Intelligent method for removing artifacts from cardiac magnetic resonance images

Abstract

The present application relates to the technical field of image processing, in particular to a kind of cardiac magnetic resonance image artifact intelligent removal method, comprising: obtaining the standard cardiac magnetic resonance gray image and the cardiac magnetic resonance gray image to be processed;Obtain the artifact graph layer quantity in the cardiac magnetic resonance gray image to be processed;Obtain the artifact graph layer superposition loss curve of the cardiac magnetic resonance gray image to be processed;Obtain the estimated blur kernel weight of the cardiac magnetic resonance gray image to be processed;According to the estimated blur kernel weight, eliminate artifact in the cardiac magnetic resonance gray image to be processed.The present application improves the artifact removal efficiency of cardiac magnetic resonance image.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to an intelligent method for removing artifacts in cardiac magnetic resonance imaging. Background Technology

[0002] In the field of magnetic resonance imaging, physiological motion artifacts occur because MR imaging takes a long time. During MR imaging, artifacts are caused by cardiac contraction, large blood vessel pulsation, respiratory movements, and blood flow. These artifacts are the most common cause of MR image quality degradation. The movement of the heart may cause blurring or distortion in the image.

[0003] Current methods reduce motion artifacts by tracking the patient's heart movement and adjusting the image data accordingly using motion correction techniques. However, these methods are preprocessing techniques that require additional sensors and data acquisition, resulting in high complexity and generally low processing accuracy. In contrast, non-blind deconvolution processing in artifact postprocessing methods is less costly, but its processing accuracy depends entirely on the accuracy of motion blur kernel estimation. Existing methods estimate the blur kernel by establishing a heart movement model based on multiple sensors or by calculating the difference from a standard image, which suffers from error accumulation. Summary of the Invention

[0004] To address the above problems, this invention provides an intelligent method for removing artifacts in cardiac magnetic resonance imaging, the method comprising:

[0005] Acquire the grayscale image of the heart magnetic resonance imaging (MRI) to be processed and the standard grayscale image of the heart MRI;

[0006] Based on the spectral energy distribution of the grayscale cardiac magnetic resonance image to be processed, the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed is obtained; based on the number of artifact layers and the difference in detail distribution between the grayscale cardiac magnetic resonance image to be processed and the standard grayscale cardiac magnetic resonance image, the artifact layer superposition loss curve of the grayscale cardiac magnetic resonance image to be processed is obtained.

[0007] Based on the artifact layer superposition loss curve and the overall structural information of the cardiac magnetic resonance grayscale image to be processed, the estimated blur kernel weight of the cardiac magnetic resonance grayscale image to be processed is obtained.

[0008] Based on the estimated fuzzy kernel weights, artifact removal is performed on the grayscale cardiac magnetic resonance image to be processed, resulting in the processed grayscale cardiac magnetic resonance image.

[0009] Preferably, the specific method for obtaining the number of artifact layers in the cardiac magnetic resonance grayscale image to be processed based on the spectral energy distribution of the image is as follows:

[0010] By performing Fourier transform on the grayscale image of the cardiac magnetic resonance imaging to be processed, the spectral center image, amplitude spectrum, and phase spectrum of the grayscale image of the cardiac magnetic resonance imaging to be processed are obtained.

[0011] Draw horizontal and vertical straight lines through the center point of the amplitude spectrum of the cardiac magnetic resonance grayscale image to be processed. Use these two straight lines as dividing lines to divide the amplitude spectrum into several regions, which are distributed in two rows and two columns. The region in the second row and first column is recorded as the lower left region of the amplitude spectrum. Obtain the values ​​of all amplitudes in the lower left region of the amplitude spectrum. Amplitudes with the same value are grouped into one category, and the information entropy of all amplitudes in the lower left region of the amplitude spectrum is obtained.

[0012] Obtain the estimated number of target artifact layers;

[0013] Based on the information entropy of all amplitudes in the lower left region of the amplitude spectrum, obtain the objective function output value for estimating the number of artifact layers for each target.

[0014] The number of target artifact layers with the smallest output value of the objective function is estimated as the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed.

[0015] Preferably, the method for obtaining the estimated number of several target artifact layers includes:

[0016] A starting quantity parameter b and a ending quantity parameter a are preset. The estimated number of artifact layers starts with b and increases by 1 step by step until the increased estimated number of artifact layers is a. The increment stops. The initial estimated number of artifact layers and the estimated number of artifact layers after each increase are recorded as the estimated number of target artifact layers. Several estimated numbers of target artifact layers are obtained.

