Multidimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation
By employing a multi-region-of-interest dynamic mask and multi-scale image transformation method, the problem of decreased feature discrimination in magnetocardiogram (MCC) signal analysis was solved. This method enables automated extraction and robustness improvement of multi-dimensional MCC features, adapts to individual differences, and enhances the accuracy and stability of MCC diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2026-02-11
- Publication Date
- 2026-06-26
AI Technical Summary
Existing magnetocardiogram (MCC) signal analysis methods ignore the two-dimensional spatial characteristics of the cardiac magnetic field, rely on manual or fixed region selection of regions of interest, resulting in decreased feature discrimination, difficulty in maintaining stability in multi-center studies, and poor robustness due to the single feature dimension.
By employing a multi-region-of-interest (MOO) dynamic mask and multi-scale image transformation method, multi-dimensional magnetocardiographic features, including first-order statistical features, shape features, and texture features, are extracted by drawing two-dimensional temporal isomagnetic maps, creating automated dynamic masks, and implementing multi-scale image transformation.
It achieves automated multi-dimensional magnetocardiogram feature extraction, with strong feature representation capabilities, high robustness, and adaptability to individual differences, thereby improving the accuracy and stability of magnetocardiogram diagnosis.
Smart Images
Figure CN122272030A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of biomedical signal analysis technology, and specifically relates to a multi-dimensional magnetic cardiomyocyte feature extraction system based on dynamic masking of multiple regions of interest and multi-scale image transformation. Background Technology
[0002] Magnetocardiography (MCG) is a non-invasive technique for diagnosing cardiac function by measuring the weak magnetic field generated by cardiac activity. Unlike electrocardiography (ECG), which records surface electrical potentials, MCG directly detects the magnetic field signal accompanying cardiac electrical currents. Because the magnetic field distribution is less affected by the non-uniform conductivity of human tissues, MCG has higher spatial resolution and can more accurately locate abnormal currents deep within the myocardium. In recent years, with the development of spin-free exchange-relaxation atomic magnetometer sensor technology, significant progress has been made in the practical application and miniaturization of magnetocardiography devices, laying the foundation for their clinical application.
[0003] Despite the promising future of magnetocardiography (MCG), the development of its signal analysis and feature extraction methods has lagged behind, severely limiting its diagnostic efficacy. Current analysis methods have several limitations: First, they often rely on the one-dimensional waveform analysis approach of electrocardiography (ECG), ignoring the inherent two-dimensional spatial characteristics of the cardiac magnetic field, and failing to effectively capture key three-dimensional dynamic information such as electromagnetic propagation patterns, eddy current formation, and extreme point trajectories. Second, they frequently depend on manual or fixed-region selection of regions of interest, which is highly subjective and cannot adapt to individual differences, leading to decreased feature discrimination. Third, the extracted features are of a single dimension and lack robustness, making it difficult to maintain stability in multi-center studies. These shortcomings prevent the full utilization of the advantages of high-dimensional spatiotemporal data from magnetocardiography, becoming a significant factor hindering its application in precision medicine. Summary of the Invention
[0004] To overcome the above-mentioned technical problems, this invention provides a multi-dimensional magnetic cardiomyocyte feature extraction system based on dynamic masks with multiple regions of interest and multi-scale image transformation. It can automatically extract multi-dimensional magnetic cardiomyocyte features and has the advantages of strong feature representation ability and high robustness.
[0005] The technical solution of the present invention is as follows:
[0006] A multi-dimensional magnetocardiogram feature extraction system based on dynamic masks for multiple regions of interest and multi-scale image transformation, characterized in that it includes:
[0007] The module for drawing two-dimensional temporal isomagnetic maps is used to construct the cardiac magnetic field distribution in the preprocessed multi-channel magnetic field signals through spatial interpolation algorithms, draw two-dimensional isomagnetic maps of the entire cycle of a single heartbeat, and arrange them in chronological order to form a complete two-dimensional isomagnetic map sequence to characterize the dynamic changes in cardiac electromagnetic activity.
