.84 0.85 0.52 0.50 1.21 1.25 0.61 0.63 58.0 60.0 023 0.90 0.89 0.41 0.42 1.46 1.44 0.79 0.81 83.0 85.0 037 0.83 0.82 0.55 0.56 1.18 1.16 0.57 0.55 49.0 47.0 046 0.86 0.86 0.48 0.49 1.29 1.30 0.67 0.66 66.0 65.0

[0181] As shown in Table 1, the predicted and measured values ​​of the anatomical conduction consistency index for each sample are consistent. The predicted value for sample 037 is 0.82, which meets the threshold of 0.82, indicating that the projection matching of the equivalent current dipole trajectory on the His bundle-Purkinje fiber physical path map can maintain the degree of path conformity. The electrophysiological instability index can be synchronously adjusted with changes in the conduction velocity characterization and magnetic field attenuation characterization. For example, the measured value of the magnetic field attenuation characterization for sample 023 is 1.46 and the predicted value is 1.44, the measured value of the electrophysiological instability index is 0.79 and the predicted value is 0.81, and the corresponding abnormality severity assessment result is 83.0 for the measured value and 85.0 for the predicted value, indicating that the dynamically coupled Planck constant analog has a stabilizing effect on the energy level resolution of the semiclassical eigenvalue sequence. The absolute errors of the five samples in the anomaly severity assessment results were 3.0, 2.0, 2.0, 2.0, and 1.0, respectively, with a mean absolute error of 2.0. The error direction remained symmetrical across different samples, indicating that the topological changes formed by the set of phase space topological feature parameters can stably characterize the structural differences between depolarized and repolarized phase space submanifolds. Combined with the statistical results of all samples, the closed-loop anatomical conduction consistency index enabled some samples to pass the threshold constraint in the second loop, reducing the perturbation of the semiclassical spectral feature set by reconstruction artifacts.

[0182] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal physiological parameters based on magnetocardiography, characterized in that, include: Multi-channel magnetocardiogram data of the subjects were collected, the spatial coordinate parameters of the corresponding channels and the environmental reference magnetic field were recorded, and preprocessing was performed to obtain the spatial-temporal magnetocardiogram tensor and signal-to-noise ratio evaluation sequence. Multivariate variational mode decomposition of the space-time magnetocardiogram tensor yields a set of multimodal tensor components; A subset of cardiac modalities is constructed based on the set of multimodal tensor components and the spatial coordinate parameters of the channels; A weighted reconstruction of a subset of cardiac modalities is performed, and the reconstruction boundary is constrained to obtain a denoised spatial-temporal cardiac magnetotensor. The equivalent current dipole trajectory is derived by inverse calculation of the denoised space-time magnetic cardiometa-tensor, and the anatomical conduction consistency index is calculated to obtain the target denoised space-time magnetic cardiometa-tensor. By processing the target noise reduction space-time cardiomagnetic tensor and equivalent current dipole trajectory using an improved semiclassical signal analysis algorithm, a semiclassical eigenvalue sequence and a semiclassical spectral feature set are obtained. A high-dimensional phase space manifold is constructed based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set. The phase space topological feature parameters are calculated, and the results of physiological abnormality parameter detection and abnormality severity assessment are output according to the structural changes of the phase space topological feature parameters.

2. The method for detecting abnormal physiological parameters based on magnetocardiography according to claim 1, characterized in that, The preprocessing specifically includes: The multi-channel magnetocardiogram (MCC) data was timestamped and aligned with the environmental reference magnetic field. Cross-channel phase consistency calibration and alignment resampling were then performed on the multi-channel MCC data to obtain a time-aligned multi-channel MCC sequence. The drift baseline is determined based on the environmental reference magnetic field, and the baseline drift is estimated and drift is canceled on the time-aligned multi-channel magnetocardiogram sequence to obtain the drift-free multi-channel magnetocardiogram sequence. Based on the spatial coordinate parameters of each channel, a spatial smoothing constraint is established, and under the spatial smoothing constraint, the drift-free multi-channel magnetocardiogram sequence is canceled by the environmental reference magnetic field to obtain a multi-channel magnetocardiogram sequence free of environmental interference. The channel amplitudes of the multi-channel magnetocardiogram (MCC) sequence were made uniform to remove environmental interference, and the initial spatial-temporal MCC tensor corresponding to the channel dimension and time dimension was established based on the spatial coordinate parameters of each channel. Based on the initial space-time magnetic cardiomyocyte tensor, the channel energy stability characterization and residual volatility characterization are calculated within a preset sliding window, and the signal-to-noise ratio evaluation sequence is calculated based on the channel energy stability characterization and residual volatility characterization. Based on the signal-to-noise ratio (SNR) evaluation sequence, low SNR segments in the initial spatial-temporal magnetic-cardiogram (MCC) tensor are identified. Under cross-channel consistency constraints, tensor completion and consistency correction are performed on the low SNR segments to form the spatial-temporal MCC tensor.

