A magnetocardiogram feature selection method based on comprehensive performance evaluation framework

Abstract

The application discloses a magnetocardiogram feature selection method based on a comprehensive performance evaluation framework, comprising: evaluating magnetocardiogram feature comprehensive performance, deleting magnetocardiogram features with strong mutual correlation, and selecting the optimal number of magnetocardiogram features. Compared with the feature selection method based on single performance evaluation, the method has the advantages that the selected magnetocardiogram features have distinguishability and anti-interference, the modeling index is higher, and the robustness is better.

Description

Technical Field

[0001] This invention belongs to the field of biomedical signal analysis and relates to a method for selecting cardiac magnetic features based on a comprehensive performance evaluation framework. Background Technology

[0002] A SERF (Spin Exchange Relaxation Free) atomic magnetometer-based magnetic resonance imaging (MRIE) device is a novel cardiac function testing instrument characterized by its non-invasiveness, radiation-free operation, non-contact nature, room temperature operation, high sensitivity, and high temporal resolution. Studies have shown that SERF-based MRIE is superior to other detection methods such as electrocardiography (ECG) in diagnosing myocardial ischemia, arrhythmias, and fetal heart conditions, demonstrating significant value for medical research and applications.

[0003] One of the crucial steps in using magnetocardiography (MCG) imaging devices for disease-assisted detection is selecting an effective set of MCG features. Traditional methods for this involve calculating the discriminability index of each MCG feature, ranking the features in the set based on their discriminability, and then selecting a subset of features with the highest discriminability index. However, this method has limitations. It doesn't account for the impact of environmental noise, subject motion artifacts, and inconsistencies in the relative spatial relationship between the sensor array and the hearts of different subjects during MCG acquisition. This results in a selected MCG feature set lacking robustness against interference, exhibiting low modeling performance, and demonstrating insufficient model robustness. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for selecting magnetic signature features based on a comprehensive performance evaluation framework. This method solves the problems of existing magnetic signature feature selection methods, such as lack of anti-interference capability, low modeling index, and insufficient model robustness.

[0005] To achieve the above and other related objectives, the technical solution adopted by the present invention is as follows: A method for selecting magnetic cardiomyocyte features based on a comprehensive performance evaluation framework includes the following steps: Step 1: Evaluate the overall performance of the magnetic cardiomyocyte features: For the input set of magnetic cardiomyocyte features, calculate the distinguishability index and anti-interference ability index of the magnetic cardiomyocyte features respectively, and calculate the overall performance evaluation index of the magnetic cardiomyocyte features based on these two indexes. Step 2: Delete strong cross-correlation features: Calculate the linear correlation coefficient between all pairs of magnetic features in the magnetic feature set, set a strong cross-correlation threshold, and delete the magnetic features with lower comprehensive performance evaluation index when the linear correlation coefficient between two magnetic features is higher than the strong cross-correlation threshold. Step 3: Select the optimal number of magnetic signature features: Plot the relationship curve between the number of magnetic signature features and the accuracy of the disease-assisted detection model, and select the number of magnetic signature features corresponding to the optimal point of the accuracy distribution curve. The optimal point of the accuracy distribution curve is the point where the accuracy distribution curve first reaches a stable state.

[0006] Furthermore, the distinguishability index of the magnetic signature is calculated using any one of the following methods: T-test, relative entropy, chi-square test, Mann-Whitney test, and Bhattacharyya distance.

[0007] Furthermore, the anti-interference capability index of the cardiac magnetic features includes three parts: anti-interference index against electromagnetic noise sources in the hospital clinical environment, anti-interference index against motion artifacts of the subjects, and anti-interference index against inconsistencies in the relative three-dimensional spatial positional relationship between the sensor array and the hearts of different subjects.

[0008] Furthermore, the method for calculating the anti-interference index against electromagnetic noise sources in the hospital clinical environment involves collecting magnetic field data for the same subject under conditions of no electromagnetic noise interference and conditions of presence of electromagnetic noise interference, and then basing the data on... Calculate the immunity index of cardiac magnetic characteristics to electromagnetic noise sources in the hospital clinical environment. ,in These are the characteristic values ​​obtained by collecting magnetocardiogram data under conditions of no electromagnetic noise interference. These are characteristic values ​​obtained by collecting magnetocardiogram data under conditions of electromagnetic noise interference.