[0008] An automated dynamic mask module is created to divide the magnetic cycle into multiple key bands based on the two-dimensional isomagnetic map sequence and the magnetic-cardiographic physiological characteristics. For each band, a dynamic mask with multiple regions of interest is automatically generated.
[0009] A multi-scale image transformation module is implemented to perform multi-scale image transformation on the two-dimensional isomagnetic map sequence, resulting in various transformed images of the two-dimensional isomagnetic map sequence;
[0010] A multi-dimensional magnetocardiogram feature extraction module is used to extract multi-dimensional magnetocardiogram feature sets from various transformed images based on dynamic masks of multiple regions of interest and two-dimensional isomagnetic sequences.
[0011] The preprocessing includes power frequency filtering, baseline correction, and bad channel removal to obtain clean, analytically usable multichannel magnetocardiogram signals. ,in n represents the channel index, and N is a positive integer, i.e., the number of channels. t represents a time point, and T is a positive integer; the spatial interpolation algorithm includes at each time point ,Will Magnetic field strength values measured by each channel As a known point, based on the known coordinates of the sensor on the measurement plane The continuous magnetic field distribution of the entire measurement area is reconstructed through spatial interpolation. .
[0012] The generation of the isomagnetic map sequence includes the interpolated continuous magnetic field distribution. At every point in time Visualization is achieved by plotting a two-dimensional isomagnetic map using colors or contour lines to represent magnetic field strength. Arranging the isomagnetic maps at all moments throughout the entire cardiac cycle, from before the P wave to after the T wave, in chronological order constitutes a complete two-dimensional temporal isomagnetic map sequence. This sequence dynamically characterizes the propagation process of electromagnetic activity during cardiac depolarization and repolarization.
[0013] The multi-channel magnetocardiogram (MCG) signal is a 64-channel MCG signal.
[0014] The key bands are the seven key bands: P-wave pre-wave, P-wave, PQ band, QRS band, ST band, T-wave, and T-wave post-wave.
[0015] The automatic generation of a dynamic mask with multiple regions of interest for each band includes: automatically extracting the maximum values of the positive and negative magnetic fields at each moment, taking 90% of the maximum values of the positive and negative magnetic fields as the regions of interest, and then combining all moments of each band into a three-dimensional volume of interest as an automated dynamic mask.
[0016] The multi-scale image transformation specifically includes: wavelet transform: using the Daubechies wavelet basis for multi-scale decomposition, and independently performing high-pass and low-pass filtering on each of the three dimensions to obtain 8 combined images, the three dimensions being spatial X dimension, spatial Y dimension and time T dimension; Gaussian filtering: enhancing the spatial continuity of the magnetic field distribution by smoothing noise; nonlinear transformation: including performing square operation, square root operation, logarithmic operation and / or exponential operation on the original two-dimensional time-series isomagnetic map; gradient calculation: calculating the gradient of the original two-dimensional time-series isomagnetic map.
[0017] The multi-dimensional magnetocardiogram feature extraction module is used to extract multi-dimensional features, including first-order statistical features, shape features, and texture features. The first-order statistical features include one or more of the following: energy, total energy, entropy, minimum eigenvalue, 10th percentile eigenvalue, 90th percentile eigenvalue, maximum eigenvalue, average eigenvalue, median eigenvalue, interquartile range, intensity range, mean absolute deviation, robust mean absolute deviation, root mean square, standard deviation, skewness, kurtosis, variance, and uniformity. The shape features include one or more of the following: grid surface, pixel surface, perimeter, perimeter ratio, sphericity, spherical non-uniformity, maximum 2D diameter, principal axis length, minor axis length, and elongation. The texture features include one or more of the following: gray-level co-occurrence matrix, gray-level size region matrix, gray-level run matrix, adjacent gray-level difference matrix, and gray-level dependency matrix.