[0017] Preferably, the specific formula for obtaining the objective function output value of the estimated number of target artifact layers based on the information entropy of all amplitudes in the lower left region of the amplitude spectrum is as follows:

[0018]

[0019] In the formula, E represents the objective function output value of the estimated number of target artifact layers; θ represents the information entropy of all amplitudes in the lower left region of the amplitude spectrum; N represents the estimated number of target artifact layers; ln() represents the logarithmic function with the natural constant as the base; || represents taking the absolute value.

[0020] Preferably, the specific method for obtaining the artifact layer superposition loss curve of the cardiac magnetic resonance grayscale image to be processed based on the number of artifact layers and the difference in detail distribution between the grayscale image to be processed and the standard cardiac magnetic resonance grayscale image is as follows:

[0021] A spectral-centered image of the standard cardiac magnetic resonance imaging (MRI) grayscale image is obtained by performing a Fourier transform on the standard cardiac MRI grayscale image. The result of the difference between the spectral-centered image of the standard cardiac MRI grayscale image and the spectral-centered image of the cardiac MRI grayscale image to be processed is denoted as the luminance residual spectrum image. For any radius in the luminance residual spectrum image, the sum of the luminance of all pixels on the radius is taken as the information loss of the radius. The luminance residual spectrum image contains multiple radii.

[0022] Based on the number of artifact layers and the amount of information loss of all radii in the brightness residual spectrum image, the artifact layer superposition loss curve of the grayscale cardiac magnetic resonance image to be processed is obtained.

[0023] Preferably, the specific method for obtaining the artifact layer superposition loss curve of the cardiac magnetic resonance grayscale image to be processed based on the number of artifact layers and the information loss of all radii in the brightness residual spectrum image is as follows:

[0024] The number of artifact layers in the grayscale cardiac MRI image to be processed is denoted as N1. N1 is used as the K value of the k-means clustering algorithm to cluster the information loss of all radii in the brightness residual spectrum image, obtaining several clusters. The mean of the information loss of all radii in each cluster is used as the average information loss of each cluster. All clusters are sorted in ascending order of average information loss, and the sorted cluster set is denoted as the cluster set of the grayscale cardiac MRI image to be processed. The cluster set of the grayscale cardiac MRI image to be processed is input into a two-dimensional coordinate system with the index of the cluster set as the x-axis and the average information loss of the cluster as the y-axis to obtain several first data points. The curve formed by fitting all the first data points by the least squares method is denoted as the artifact layer superposition loss curve of the grayscale cardiac MRI image to be processed.

[0025] Preferably, the specific method for obtaining the estimated blur kernel weights of the cardiac magnetic resonance grayscale image to be processed based on the artifact layer overlay loss curve and the overall structural information of the cardiac magnetic resonance grayscale image to be processed includes:

[0026] A starting decomposition quantity parameter b1 and a ending decomposition quantity parameter b2 are preset. The decomposition quantity starts with b1 and increases sequentially with a step size of 1 until the increased decomposition quantity is b2, at which point the increment stops. The starting decomposition quantity and the decomposition quantity after each increase are recorded as the target decomposition quantity, and several target decomposition quantities are obtained.

[0027] For any target decomposition quantity, the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is decomposed using the independent component analysis algorithm according to the target decomposition quantity to obtain several phase sub-spectrums under the target decomposition quantity; and the motion change curve of the artifact layer under the target decomposition quantity is obtained.

[0028] Based on the artifact layer overlay loss curve and the artifact layer motion change curve under the target decomposition number, the objective function value under each target decomposition number is obtained; the set of all phase sub-spectrums corresponding to the target decomposition number with the smallest objective function value is taken as the optimal set of phase sub-spectrums for the cardiac magnetic resonance grayscale image to be processed.

[0029] The Euclidean distance between the centroid of each phase subspectrum in the optimal phase subspectrum set and the centroid of the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is denoted as the second distance of each phase subspectrum in the optimal phase subspectrum set. The second distances of all phase subspectrums in the optimal phase subspectrum set are linearly normalized to obtain all normalized second distances. The standard deviation of all normalized second distances is used as the estimation fuzzy kernel weight of the cardiac magnetic resonance grayscale image to be processed.

[0030] Preferably, the specific method for obtaining the motion change curve of the artifact layer under the target decomposition quantity is as follows:

[0031] The Euclidean distance between the centroid of each phase sub-spectrum under the target decomposition quantity and the centroid of the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is taken as the first distance of each phase sub-spectrum under the target decomposition quantity. All phase sub-spectrums under the target decomposition quantity are sorted in ascending order of the first distance to obtain the sorted phase sub-spectrum set and recorded as the phase sub-spectrum set under the target decomposition quantity. Using the index in the phase sub-spectrum set as the abscissa and the first distance of the phase sub-spectrum as the ordinate, the phase sub-spectrum set under the target decomposition quantity is input into a two-dimensional coordinate system to obtain several second data points. All second data points are fitted using the least squares method to form the motion change curve of the artifact layer under the target decomposition quantity.