[0009] Furthermore, the method for calculating the anti-interference index of motion artifacts in the subject is as follows: for the same subject, magnetic field data are collected under conditions of no motion artifacts and with motion artifacts, and the results are calculated based on... Calculate the resistance index of magnetocardiographic features to motion artifacts in subjects. ,in These are the feature values ​​obtained by acquiring magnetocardiogram data under conditions free from motion artifact interference. These are feature values ​​obtained by collecting magnetocardiogram data under conditions of motion artifact interference. The motion artifacts of the subjects include head movement, limb movement, deep breathing, and coughing.

[0010] Furthermore, the method for calculating the anti-interference index of the inconsistency in the relative three-dimensional spatial positional relationship between the sensor array and the hearts of different subjects is as follows: For the same subject, the inconsistency in the relative three-dimensional spatial positional relationship between the sensor array and the hearts of different subjects is simulated by adjusting the horizontal front-back and left-right distances and vertical up-down distances between the sensor array and the subject. Cardiac magnetic data are collected at the initial position of the sensor array and under conditions of horizontal front-back and left-right and vertical up-down movement of the sensor array, and based on... Calculate the anti-interference index of the inconsistency between the magnetic signature of the heart and the relative three-dimensional spatial position between the sensor array and the hearts of different subjects. ,in These are feature values ​​obtained by acquiring magnetocardiogram data under the initial position of the sensor array. These are the characteristic values ​​obtained by collecting magnetocardiogram data when the sensor array moves horizontally forward and backward, left and right, and vertically up and down.

[0011] Furthermore, the comprehensive performance evaluation indicators ,in To assign to the distinguishability index and anti-interference capability indicators The weight, The anti-interference capability index A is a comprehensive value calculated based on A1, A2 and A3.

[0012] Compared with the prior art, the present invention has the following advantages: This invention considers the impact of three interference sources on magnetic field features during the acquisition of magnetic field data: the first interference source is electromagnetic interference encountered by the magnetic field imaging device in a clinical environment, which interferes with the quality of magnetic field data and thus affects the stable distribution of magnetic field features; the second interference source is head movement, limb movement, deep breathing, and coughing by the subject during the acquisition process, which introduces motion artifacts and affects the stable distribution of magnetic field features; the third interference source is the inconsistency in the relative three-dimensional spatial position between the sensor array and the hearts of different subjects, which can lead to instability in the distribution of magnetic field features in healthy individuals due to differences in subject body size. In this invention, by using comprehensive performance evaluation features, the selected feature set simultaneously possesses discriminability and anti-interference ability, improves modeling indicators, and enhances robustness. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for selecting magnetic signature features based on a comprehensive performance evaluation framework according to the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] like Figure 1 As shown, the magnetic signature selection method based on a comprehensive performance evaluation framework of the present invention includes the following steps: The first step is to assess the overall performance of the magnetic signature.

[0016] For the input set of magnetic cardiomyocyte features, calculate the distinguishability index and anti-interference ability index of the magnetic cardiomyocyte features respectively, and calculate the comprehensive performance evaluation index of the magnetic cardiomyocyte features based on these two indexes. Finally, sort the magnetic cardiomyocyte features according to the comprehensive performance evaluation index.

[0017] The set of magnetic cardiomyocyte features refers to multiple categories of magnetic cardiomyocyte features extracted from one-dimensional butterfly diagrams, two-dimensional isomagnetic diagrams, and two-dimensional pseudo-current density diagrams.

[0018] The discriminability index of the magnetic signature can be calculated using the T-test, relative entropy, chi-square test, Mann-Whitney test, and Bhattacharyya distance.