[0018] The technical effects of this invention are as follows: This invention is a multi-dimensional magnetocardiogram (MCC) feature extraction system based on dynamic masks of multiple regions of interest and multi-scale image transformation. It can achieve automated multi-dimensional MCC feature extraction by extracting MCC features based on multi-scale image transformation and regions of interest. It has the advantages of strong feature representation ability and high robustness. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the process executed by the multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation of the present invention. Figure 1 The process includes: Step 1, drawing a two-dimensional temporal isomagnetic map; Step 2, creating an automated dynamic mask; Step 3, implementing multi-scale image transformation; and Step 4, extracting multi-dimensional magnetocardiographic features. Detailed Implementation
[0020] The following is in conjunction with the attached diagram ( Figure 1 The present invention will be described in conjunction with the embodiments.
[0021] Figure 1 This is a schematic diagram of the workflow executed by the multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic masks and multi-scale image transformation according to the present invention. (Reference) Figure 1As shown, a multi-dimensional magnetocardiogram (MCC) feature extraction system based on multi-region-of-interest (MOI) dynamic masks and multi-scale image transformation includes: a module for drawing two-dimensional temporal isomagnetic maps (MTAs), used to construct the cardiac magnetic field distribution in preprocessed multi-channel MMC signals using spatial interpolation algorithms, drawing two-dimensional MMAs of the entire single-heartbeat cycle, and arranging them in chronological order to form a complete two-dimensional MMA sequence to characterize the dynamic changes in cardiac electromagnetic activity; an automated dynamic mask creation module, used to divide the MMC cycle into multiple key bands based on the two-dimensional MMA sequence and MMC physiological characteristics, and automatically generate multi-region-of-interest (MOI) dynamic masks for each band; a multi-scale image transformation module, used to perform multi-scale image transformations on the two-dimensional MMA sequence to obtain various transformed images of the two-dimensional MMA sequence; and a multi-dimensional MMC feature extraction module, used to extract a multi-dimensional MMC feature set based on the multi-MOI dynamic masks and various transformed images of the two-dimensional MMA sequence.
[0022] The preprocessing includes power frequency filtering, baseline correction, and bad channel removal to obtain clean, analytically usable multichannel magnetocardiogram signals. ,in n represents the channel index, and N is a positive integer, i.e., the number of channels. t represents a time point, and T is a positive integer; the spatial interpolation algorithm includes at each time point ,Will Magnetic field strength values measured by each channel As a known point, based on the known coordinates of the sensor on the measurement plane The continuous magnetic field distribution of the entire measurement area is reconstructed through spatial interpolation. .
[0023] The generation of the isomagnetic map sequence includes the interpolated continuous magnetic field distribution. At every point in time Visualization is achieved by plotting a two-dimensional isomagnetic map using colors or contour lines to represent magnetic field strength. Arranging the isomagnetic maps at all moments throughout the entire cardiac cycle, from before the P wave to after the T wave, in chronological order constitutes a complete two-dimensional temporal isomagnetic map sequence. This sequence dynamically characterizes the propagation process of electromagnetic activity during cardiac depolarization and repolarization.
[0024] The multi-channel magnetic resonance imaging (MRI) signal is a 64-channel MRI signal. The multiple key segments are the seven key segments: pre-P wave, P wave, PQ segment, QRS complex, ST segment, T wave, and post-T wave (each letter is a component of the unified naming and definition of each wave and segment in the electrocardiogram).
[0025] The automatic generation of a dynamic mask with multiple regions of interest for each band includes: automatically extracting the maximum values of the positive and negative magnetic fields at each moment, taking 90% of the maximum values of the positive and negative magnetic fields as the regions of interest, and then combining all moments of each band into a three-dimensional volume of interest as an automated dynamic mask.
[0026] The multi-scale image transformation specifically includes: wavelet transform: using the Daubechies wavelet basis for multi-scale decomposition, and independently performing high-pass and low-pass filtering on each of the three dimensions to obtain 8 combined images, the three dimensions being spatial X dimension, spatial Y dimension and time T dimension; Gaussian filtering: enhancing the spatial continuity of the magnetic field distribution by smoothing noise; nonlinear transformation: including performing square, square root (using the absolute value of the original image), logarithmic (using the absolute value of the original image + 1) and exponential operations on the original two-dimensional time-series isomagnetic map, scaling the value to the original range; gradient calculation: calculating the gradient of the original two-dimensional time-series isomagnetic map.