3. The method for detecting abnormal physiological parameters based on magnetocardiography according to claim 1, characterized in that, The multivariate variational mode decomposition is specifically as follows: A Hilbert transform is performed along the time dimension of the spacetime magnetocardiogram tensor, followed by energy normalization, to obtain the normalized complex analytic spacetime magnetocardiogram tensor. The distance matrix between channels is calculated based on the spatial coordinate parameters of each channel, and a spatial correlation constraint matrix is ​​constructed. An initial set of modal component values ​​is established based on a preset number of modes, and the preset bandwidth constraint parameters are adaptively adjusted based on the signal-to-noise ratio evaluation sequence to obtain adaptive bandwidth constraint parameters. An augmented optimization objective function is constructed based on the normalized complex analytic space-time magnetocardiogram tensor, the initial set of modal components, the spatial correlation constraint matrix, and the adaptive bandwidth constraint parameters. The augmented optimization objective function is iteratively updated in alternating directions in the frequency domain, and cross-channel spatial consistency constraints are applied to the modal components based on the spatial correlation constraint matrix in each update to obtain the set of frequency domain modal components. By performing an inverse transformation and consistency constraints on the frequency domain modal component set, a multimodal tensor component set is obtained.

4. The method for detecting abnormal physiological parameters based on magnetocardiography according to claim 1, characterized in that, The specific subset of cardiac modalities is as follows: Energy normalization is performed on the multimodal tensor component set, and the corresponding modal time series are extracted in the time dimension to obtain the modal space energy distribution vector set and the modal time series set; Based on the spatial coordinate parameters of each channel, a set of candidate dipole positions is established in the preset thoracic cavity neighborhood, and a set of candidate dipole spatial response templates is generated based on the set of candidate dipole positions and the spatial coordinate parameters of each channel. Channel-level matching calculations are performed on the modal space energy distribution vector set and the candidate dipole space response template set to obtain the spatial dipole consistency index sequence, and the corresponding optimal candidate dipole position identifier is determined to obtain the spatial dipole consistency result set. For each modal time series in the modal time series set, an amplitude envelope sequence and a main period interval sequence are constructed. The envelope stability characterization quantity and the period stability characterization quantity are calculated within a preset sliding window to obtain the temporal cardiac origin consistency index sequence. The screening thresholds for the spatial dipole consistency index sequence and the temporal cardiac source consistency index sequence are adjusted based on the signal-to-noise ratio evaluation sequence to obtain an adaptive screening threshold set. Under the constraint of the adaptive screening threshold set, the multimodal tensor components that simultaneously satisfy the screening conditions of the spatial dipole consistency result set and the temporal cardiac source consistency index sequence are selected as the cardiac source mode subset.

5. The method for detecting abnormal physiological parameters based on magnetocardiography according to claim 1, characterized in that, The weighted reconstruction of the subset of cardiac modalities, and the application of continuity constraints to the reconstruction boundary, specifically involves: Energy normalization is performed on each multimodal tensor component in the cardiac modality subset, and credibility is calibrated based on the signal-to-noise ratio evaluation sequence to obtain a set of modal credibility weight sequences. A spatial consistency constraint weight matrix is ​​established based on the spatial correlation constraint matrix, and cross-channel phase consistency correction is performed on each multimodal tensor component in the cardiac modality subset to obtain a set of spatial consistency corrected multimodal tensor components. The set of multimodal tensor components for spatial consistency correction is jointly weighted based on the set of modal confidence weight sequences and the weight matrix of spatial consistency constraints to obtain the set of multimodal tensor components with joint weights. The initial reconstructed spatial-temporal magnetocardiogram tensor is obtained by weighted superposition and reconstruction of the joint weighted multimodal tensor component set. Within a preset sliding window, the amplitude envelope continuity constraint and phase continuity constraint are applied to the adjacent window boundaries of the initial reconstructed spatial-temporal magnetic-oscelematic tensor, and the windows are spliced ​​to obtain the boundary-continuous reconstructed spatial-temporal magnetic-oscelematic tensor. Cross-channel consistency correction is performed on the boundary continuously reconstructed spatial-temporal magnetic-cardiogram tensor under the constraint of the spatial consistency constraint weight matrix to obtain the denoised spatial-temporal magnetic-cardiogram tensor.

6. The method for detecting abnormal physiological parameters based on magnetocardiography according to claim 1, characterized in that, The method of inversely calculating the equivalent current dipole trajectory based on the denoised space-time magnetic cardiomyocyte tensor and calculating the anatomical conduction consistency index is as follows: The equivalent current dipole forward magnetic field response relationship is established based on the noise-reduced space-time magnetic tensor and the spatial coordinate parameters of each channel, and the serialized inversion solution is performed along the time dimension to obtain the initial equivalent current dipole trajectory. In the preset cardiac anatomical coordinate system, the conduction center line of the His bundle-Purkinje fiber is represented as a set of directed physical paths with bifurcations, and the conduction direction vector and path length parameters are labeled to obtain the His bundle-Purkinje fiber physical path diagram. Based on the dipole position at each moment of the initial equivalent current dipole trajectory, the path projection matching is performed with the His bundle-Purkinje fiber physical path map to obtain the path projection point sequence. A path travel sequence is generated along the path projection point sequence and the path length parameter conduction direction, and a displacement direction sequence is generated based on the initial equivalent current dipole trajectory. Calculate the anatomical conduction consistency index based on the path travel sequence, displacement direction sequence, and conduction direction vector; When the anatomical conduction consistency index does not meet the preset threshold, the preset bandwidth constraint parameters are adjusted and the multivariate variational mode decomposition, the cardiac source mode subset and weighted reconstruction are re-executed to obtain the target noise-reduced spatial-temporal cardiac magnetic tensor and the equivalent current dipole trajectory that pass through the anatomical constraint.