[0032] Preferably, the specific method for obtaining the objective function value for each target decomposition number based on the artifact layer overlay loss curve and the artifact layer motion change curve under the target decomposition number is as follows:

[0033] Obtain the brightness of all pixels in each phase subspectrum, and group pixels with the same brightness into the same category; the method for calculating the objective function value under any number of target decompositions is as follows:

[0034]

[0035] In the formula, W represents the objective function value under any target decomposition number; MSE represents the mean square error between the artifact layer superposition loss curve of the cardiac magnetic resonance grayscale image to be processed and the artifact layer motion change curve under the target decomposition number; N2 represents the total number of all phase sub-spectrums under the target decomposition number; μ i This represents the average brightness of all class pixels in the i-th phase subspectrum under the target decomposition number; σ represents the mean brightness of all pixel-like points in all phase sub-spectrums under the target decomposition quantity; i This represents the standard deviation of all class pixels in the i-th phase subspectrum under the target decomposition number; represents the standard deviation of all pixel-like points in all phase sub-spectrums under the target decomposition quantity; | represents taking the absolute value.

[0036] Preferably, the specific method for performing artifact removal on the cardiac magnetic resonance grayscale image to be processed based on the estimated fuzzy kernel weights to obtain the processed cardiac magnetic resonance grayscale image includes:

[0037] A preset size parameter L is used to set an L×L Gaussian kernel. The weight distribution of the Gaussian kernel follows the estimated blur kernel weight of the cardiac magnetic resonance grayscale image to be processed, and the Gaussian kernel is used as the blur kernel of the cardiac magnetic resonance grayscale image to be processed. The blur kernel is then used to perform non-blind deconvolution on the cardiac magnetic resonance grayscale image to be processed to obtain the processed cardiac magnetic resonance grayscale image.

[0038] The beneficial effects of the technical solution of the present invention are as follows: The present invention obtains the estimated blur kernel weight of the cardiac magnetic resonance grayscale image to be processed based on the artifact layer superposition loss curve and the overall structural information of the cardiac magnetic resonance grayscale image to be processed; since the blur kernel actually carries the motion information of all artifact layers in the cardiac magnetic resonance grayscale image to be processed, it can effectively eliminate physiological motion artifacts generated by heartbeat; by eliminating artifacts in the cardiac magnetic resonance grayscale image to be processed based on the estimated blur kernel weight, compared with the method of using other sensors to model the cardiac motion state to estimate the blur kernel or perform motion compensation, the present invention does not need to introduce redundant information sources, thereby avoiding the problem of error accumulation, thus improving the artifact removal efficiency of cardiac magnetic resonance images. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of the steps of an intelligent method for removing artifacts in cardiac magnetic resonance imaging according to the present invention;

[0041] Figure 2 This is a flowchart illustrating the feature relationships of an intelligent method for removing artifacts in cardiac magnetic resonance imaging according to the present invention. Detailed Implementation

[0042] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for intelligent removal of artifacts in cardiac magnetic resonance imaging based on the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0044] The following describes in detail, with reference to the accompanying drawings, a specific scheme for the intelligent removal of artifacts in cardiac magnetic resonance imaging provided by the present invention.

[0045] Please see Figure 1 The diagram illustrates a flowchart of a method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to an embodiment of the present invention. The method includes the following steps:

[0046] Step S001: Obtain the grayscale image of the heart magnetic resonance imaging (MRI) to be processed and the standard grayscale image of the heart magnetic resonance imaging (MRI).

[0047] Specifically, the first step is to acquire a grayscale image of the heart magnetic resonance imaging (MRI) to be processed. The specific process is as follows:

[0048] A cardiac magnetic resonance imaging (MRI) image with artifacts is acquired using an MIR device and designated as the cardiac MRI image to be processed. Median filtering and grayscale conversion are performed on the cardiac MRI image to be processed to obtain a grayscale cardiac MRI image. A cardiac MRI image without artifacts is acquired using an MIR device and designated as the standard cardiac MRI image. Median filtering and grayscale conversion are performed on the standard cardiac MRI image to obtain a standard cardiac MRI grayscale image.

[0049] Median filtering and grayscale conversion are existing technologies, and will not be described in detail here.

[0050] Thus, the grayscale image of the heart magnetic resonance imaging (MRI) to be processed and the standard grayscale image of the heart magnetic resonance imaging (MRI) are obtained through the above method.

[0051] Step S002: Obtain the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed; obtain the artifact layer superposition loss curve of the grayscale cardiac magnetic resonance image to be processed;

[0052] 1. Obtain the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed.