[0027] The multi-dimensional magnetocardiogram feature extraction module is used to extract multi-dimensional features, including first-order statistical features, shape features, and texture features. The first-order statistical features include one or more of the following: energy, total energy, entropy, minimum eigenvalue, 10th percentile eigenvalue, 90th percentile eigenvalue, maximum eigenvalue, average eigenvalue, median eigenvalue, interquartile range, intensity range, mean absolute deviation, robust mean absolute deviation, root mean square, standard deviation, skewness, kurtosis, variance, and uniformity. The shape features include one or more of the following: grid surface, pixel surface, perimeter, perimeter ratio, sphericity, spherical non-uniformity, maximum 2D diameter, principal axis length, minor axis length, and elongation. The texture features include one or more of the following: gray-level co-occurrence matrix, gray-level size region matrix, gray-level run matrix, adjacent gray-level difference matrix, and gray-level dependency matrix.
[0028] This invention discloses a multi-dimensional magnetocardiogram (MCG) feature extraction system based on a dynamic mask with multiple regions of interest (ROIs) and multi-scale image transformation. The system performs the following steps: 1) Drawing a two-dimensional temporal isomagnetic map: Based on preprocessed multi-channel MCG signals, the cardiac magnetic field distribution is constructed through spatial interpolation, and a two-dimensional isomagnetic map sequence for the entire single-heartbeat cycle is drawn; 2) Creating an automated dynamic mask: The MCG cycle is divided into 7 key bands according to the physiological characteristics of MCG, and a dynamic mask with multiple regions of interest (ROIs) is automatically generated for each band; 3) Performing multi-scale image transformation: Wavelet transform, Gaussian filtering, nonlinear transformation, and gradient calculation are performed on the two-dimensional temporal isomagnetic map; 4) Extracting multi-dimensional MCG features: Based on the dynamic ROI mask, a multi-dimensional MCG feature set including first-order statistical features, shape features, and texture features is extracted. Compared with traditional methods, this invention can automatically extract multi-dimensional MCG features and has the advantages of strong feature representation ability and high robustness.
[0029] like Figure 1 As shown, the multi-dimensional magnetocardiogram feature extraction method based on multi-region-of-interest dynamic mask and multi-scale image transformation of the present invention includes the following steps:
[0030] The first step is to draw a two-dimensional temporal isomagnetic map.
[0031] First, data preprocessing is performed: the raw 64-channel magnetocardiogram (MCC) signals are preprocessed, including power frequency filtering, baseline correction, and bad sector removal, to obtain clean, analysis-ready multi-channel MCC signals. ),in Indicates the channel index. Indicates a point in time.
[0032] Spatial interpolation: at each time point ,Will Magnetic field strength values measured by each channel As a known point, based on the known coordinates of the sensor on the measurement plane The continuous magnetic field distribution of the entire measurement area is reconstructed using a spatial interpolation algorithm. .
[0033] Generating an isomagnetic map sequence: This involves analyzing the interpolated magnetic field distribution. At every point in time Visualize the data and plot it as a two-dimensional isomagnetic map (i.e., using color or contour lines to represent magnetic field strength). Arrange the isomagnetic maps of all moments throughout the entire cardiac cycle (from before the P wave to after the T wave) in chronological order to form a complete two-dimensional temporal isomagnetic map sequence. This sequence dynamically characterizes the propagation process of electromagnetic activity during cardiac depolarization and repolarization.
[0034] The second step is to create an automated dynamic mask.
[0035] Band segmentation: Based on synchronously recorded electrocardiograms or typical waveforms extracted from magnetocardiogram signals, a complete cardiac cycle is automatically divided into the following 7 key bands: (P wave preamp) (P wave) (PQ segment), (QRS wave) (ST section) (T wave) (After the T wave).
[0036] Region of Interest Extraction at Determined Time: For Bands any specific moment within isomagnetic map Find the global maximum value of the magnetic field strength in the graph. (Positive) and global minimum (Negative pole). The positive pole region of interest is the region containing all magnetic field strength values in the graph. The pixels are marked as 1, and the rest are marked as 0, forming the positive binary mask at that moment. Similarly, the region of interest at the negative pole is the region containing all magnetic field strength values in the figure. The pixels are marked as 1, and the rest are marked as 0, forming the negative binary mask at that moment. .