7. The method for detecting abnormal physiological parameters based on magnetocardiography according to claim 1, characterized in that, The improved semi-classical signal analysis algorithm is specifically as follows: Based on the equivalent current dipole trajectory and the spatial coordinate parameters of each channel, the target noise reduction spatial-temporal magnetic-cardiogram tensor is weighted and aggregated across channels to obtain the heart source aggregated magnetic-cardiogram time series. The amplitude envelope sequence and local curvature sequence were calculated based on the heart-source aggregated magnetocardiogram time series, and normalized and calibrated to obtain the envelope potential construction sequence and curvature potential construction sequence. Construct an adaptive semiclassical potential function sequence based on the envelope potential and curvature potential sequences; Construct a dynamically coupled analog of Planck's constant; A semiclassical spectral operator is constructed based on an adaptive semiclassical potential function sequence and a dynamically coupled Planck constant analog quantity. The semiclassical spectral operator is then discretized in the time dimension to form a discrete representation of the spectral operator. The discrete representation of the spectral operator is decomposed to obtain a semiclassical eigenvalue sequence, and the semiclassical spectral feature set is calculated based on the semiclassical eigenvalue sequence and the adaptive semiclassical potential function sequence.

8. The method for detecting abnormal physiological parameters based on magnetocardiography according to claim 1, characterized in that, The method of outputting physiological abnormality parameter detection results and abnormality severity assessment results based on structural changes in phase space topological feature parameters is as follows: Based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set, a phase space state vector sequence is constructed under a preset delay dimension and a preset embedding dimension, and then scale normalization and coordinate consistency are performed. A high-dimensional phase space manifold is constructed based on the phase space state vector sequence, and local neighborhood graph connectivity and geodesic distance consistency correction are performed to obtain a measurable high-dimensional phase space manifold. Based on the semiclassical eigenvalue sequence and the semiclassical spectral feature set, measurable high-dimensional phase space manifolds are segmented into stages to obtain depolarized phase space submanifolds and complex polarized phase space submanifolds. The set of topological characteristic parameters of phase space is calculated based on depolarized phase space submanifolds and complex polarized phase space submanifolds, respectively. The topological structure change is calculated based on the set of phase space topological characteristic parameters, and the electrophysiological instability index is calculated based on the topological structure change. The results of physiological abnormality parameter detection and abnormality severity assessment are output based on the electrophysiological instability index.

9. A system for detecting abnormal physiological parameters based on magnetocardiography, comprising the method for detecting abnormal physiological parameters based on magnetocardiography as described in any one of claims 1 to 8, characterized in that, include: The magnetocardiogram (MCC) acquisition unit is used to acquire multi-channel MCC data of the subject and record spatial coordinate parameters and environmental reference magnetic field. The preprocessing unit preprocesses the multi-channel magnetocardiogram data and the environmental reference magnetic field, and outputs time-aligned magnetocardiogram tensor and signal-to-noise ratio evaluation sequences. The mode decomposition unit performs multivariate variational mode decomposition on the space-time magnetocardiogram tensor, outputs a set of multimodal tensor components, and calculates the correlation constraint matrix; The cardiac modality subset construction unit constructs cardiac modality subsets based on modal components and spatial coordinate parameters, and performs energy normalization, confidence calibration, and consistency correction. The weighted reconstruction unit performs weighted reconstruction on a subset of cardiac modalities, applies boundary continuity constraints, and outputs a denoised space-time cardiac magnetic tensor. The equivalent current dipole inversion unit, based on the denoised space-time magnetic cardiotensor and spatial coordinate parameters, inversely calculates the equivalent current dipole trajectory and the anatomical conduction consistency index. The bandwidth constraint adjustment unit adjusts the bandwidth constraint parameters and triggers multivariate variational mode decomposition and cardiac modality subset reconstruction when the anatomical conduction consistency index does not meet the threshold. The semiclassical signal analysis unit performs cross-channel weighted aggregation based on the equivalent current dipole trajectory and spatial coordinate parameters to generate a cardiac source aggregated time series, and performs envelope and curvature calculations. The phase space manifold and topology analysis unit constructs a high-dimensional phase space manifold based on the eigenvalue sequence and spectral feature set, calculates topological feature parameters, and outputs physiological abnormality detection results and abnormality severity assessment.

Images