[0053] It should be noted that when motion blur occurs, its formation principle is essentially due to the mixing of images from different points in time within the exposure period. Therefore, the image presents a state of multiple layers superimposed, which corresponds to the superimposed state of multiple similar pixel distributions. In the amplitude spectrum, the amplitude of each pixel is necessarily larger than that of an image without motion blur, because the superposition of layers leads to an increase in the value of all pixels in the image. In the phase spectrum, this is characterized by an increase in phase types and a large skewness in the phase distribution. This is because, assuming that when layers are superimposed at different positions, the positional relationship between the information within the layers hardly changes, but rather that layers with two similar phase distributions are superimposed in an alternating manner. When multiple distributions are superimposed, a distribution skewness will occur.

[0054] Specifically, by performing a Fourier transform on the grayscale image of the cardiac magnetic resonance imaging to be processed, the spectral centering image, amplitude spectrum, and phase spectrum of the grayscale image of the cardiac magnetic resonance imaging to be processed are obtained; wherein, each pixel in the amplitude spectrum and phase spectrum corresponds to a frequency, and each pixel in the spectral centering image corresponds to a brightness.

[0055] The acquisition of the spectral center image, amplitude spectrum, and phase spectrum through Fourier transform is well-known, and will not be elaborated further in this embodiment.

[0056] It should be noted that the horizontal axis of the amplitude spectrum represents different frequencies in the horizontal direction of the grayscale cardiac magnetic resonance image to be processed, and the vertical axis represents different frequencies in the vertical direction of the grayscale cardiac magnetic resonance image to be processed. The left side of the amplitude spectrum represents low-frequency components, while the right side of the amplitude spectrum represents high-frequency components. The bottom of the amplitude spectrum represents low-frequency components, while the top of the amplitude spectrum represents high-frequency components. Furthermore, when motion blur occurs, the low-frequency components will increase.

[0057] Specifically, the amplitude spectrum of the cardiac magnetic resonance grayscale image to be processed is divided into several regions by drawing horizontal and vertical straight lines through the center point. These two lines are used as dividing lines. The regions are distributed in two rows and two columns. The region in the second row and first column is designated as the lower left region of the amplitude spectrum. The values ​​of all amplitudes in the lower left region of the amplitude spectrum are obtained. Amplitudes with the same value are grouped together, and the information entropy of all amplitudes in the lower left region of the amplitude spectrum is obtained.

[0058] Among them, obtaining information entropy is a well-known technology, and will not be elaborated on in this implementation.

[0059] It should be noted that, because the amplitude of motion varies at different locations in the grayscale cardiac magnetic resonance image being processed during heartbeat, the physiological motion artifacts in the grayscale cardiac magnetic resonance image being processed are asynchronous at different locations. The superposition of artifacts at each pixel is different. The same amplitude in the amplitude spectrum can only occur at pixels in the same region of the grayscale cardiac magnetic resonance image being processed, and where the artifact superposition is the same. That is, when two pixels in the grayscale cardiac magnetic resonance image being processed undergo the same motion and are located in the same connected region, these two pixels can have the same amplitude. When the motions of two pixels are different, i.e., the artifact superposition is different, the same amplitude will not occur.

[0060] A starting quantity parameter b and a ending quantity parameter a are preset. In this embodiment, a = 25 and b = 1 are used as examples. This embodiment does not impose specific limitations. The values ​​of a and b are determined according to the specific implementation.

[0061] Specifically, starting with b, the estimated number of artifact layers is incremented by 1. The initial estimated number of artifact layers and the estimated number of artifact layers after each increment are recorded as the estimated number of target artifact layers. The incrementing stops when the estimated number of target artifact layers is a, and several estimated numbers of target artifact layers are obtained.

[0062] As an example, the method for calculating the output value of the objective function to obtain the estimated number of artifact layers for each target is as follows:

[0063]

[0064] In the formula, E represents the objective function output value of the estimated number of target artifact layers; θ represents the information entropy of all amplitudes in the lower left region of the amplitude spectrum; N represents the estimated number of target artifact layers; ln() represents the logarithmic function with the natural constant as the base; || represents taking the absolute value.

[0065] It should be noted that the information entropy of all amplitudes in the lower left region of the amplitude spectrum represents the expected total information contained in all amplitudes in the lower left region. Assuming that the lower left region of the amplitude spectrum contains the vast majority of artifact information, and that the amplitudes appearing in this region are due to the superposition of N layers; lnN represents the entropy limit generated when all amplitudes in the lower left region of the amplitude spectrum are divided into N layers, i.e., the maximum expected information. Then, when there exists a value of N whose maximum expected information is closest to the total expected information of the lower left region, it means that this value of N is the minimum number of layers superimposed in the cardiac magnetic resonance grayscale image to be processed.