[0037] Constructing a 3D dynamic mask: using a band The positive electrode masks at all times within the time frame are logically ORed and merged into a three-dimensional dynamic positive electrode mask. Similarly, the negative electrode dynamic mask for this band is obtained. .
[0038]
[0039]
[0040] Ultimately, 14 (7 positive + 7 negative) three-dimensional automated dynamic masks with 7 bands were obtained for subsequent feature extraction.
[0041] The third step is to perform multi-scale image transformation.
[0042] 1. Wavelet Transform: The original image sequence is processed using the Daubechies wavelet basis. Perform a three-dimensional discrete wavelet transform. Apply a single high-pass (H) and low-pass (L) filter decomposition to the three dimensions of space X, space Y, and time T respectively. The decomposition yields... The sub-band coefficients are denoted as: LLL, LLH, LHL, LHH, HLL, HLH, HHL, HHH. Each letter represents the filter type in one dimension (L: low-pass, H: high-pass), in the order (X, Y, T). Reconstructing each sub-band coefficient yields wavelet transform images of eight different frequency combinations. ,in Subband is a subband, and wavelet is a wavelet transform.
[0043] 2. Gaussian filtering: Convolves the original image with a three-dimensional Gaussian kernel to smooth noise and enhance spatial continuity.
[0044] in, The standard deviation is A three-dimensional Gaussian function, where * denotes the convolution operation. This is the image after Gaussian filtering.
[0045] 3. Nonlinear Transformation: Applying a nonlinear operation to each pixel value I of the original image sequence generates four types of transformed images:
[0046] Square Transform Image expression:
[0047] Square root transform image expression:
[0048] Logarithmic Transformation Image expression: (Add 1 to avoid taking the logarithm of 0)
[0049] Exponential Transform Image expression:
[0050] in This is the original image. To ensure numerical stability, all transformed images are rescaled to the data range of the original image.
[0051] 4. Gradient Calculation
[0052] Calculate the spatial gradient of the original image sequence This is to highlight the boundaries and regions of magnetic field changes.
[0053]
[0054] Where I represents the pixel values of the original image sequence. In practical digital image processing, gradient operators such as the Sobel operator or the Prewitt operator are typically used for approximate calculations. The final result is an image of the gradient magnitude. .
[0055] The fourth step involves extracting multi-dimensional features based on multi-scale image transformation and regions of interest.
[0056] 1. Feature Extraction
[0057] For each region of interest mask and the corresponding image (e.g.) Features are extracted from the following three dimensions:
[0058] First-order statistical features: calculated from the histogram of image pixel values, without considering spatial relationships.
[0059] energy expression: Where N is the number of channels and i is the sequence number. It refers to image intensity.
[0060] entropy expression: ,in It is strength The probability of.
[0061] Skewness expression: ,in It is the mean. It is the standard deviation.
[0062] Shape features: Calculated from the geometric properties of the region of interest mask itself, independent of image intensity.
[0063] Mesh surface: The sum of the surface areas of all voxels within the region of interest.
[0064] sphericity expression: ,in It is volume. It is the surface area. It measures how close the region of interest is to a sphere.
[0065] Elongation: Calculated based on the principal and minor axis lengths of the smallest circumscribed ellipsoid of the region of interest.
[0066] Texture features: Describe the spatial distribution pattern of pixel intensity.
[0067] Gray-level co-occurrence matrix features: such as contrast Where i is the grayscale value of the first pixel, and j is the grayscale value of the second pixel. It includes gray-level co-occurrence matrix, correlation, energy, homogeneity, etc.
[0068] Features of the grayscale run-length matrix: such as the advantages of short runs and the advantages of long runs.
[0069] The final feature set combines the first-order, shape, and texture features extracted from all images (original + transformed) of all regions of interest (ROIs) to form a comprehensive, multi-dimensional magnetocardiogram (MCG) feature vector based on multi-scale image transformations and ROIs. This vector is used for subsequent machine learning model training to assist in the diagnosis, risk stratification, or efficacy assessment of cardiac diseases (such as myocardial ischemia, arrhythmia, cardiomyopathy, etc.). In practical applications, feature selection or dimensionality reduction is usually required to remove redundancy and construct the optimal classification or regression model.