[0019] The anti-interference capability index of the cardiac magnetic features includes three parts: anti-interference of electromagnetic noise sources in the hospital clinical environment (interference source 1), anti-interference of motion artifacts of the subjects (interference source 2), and anti-interference of the inconsistency of the relative three-dimensional spatial position relationship between the sensor array and the hearts of different subjects (interference source 3).

[0020] The method for evaluating the anti-interference index of the electromagnetic noise source (interference source 1) in the hospital clinical environment is as follows: For the same subject, magnetic field data are collected under conditions of no electromagnetic noise interference and conditions of presence of electromagnetic noise interference, and then... Calculate the anti-interference index of magnetic resonance imaging characteristics against interference source 1. ),in These are the characteristic values ​​obtained by collecting magnetocardiogram data under conditions of no electromagnetic noise interference. These are characteristic values ​​obtained by collecting magnetocardiogram data under conditions of electromagnetic noise interference.

[0021] The method for evaluating the anti-interference index of motion artifacts (interference source 2) in the subject is as follows: for the same subject, magnetic field data are collected under conditions of no motion artifacts and with motion artifacts, and then... Calculate the anti-interference index of magnetic resonance imaging characteristics against interference source 2 ( ),in These are the feature values ​​obtained by acquiring magnetocardiogram data under conditions free from motion artifact interference. These are feature values ​​obtained from magnetocardiogram data collected under conditions of motion artifact interference. Interference source 2 includes head movement, limb movement, deep breathing, and coughing.

[0022] The anti-interference index evaluation method for the inconsistency of the relative three-dimensional spatial position relationship between the sensor array and the hearts of different subjects (interference source 3) is as follows: For the same subject, the inconsistency of the relative three-dimensional spatial position relationship between the sensor array and the hearts of different subjects is simulated by adjusting the horizontal front-back and left-right distance and the vertical up-down distance between the sensor array and the subject. Cardiac magnetic data are collected under the conditions of the initial position of the sensor array and the horizontal front-back and left-right distance and vertical up-down distance of the sensor array, respectively, and based on... Calculate the anti-interference index of magnetic resonance imaging characteristics against interference source 3 ( ),in These are feature values ​​obtained by acquiring magnetocardiogram data under the initial position of the sensor array. These are the characteristic values ​​obtained by collecting magnetocardiogram data under conditions of horizontal forward and backward, left and right, and vertical up and down movement of the sensor array.

[0023] The comprehensive performance evaluation index (C) of the magnetic signature features refers to the simultaneous consideration of the distinguishability (D) and anti-interference index (A) of the magnetic signature features. ,in To be assigned and The weight, , The optimal value needs to be obtained through experimental calibration, where the anti-interference capability index A is a comprehensive value calculated based on A1, A2 and A3.

[0024] The second step is to remove strongly correlated magnetic features.

[0025] Calculate the linear correlation coefficients between all pairs of features in the set of magnetic cardiomyocyte features output in the first step, and then delete the strongly cross-correlated features.

[0026] The linear correlation coefficient refers to the Pearson linear correlation coefficient.

[0027] The strong cross-correlation is defined with a linear correlation coefficient of 0.5 as the threshold for strong cross-correlation.

[0028] The process of deleting strongly cross-correlated features requires consideration of the combined performance index of the two features, and features with lower combined performance index should be deleted.

[0029] The third step is to select the optimal number of magnetic signature features.

[0030] Plot the relationship curve between the number of magnetic signature features and the accuracy of the disease-aided detection model, and select the number of magnetic signature features corresponding to the optimal point of the accuracy distribution curve.

[0031] The optimal point of the accuracy distribution curve is the point where the accuracy distribution curve first reaches a stable state.

[0032] In summary, the magnetic resonance imaging (MRI) feature selection method based on a comprehensive performance evaluation framework described in this invention sequentially evaluates the comprehensive performance of MRI features, removes strongly correlated MRI features, and selects the optimal number of MRI features, ultimately achieving effective MRI feature selection. Its advantages include selectable and interference-resistant MRI features, higher modeling metrics, and superior robustness. Therefore, this invention effectively overcomes the shortcomings of existing technologies and has high medical application value.