[0066] Specifically, obtain the output value of the objective function for the estimated number of all target artifact layers, and take the estimated number of target artifact layers with the smallest objective function output value as the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed.

[0067] At this point, the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed is obtained.

[0068] 2. Obtain the artifact layer overlay loss curve of the grayscale image of the heart magnetic resonance imaging to be processed.

[0069] It should be noted that the brightness of pixels in the spectral-centered image represents the amount of high-frequency and low-frequency information in the grayscale cardiac magnetic resonance image to be processed. The closer to the center of the spectral-centered image and the higher the brightness, the lower the frequency. The closer to the edge of the spectral-centered image and the darker the brightness, the higher the frequency. Pixels at different radii in the spectral-centered image represent different levels of high and low frequency information.

[0070] Specifically, a spectral-centered image of the standard cardiac magnetic resonance grayscale image is obtained by performing a Fourier transform on the standard cardiac magnetic resonance grayscale image; the result of the difference between the spectral-centered image of the standard cardiac magnetic resonance grayscale image and the spectral-centered image of the cardiac magnetic resonance grayscale image to be processed is denoted as the luminance residual spectrum image; for any radius in the luminance residual spectrum image, the sum of the luminance of all pixels on the radius is taken as the information loss of the radius; the luminance residual spectrum image contains multiple radii.

[0071] The spectral-centered image is composed of several concentric circles with different radii. During the phase difference process between the spectral-centered image of the standard cardiac magnetic resonance grayscale image and the spectral-centered image of the cardiac magnetic resonance grayscale image to be processed, if the brightness of a pixel is negative after the phase difference, the brightness of that pixel is set to 0.

[0072] Furthermore, the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed is denoted as N1. N1 is used as the K value of the k-means clustering algorithm to cluster the information loss of all radii in the brightness residual spectrum image, obtaining several clusters. The mean of the information loss of all radii in each cluster is used as the average information loss of each cluster. All clusters are sorted in ascending order of average information loss, and the sorted cluster set is denoted as the cluster set of the grayscale cardiac magnetic resonance image to be processed. The cluster set of the grayscale cardiac magnetic resonance image to be processed is input into a two-dimensional coordinate system with the index of the cluster set as the x-axis and the average information loss of the cluster as the y-axis to obtain several first data points. The curve formed by fitting all the first data points by the least squares method is denoted as the artifact layer superposition loss curve of the grayscale cardiac magnetic resonance image to be processed.

[0073] The clustering process based on the K value is a well-known process, and the least squares method is a well-known technique. Therefore, this embodiment will not elaborate further on it here.

[0074] Thus, the artifact layer stacking loss curve of the grayscale cardiac magnetic resonance image to be processed is obtained.

[0075] Step S003: Obtain the estimated fuzzy kernel weights of the grayscale cardiac magnetic resonance image to be processed.

[0076] Specifically, a starting decomposition quantity parameter b1 and a ending decomposition quantity parameter b2 are preset. In this embodiment, b1 = 5 and b2 = 20 are used as examples. This embodiment does not impose specific limitations, and b1 and b2 are determined according to the specific implementation.

[0077] Specifically, starting with b1 as the initial decomposition quantity, the decomposition quantity is incremented sequentially with a step size of 1. The initial decomposition quantity and the decomposition quantity after each increment are recorded as the target decomposition quantity, until the target decomposition quantity is b2, at which point the increment stops, and several target decomposition quantities are obtained.

[0078] For any target decomposition quantity, the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is decomposed using the independent component analysis algorithm according to the target decomposition quantity to obtain several phase sub-spectrums under the target decomposition quantity.

[0079] It should be noted that the phase spectrum cannot be directly decomposed. This step treats the phase spectrum as an image matrix, where the brightness at each position does not have a real numerical meaning. Matrix pixels with different brightness only represent different phases. The decomposition of the phase spectrum is regarded as the decomposition of all matrix pixels on the phase spectrum image matrix. Matrix pixels exist only at some positions on each phase sub-spectrum, and not at the remaining positions. Independent component analysis algorithm is existing technology, and will not be described in detail here.

[0080] Specifically, for any target decomposition quantity, the Euclidean distance between the centroid of each phase sub-spectrum under the target decomposition quantity and the centroid of the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is taken as the first distance of each phase sub-spectrum; all phase sub-spectrums are sorted in ascending order of the first distance to obtain the sorted phase sub-spectrum set and recorded as the phase sub-spectrum set under the target decomposition quantity; the phase sub-spectrum set under the target decomposition quantity is input into a two-dimensional coordinate system with the index in the phase sub-spectrum set as the horizontal axis and the first distance of the phase sub-spectrum as the vertical axis to obtain several second data points; and all second data points are fitted using the least squares method to form the motion change curve of the artifact layer under the target decomposition quantity.