[0070] A multi-dimensional magnetocardiogram feature extraction method based on multi-region-of-interest dynamic masks and multi-scale image transformation includes the following steps:
[0071] Step 1: Draw a two-dimensional temporal isomagnetic map: Based on the preprocessed multi-channel cardiac magnetic signals, the cardiac magnetic field distribution is constructed through a spatial interpolation algorithm, and a two-dimensional isomagnetic map of the entire cycle of a single heartbeat is drawn. The maps are arranged in chronological order to form a complete two-dimensional isomagnetic map sequence to characterize the dynamic changes in cardiac electromagnetic activity.
[0072] Step 2: Create an automated dynamic mask: Based on the physiological characteristics of the electrocardiogram signal, the electrocardiogram cycle is divided into 7 key segments: pre-P wave, P wave, PQ segment, QRS complex, ST segment, T wave, and post-T wave (each letter is a component that is uniformly named and defined for each wave and segment of the electrocardiogram); for each segment, an automated dynamic mask with multiple regions of interest is automatically generated.
[0073] Step 3: Perform multi-scale image transformation: To improve the robustness of feature extraction, perform multi-scale image transformation on the two-dimensional temporal isomagnetic map;
[0074] Step 4: Extract multi-dimensional magnetocardiogram features: Extract multi-dimensional magnetocardiogram feature sets based on dynamic region of interest masks.
[0075] Furthermore, in step 2, the region of interest is obtained by automatically extracting the maximum values of the positive and negative magnetic fields at each moment, taking 90% of the maximum values of the positive and negative magnetic fields as the region of interest, and then combining all moments of each band into a three-dimensional region of interest as an automated dynamic mask.
[0076] Further, step 3, the multi-scale image transformation, specifically includes: wavelet transform: using the Daubechies wavelet basis for multi-scale decomposition, performing high-pass and low-pass filtering independently on each of the three dimensions (spatial X, spatial Y, and time T) to obtain 8 combined images; Gaussian filtering: smoothing noise and enhancing the spatial continuity of the magnetic field distribution; nonlinear transformation: including performing squaring, square root (using the absolute value of the original image), logarithmic (using the absolute value of the original image + 1), and exponential operations on the original two-dimensional time-series isomagnetic map, scaling the value to the original range; gradient calculation: calculating the gradient of the original two-dimensional time-series isomagnetic map.
[0077] Further, step 4, the multi-dimensional magnetic field features, includes first-order statistical features, shape features, and texture features: First-order statistical features include energy, total energy, entropy, minimum eigenvalue, 10th percentile eigenvalue, 90th percentile eigenvalue, maximum eigenvalue, average eigenvalue, median eigenvalue, interquartile range, intensity range, mean absolute deviation, robust mean absolute deviation, root mean square, standard deviation, skewness, kurtosis, variance, and uniformity. Shape features include grid surface, pixel surface, perimeter, perimeter ratio, sphericity, spherical inhomogeneity, maximum 2D diameter (i.e., the maximum paired Euclidean distance between grid vertices on the tumor surface in the slice plane), principal axis length, minor axis length, and elongation. Texture features include gray-level co-occurrence matrix, gray-level size region matrix, gray-level run matrix, adjacent gray-level difference matrix, and gray-level dependency matrix.
[0078] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, and / or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.
Claims
1. A multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic masks and multi-scale image transformation, characterized in that, include: The module for drawing two-dimensional temporal isomagnetic maps is used to construct the cardiac magnetic field distribution in the preprocessed multi-channel magnetic field signals through spatial interpolation algorithms, draw two-dimensional isomagnetic maps of the entire cycle of a single heartbeat, and arrange them in chronological order to form a complete two-dimensional isomagnetic map sequence to characterize the dynamic changes in cardiac electromagnetic activity. An automated dynamic mask module is created to divide the magnetic cycle into multiple key bands based on the two-dimensional isomagnetic map sequence and the magnetic-cardiographic physiological characteristics. For each band, a dynamic mask with multiple regions of interest is automatically generated. A multi-scale image transformation module is implemented to perform multi-scale image transformation on the two-dimensional isomagnetic map sequence, resulting in various transformed images of the two-dimensional isomagnetic map sequence; A multi-dimensional magnetocardiogram feature extraction module is used to extract multi-dimensional magnetocardiogram feature sets from various transformed images based on dynamic masks of multiple regions of interest and two-dimensional isomagnetic sequences.