[0033] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for selecting magnetic signature features based on a comprehensive performance evaluation framework, characterized in that, The method includes the following steps: Step 1: Evaluate the comprehensive performance of the cardiac magnetic features: For the input set of cardiac magnetic features, calculate the distinguishability index and anti-interference ability index of the cardiac magnetic features respectively, and calculate the comprehensive performance evaluation index of the cardiac magnetic features based on these two indexes; The anti-interference ability index of the cardiac magnetic features includes three parts, namely, the anti-interference index against electromagnetic noise sources in the hospital clinical environment, the anti-interference index against motion artifacts of the subjects, and the anti-interference index against the inconsistency of the relative three-dimensional spatial position relationship between the sensor array and the hearts of different subjects; Step 2: Delete strong cross-correlation magnetic features: Calculate the linear correlation coefficient between all pairs of magnetic features in the magnetic feature set, set a strong cross-correlation threshold, and delete the magnetic features with lower comprehensive performance evaluation index when the linear correlation coefficient between two magnetic features is higher than the strong cross-correlation threshold. Step 3: Select the optimal number of cardiac magnetic features: Plot the relationship curve between the number of cardiac magnetic features and the accuracy of the disease-assisted detection model, and select the number of cardiac magnetic features corresponding to the optimal point of the accuracy distribution curve. The optimal point of the accuracy distribution curve is the point where the accuracy distribution curve first reaches a stable state. The method for calculating the anti-interference index against electromagnetic noise sources in the hospital clinical environment involves collecting magnetic field data from the same subject under conditions of no electromagnetic noise interference and conditions of presence of electromagnetic noise interference, and then basing the data on... Calculate the immunity index of cardiac magnetic characteristics to electromagnetic noise sources in the hospital clinical environment. ,in These are the characteristic values ​​obtained by collecting magnetocardiogram data under conditions of no electromagnetic noise interference. These are the characteristic values ​​obtained by collecting magnetocardiogram data under conditions of electromagnetic noise interference. The method for calculating the anti-interference index of motion artifacts in the subjects is as follows: for the same subject, magnetic field data are collected under conditions of no motion artifacts and with motion artifacts, and then... Calculate the resistance index of magnetocardiographic features to motion artifacts in subjects. ,in These are the feature values ​​obtained from magnetocardiogram data acquired in the absence of motion artifact interference sources. These are the feature values ​​obtained by collecting magnetocardiogram data under conditions of motion artifact interference. The motion artifacts of the subjects include head movement, limb movement, deep breathing, and coughing. The method for calculating the anti-interference index of the inconsistency in the relative three-dimensional spatial positional relationship between the sensor array and the hearts of different subjects is as follows: For the same subject, the inconsistency in the relative three-dimensional spatial positional relationship between the sensor array and the heart of different subjects is simulated by adjusting the horizontal front-back and left-right distances and vertical up-down distances between the sensor array and the subject. Cardiac magnetic data are collected at the initial position of the sensor array and under conditions of horizontal front-back and left-right and vertical up-down movement of the sensor array, and based on... Calculate the anti-interference index of the inconsistency between the magnetic signature of the heart and the relative three-dimensional spatial position between the sensor array and the hearts of different subjects. ,in These are feature values ​​obtained by acquiring magnetocardiogram data under the initial position of the sensor array. These are the characteristic values ​​obtained by collecting magnetocardiogram data when the sensor array moves horizontally forward and backward, left and right, and vertically up and down.

2. The method for selecting magnetic signature features based on a comprehensive performance evaluation framework according to claim 1, characterized in that, The discriminability index of the magnetic signature was calculated using any one of the following methods: T-test, relative entropy, chi-square test, Mann-Whitney test, and Bhattacharyya distance.

3. The method for selecting magnetic signature features based on a comprehensive performance evaluation framework according to claim 1, characterized in that, The comprehensive performance evaluation indicators ,in To assign to the distinguishability index and anti-interference capability indicators The weight, The anti-interference capability index A is a comprehensive value calculated based on A1, A2 and A3.

Images