[0081] Furthermore, the brightness of all pixels in each phase sub-spectrum is obtained, pixels with the same brightness are grouped into one class of pixels, and the standard deviation and mean of all classes of pixels in each phase sub-spectrum are obtained.

[0082] As an example, the method for calculating the objective function value under any objective decomposition quantity is as follows:

[0083]

[0084] In the formula, W represents the objective function value under any target decomposition number; MSE represents the mean square error between the artifact layer superposition loss curve of the cardiac magnetic resonance grayscale image to be processed and the artifact layer motion change curve under the target decomposition number; N2 represents the total number of all phase sub-spectrums under the target decomposition number; μ i This represents the average brightness of all class pixels in the i-th phase subspectrum under the target decomposition number; σ represents the mean brightness of all pixel-like points in all phase sub-spectrums under the target decomposition quantity; i This represents the standard deviation of all class pixels in the i-th phase subspectrum under the target decomposition number; represents the standard deviation of all pixel-like points in all phase sub-spectrums under the target decomposition quantity; | represents taking the absolute value.

[0085] It should be noted that since the superposition loss between artifact layers depends on the relative motion of each layer with respect to the original image, the greater the motion, the larger the superposition area and the greater the superposition loss. If we consider the two to be linearly related, then the two curves should have the same trend. Therefore, a larger MSE indicates that the phase sub-spectrums of the phase spectrum decomposition are closer to the phase information of the actual artifact layer. In the case of the cardiac magnetic resonance grayscale image being processed, each movement generates an overall positional shift between each artifact layer, but the phase relationship within each artifact layer is fixed. For example, when motion blur occurs in a human facial image, the distance between the eyes in any artifact layer is fixed. Therefore, the phase distribution information of all phase sub-spectrums is the same, and the brightness distribution information of all matrix pixels in all phase sub-spectrums should conform to similar or identical standard deviations and means. This represents the phase distribution dispersion of the i-th phase sub-spectrum under the target decomposition number; This represents the phase distribution dispersion of all phase sub-spectrums under the target decomposition quantity. The smaller this value, the more the phase distribution of all phase sub-spectrums conforms to the same standard deviation and mean, and the closer the decomposed phase sub-spectrums are to the phase information of the actual artifact layer.

[0086] Furthermore, the objective function value is obtained for all objective decomposition quantities, and the set of all phase sub-spectrums corresponding to the objective decomposition quantity with the smallest objective function value is taken as the optimal set of phase sub-spectrums for the cardiac magnetic resonance grayscale image to be processed.

[0087] The Euclidean distance between the centroid of each phase subspectrum in the optimal phase subspectrum set and the centroid of the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is denoted as the second distance of each phase subspectrum. The second distances of all phase subspectrums in the optimal phase subspectrum set are linearly normalized to obtain all normalized second distances. The standard deviation of all normalized second distances is used as the estimation fuzzy kernel weight of the cardiac magnetic resonance grayscale image to be processed.

[0088] Thus, the estimated fuzzy kernel weights of the cardiac magnetic resonance grayscale image to be processed are obtained through the above method.

[0089] Step S004: Perform artifact removal operation on the grayscale cardiac magnetic resonance image to be processed according to the estimated fuzzy kernel weights to obtain the processed grayscale cardiac magnetic resonance image.

[0090] It should be noted that the blur kernel actually carries the motion information of all artifact layers in the target image, thus effectively eliminating physiological motion artifacts caused by heartbeats. Compared with using other sensors to model the heart's motion state to estimate the blur kernel or perform motion compensation, it does not require the introduction of extra information sources, thereby avoiding the problem of error accumulation and improving the processing efficiency of cardiac magnetic resonance images.

[0091] A size parameter L is preset. In this embodiment, L=7 is used as an example. This embodiment does not impose specific limitations. L is determined according to the specific implementation.

[0092] Specifically, an L×L Gaussian kernel is set, and the weight distribution of the Gaussian kernel follows the estimated blur kernel weights of the cardiac magnetic resonance grayscale image to be processed. The Gaussian kernel is used as the blur kernel of the cardiac magnetic resonance grayscale image to be processed. The blur kernel is then used to perform non-blind deconvolution on the cardiac magnetic resonance grayscale image to be processed to obtain the processed cardiac magnetic resonance grayscale image.