2. The multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation according to claim 1, characterized in that, The preprocessing includes power frequency filtering, baseline correction, and bad channel removal to obtain clean, analytically usable multichannel magnetocardiogram signals. ,in n represents the channel index, and N is a positive integer, i.e., the number of channels. t represents a time point, and T is a positive integer; the spatial interpolation algorithm includes at each time point ,Will Magnetic field strength values measured by each channel As a known point, based on the known coordinates of the sensor on the measurement plane The continuous magnetic field distribution of the entire measurement area is reconstructed through spatial interpolation. .
3. The multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation according to claim 1, characterized in that, The generation of the isomagnetic map sequence includes the interpolated continuous magnetic field distribution. At every point in time Visualization is achieved by plotting a two-dimensional isomagnetic map using colors or contour lines to represent magnetic field strength. Arranging the isomagnetic maps at all moments throughout the entire cardiac cycle, from before the P wave to after the T wave, in chronological order constitutes a complete two-dimensional temporal isomagnetic map sequence. This sequence dynamically characterizes the propagation process of electromagnetic activity during cardiac depolarization and repolarization.
4. The multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation according to claim 1, characterized in that, The multi-channel magnetocardiogram (MCG) signal is a 64-channel MCG signal.
5. The multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation according to claim 1, characterized in that, The key bands are the seven key bands: P-wave pre-wave, P-wave, PQ band, QRS band, ST band, T-wave, and T-wave post-wave.
6. The multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation according to claim 1, characterized in that, The automatic generation of a dynamic mask with multiple regions of interest for each band includes: automatically extracting the maximum values of the positive and negative magnetic fields at each moment, taking 90% of the maximum values of the positive and negative magnetic fields as the regions of interest, and then combining all moments of each band into a three-dimensional volume of interest as an automated dynamic mask.
7. The multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation according to claim 1, characterized in that, The multi-scale image transformation specifically includes: wavelet transform: using the Daubechies wavelet basis for multi-scale decomposition, and independently performing high-pass and low-pass filtering on each of the three dimensions to obtain 8 combined images, the three dimensions being spatial X dimension, spatial Y dimension and time T dimension; Gaussian filtering: enhancing the spatial continuity of the magnetic field distribution by smoothing noise; nonlinear transformation: including performing square operation, square root operation, logarithmic operation and / or exponential operation on the original two-dimensional time-series isomagnetic map; gradient calculation: calculating the gradient of the original two-dimensional time-series isomagnetic map.
8. The multi-dimensional magnetocardiogram feature extraction system based on multi-region-of-interest dynamic mask and multi-scale image transformation according to claim 1, characterized in that, The multi-dimensional magnetocardiogram (MCC) feature extraction module is used to extract multi-dimensional MCC features, including first-order statistical features, shape features, and texture features. The first-order statistical features include one or more of the following: energy, total energy, entropy, minimum eigenvalue, 10th percentile eigenvalue, 90th percentile eigenvalue, maximum eigenvalue, average eigenvalue, median eigenvalue, interquartile range, intensity range, mean absolute deviation, robust mean absolute deviation, root mean square, standard deviation, skewness, kurtosis, variance, and uniformity. The shape features include one or more of the following: mesh surface, pixel surface, perimeter, perimeter ratio, sphericity, spherical non-uniformity, maximum 2D diameter, principal axis length, minor axis length, and elongation. The texture features include one or more of the following: gray-level co-occurrence matrix, gray-level size region matrix, gray-level run matrix, adjacent gray-level difference matrix, and gray-level dependency matrix.