[0093] This concludes the embodiment; please refer to [link / reference]. Figure 2 It shows a flowchart of the feature relationships of an intelligent method for removing artifacts in cardiac magnetic resonance imaging.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent removal of artifacts in cardiac magnetic resonance imaging, characterized in that, The method includes the following steps: Acquire the grayscale image of the heart magnetic resonance imaging (MRI) to be processed and the standard grayscale image of the heart MRI; Based on the spectral energy distribution of the grayscale cardiac magnetic resonance image to be processed, the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed is obtained; based on the number of artifact layers and the difference in detail distribution between the grayscale cardiac magnetic resonance image to be processed and the standard grayscale cardiac magnetic resonance image, the artifact layer superposition loss curve of the grayscale cardiac magnetic resonance image to be processed is obtained. Based on the artifact layer superposition loss curve and the overall structural information of the cardiac magnetic resonance grayscale image to be processed, the estimated blur kernel weight of the cardiac magnetic resonance grayscale image to be processed is obtained. Based on the estimated fuzzy kernel weights, artifact removal is performed on the grayscale cardiac magnetic resonance image to be processed, resulting in the processed grayscale cardiac magnetic resonance image.

2. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 1, characterized in that, The method for obtaining the number of artifact layers in the grayscale cardiac magnetic resonance image based on the spectral energy distribution of the image to be processed includes the following specific methods: By performing Fourier transform on the grayscale image of the cardiac magnetic resonance imaging to be processed, the spectral center image, amplitude spectrum, and phase spectrum of the grayscale image of the cardiac magnetic resonance imaging to be processed are obtained. Draw horizontal and vertical straight lines through the center point of the amplitude spectrum of the cardiac magnetic resonance grayscale image to be processed. Use these two straight lines as dividing lines to divide the amplitude spectrum into several regions, which are distributed in two rows and two columns. The region in the second row and first column is recorded as the lower left region of the amplitude spectrum. Obtain the values ​​of all amplitudes in the lower left region of the amplitude spectrum. Amplitudes with the same value are grouped into one category, and the information entropy of all amplitudes in the lower left region of the amplitude spectrum is obtained. Obtain the estimated number of target artifact layers; Based on the information entropy of all amplitudes in the lower left region of the amplitude spectrum, obtain the objective function output value for estimating the number of artifact layers for each target. The number of target artifact layers with the smallest output value of the objective function is estimated as the number of artifact layers in the grayscale cardiac magnetic resonance image to be processed.

3. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 2, characterized in that, The specific method for obtaining the estimated number of several target artifact layers is as follows: A starting quantity parameter b and a ending quantity parameter a are preset. The estimated number of artifact layers starts with b and increases by 1 step by step until the increased estimated number of artifact layers is a. The increment stops. The initial estimated number of artifact layers and the estimated number of artifact layers after each increase are recorded as the estimated number of target artifact layers. Several estimated numbers of target artifact layers are obtained.

4. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 2, characterized in that, The specific formula for obtaining the objective function output value of the estimated number of artifact layers for each target based on the information entropy of all amplitudes in the lower left region of the amplitude spectrum is as follows: In the formula, E represents the objective function output value for estimating the number of artifact layers for any target; θ represents the information entropy of all amplitudes in the lower left region of the amplitude spectrum; N represents the estimated number of target artifact layers; ln() represents the logarithmic function with the natural constant as the base; || represents taking the absolute value.

5. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 2, characterized in that, The method for obtaining the artifact layer superposition loss curve of the cardiac magnetic resonance grayscale image to be processed based on the number of artifact layers and the difference in detail distribution between the grayscale image to be processed and the standard cardiac magnetic resonance grayscale image includes the following specific methods: By performing Fourier transform on a standard cardiac magnetic resonance grayscale image, a spectral-centered image of the standard cardiac magnetic resonance grayscale image is obtained; the result of the phase difference between the spectral-centered image of the standard cardiac magnetic resonance grayscale image and the spectral-centered image of the cardiac magnetic resonance grayscale image to be processed is denoted as the brightness residual spectrum image. For any radius in the luminance residual spectrum image, the sum of the luminance of all pixels on the radius is taken as the information loss of the radius; the luminance residual spectrum image contains multiple radii; Based on the number of artifact layers and the amount of information loss of all radii in the brightness residual spectrum image, the artifact layer superposition loss curve of the grayscale cardiac magnetic resonance image to be processed is obtained.

6. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 5, characterized in that, The method for obtaining the artifact layer superposition loss curve of the cardiac magnetic resonance grayscale image to be processed based on the number of artifact layers and the information loss of all radii in the brightness residual spectrum image includes the following specific methods: The number of artifact layers in the grayscale cardiac MRI image to be processed is denoted as N1. N1 is used as the K value of the k-means clustering algorithm to cluster the information loss of all radii in the brightness residual spectrum image, obtaining several clusters. The mean of the information loss of all radii in each cluster is used as the average information loss of each cluster. All clusters are sorted in ascending order of average information loss, and the sorted cluster set is denoted as the cluster set of the grayscale cardiac MRI image to be processed. The cluster set of the grayscale cardiac MRI image to be processed is input into a two-dimensional coordinate system with the index of the cluster set as the x-axis and the average information loss of the cluster as the y-axis to obtain several first data points. The curve formed by fitting all the first data points by the least squares method is denoted as the artifact layer superposition loss curve of the grayscale cardiac MRI image to be processed.

7. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 2, characterized in that, The specific method for obtaining the estimated blur kernel weights of the cardiac magnetic resonance grayscale image to be processed based on the artifact layer superposition loss curve and the overall structural information of the image is as follows: A starting decomposition quantity parameter b1 and a ending decomposition quantity parameter b2 are preset. The decomposition quantity starts with b1 and increases sequentially with a step size of 1 until the increased decomposition quantity is b2, at which point the increment stops. The starting decomposition quantity and the decomposition quantity after each increase are recorded as the target decomposition quantity, and several target decomposition quantities are obtained. For any target decomposition quantity, the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is decomposed using the independent component analysis algorithm according to the target decomposition quantity to obtain several phase sub-spectrums under the target decomposition quantity; and the motion change curve of the artifact layer under the target decomposition quantity is obtained. Based on the artifact layer overlay loss curve and the artifact layer motion change curve under the target decomposition number, the objective function value under each target decomposition number is obtained; the set of all phase sub-spectrums corresponding to the target decomposition number with the smallest objective function value is taken as the optimal set of phase sub-spectrums for the cardiac magnetic resonance grayscale image to be processed. The Euclidean distance between the centroid of each phase subspectrum in the optimal phase subspectrum set and the centroid of the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is denoted as the second distance of each phase subspectrum in the optimal phase subspectrum set. The second distances of all phase subspectrums in the optimal phase subspectrum set are linearly normalized to obtain all normalized second distances. The standard deviation of all normalized second distances is used as the estimation fuzzy kernel weight of the cardiac magnetic resonance grayscale image to be processed.

8. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 7, characterized in that, The specific method for obtaining the motion change curve of the artifact layer under the target decomposition quantity is as follows: The Euclidean distance between the centroid of each phase sub-spectrum under the target decomposition quantity and the centroid of the phase spectrum of the cardiac magnetic resonance grayscale image to be processed is taken as the first distance of each phase sub-spectrum under the target decomposition quantity. All phase sub-spectrums under the target decomposition quantity are sorted in ascending order of the first distance to obtain the sorted phase sub-spectrum set and recorded as the phase sub-spectrum set under the target decomposition quantity. Using the index in the phase sub-spectrum set as the abscissa and the first distance of the phase sub-spectrum as the ordinate, the phase sub-spectrum set under the target decomposition quantity is input into a two-dimensional coordinate system to obtain several second data points. All second data points are fitted using the least squares method to form the motion change curve of the artifact layer under the target decomposition quantity.

9. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 7, characterized in that, The specific method for obtaining the objective function value for each target decomposition number based on the artifact layer superposition loss curve and the artifact layer motion change curve under the target decomposition number is as follows: Obtain the brightness of all pixels in each phase subspectrum, and group pixels with the same brightness into the same category; the method for calculating the objective function value under any number of target decompositions is as follows: In the formula, W represents the objective function value under any target decomposition number; MSE represents the mean square error between the artifact layer superposition loss curve of the cardiac magnetic resonance grayscale image to be processed and the artifact layer motion change curve under the target decomposition number; N2 represents the total number of all phase sub-spectrums under the target decomposition number; μ i This represents the average brightness of all class pixels in the i-th phase subspectrum under the target decomposition number; σ represents the mean brightness of all pixel-like points in all phase sub-spectrums under the target decomposition quantity; i This represents the standard deviation of all class pixels in the i-th phase subspectrum under the target decomposition number; The standard deviation of all pixel-like points in all phase sub-spectrums under the target decomposition quantity is represented by ||; || represents taking the absolute value.

10. The method for intelligent removal of artifacts in cardiac magnetic resonance imaging according to claim 1, characterized in that, The specific method for performing artifact removal on the grayscale cardiac magnetic resonance image to be processed based on the estimated fuzzy kernel weights to obtain the processed grayscale cardiac magnetic resonance image includes: A preset size parameter L is used to set an L×L Gaussian kernel. The weight distribution of the Gaussian kernel follows the estimated blur kernel weight of the cardiac magnetic resonance grayscale image to be processed, and the Gaussian kernel is used as the blur kernel of the cardiac magnetic resonance grayscale image to be processed. The blur kernel is then used to perform non-blind deconvolution on the cardiac magnetic resonance grayscale image to be processed to obtain the processed cardiac magnetic resonance grayscale image.

Images