A DWI-EPI image Nyquist ghost correction method

Abstract

This invention relates to a method for Nyquist ghosting correction in DWI-EPI images, comprising the following steps: acquiring raw magnetic resonance imaging (MRI) data; segmenting the raw MRI data; extracting reference scan data for reference scan correction; filling the remaining data into corresponding k-space locations to form k-space data in the (Kx,Ky) domain; transforming both the k-space data in the (Kx,Ky) domain and the reference scan data into (x,Ky) mixed-domain data; performing reference scan correction on the (x,Ky) mixed-domain data to obtain first corrected data; transforming the first corrected data in the (x,Ky) mixed-domain into (Kx,Ky) domain data; performing referenceless scan correction on the (Kx,Ky) domain data to obtain second corrected data; performing zero-padding on the second corrected data to obtain third corrected data; and performing a two-dimensional inverse Fourier transform on the third corrected data to obtain an artifact-free image. Compared with existing technologies, this invention has advantages such as good correction effect and short correction time.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a method for correcting Nyquist ghosting in DWI-EPI images. Background Technology

[0002] Globally, nuclear magnetic resonance (NMR) technology has developed rapidly and has been widely applied in many fields. For example, low-field NMR is increasingly used in food science, agriculture, petroleum energy, materials science, and textile chemicals. On the other hand, high-field NMR is generally used for human examinations. After years of technological development and practical application research, NMR examinations are now used for almost the entire human body. To meet more examination needs, functional magnetic resonance imaging (fMRI) has been developed based on NMR.

[0003] High-field magnetic resonance (>=1.5T) diffusion-weighted imaging (DWI) is a type of functional magnetic resonance imaging (fMRI). It is a component of modern, state-of-the-art MRI and is indispensable in neuroimaging and oncology. DWI is a rapidly developing field with increasingly diverse applications. Currently, it is used for acute ischemic attacks, brain tumors, head and neck malignancies, thoracic malignancies, breast cancer, and abdominal malignancies.

[0004] EPI (Echo Planar Imaging) sequences are indispensable in diffusion-weighted imaging. Due to the extremely fast acquisition speed of a single-shot EPI sequence (capable of acquiring an image in milliseconds, from the time required for reconstruction to the speed required for reconstruction),... Spatial data), clinical DWI almost entirely uses it as the acquisition method. Although EPI has extremely fast imaging speed, it is highly susceptible to various artifacts caused by magnetic field inhomogeneities, mainly including three types: Nyquist ghosting, chemical shift artifacts, and pattern distortion artifacts. Among these, Nyquist ghosting is the most common, because artifacts such as gradient eddies, Field inhomogeneity, receiver chain and gradient amplifier group delay, adjoint field, amplitude modulation and Issues such as spatial data displacement can cause Nyquist ghosting, which affects imaging accuracy. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a DWI-EPI image Nyquist ghosting correction method with good correction effect and short correction time, especially for images with complex phase information that are difficult to correct using traditional methods.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for Nyquist ghosting correction in DWI-EPI images includes the following steps:

[0008] Acquire raw magnetic resonance data, segment the raw magnetic resonance data, extract reference scan data for reference scan correction, and fill the remaining data into the corresponding k-space positions to form k-space data in the (Kx,Ky) domain.

[0009] Both the k-space data and the reference scan data in the (Kx,Ky) domain are transformed into (x,Ky) mixed domain data;

[0010] Perform reference scan correction on the (x,Ky) mixed domain data to obtain the first corrected data;

[0011] Transform the first corrected data in the (x,Ky) mixed domain into (Kx,Ky) domain data;

[0012] The (Kx,Ky) domain data is corrected using a no-reference scan correction method to obtain the second corrected data;

[0013] The second correction data is zero-padding to obtain the third correction data, and the third correction data is subjected to a two-dimensional inverse Fourier transform to obtain an artifact-free image.

[0014] Furthermore, the segmentation of the raw magnetic resonance data specifically involves:

[0015] Based on the sequence gradient information, acquisition frequency, and system delay information, noise points and signal data acquired during gradient ramps are discarded. Only data points in the steady gradient segment during cyclic acquisition are taken, and data without phase encoding are removed, so that the data matrix size is consistent with the matrix size set during scanning.

[0016] Furthermore, the reference scan data is the data with Ky=0 in the scanned data matrix.

[0017] Furthermore, when filling the corresponding k-space positions, the negative gradient data in the k-space data are reversed left and right so that the degree directions of the k-space data are consistent.

[0018] Furthermore, the k-space data and the reference scan data are respectively subjected to inverse Fourier transform along the Kx direction to transform each signal with different phase encoding to the (x,Ky) mixed domain.

[0019] Furthermore, the reference scan correction specifically includes:

[0020] The reference scan data of the (x,Ky) mixed domain contains three rows of data. The interpolation row is obtained by averaging the corresponding positions of the first and third rows. The interpolation row is multiplied by the conjugate of the second row of the reference scan data of the (x,Ky) mixed domain to obtain a complex number of rows of data.

[0021] Phase difference information row data is obtained by fitting the complex row data using a first-order linear fitting method.

[0022] The phase difference information row data is removed from the positive values ​​in the image data in the (x,Ky) mixed domain to obtain the first correction data.

[0023] Furthermore, the first-order straight-line fitting method is the least squares method.

[0024] Furthermore, a Fourier transform is applied to the x-direction of the first corrected data in the (x,Ky) mixed domain to obtain (Kx,Ky) domain data.

[0025] Furthermore, the referenceless scan correction method is an improved information entropy correction method, and the correction process includes:

[0026] The parameter range of the constant parameter b and the first-order linear parameter k of the information entropy correction method is generated. Multiple parameter combinations are randomly generated within the parameter range, and each parameter combination corresponds to a first-order linear phase error.

[0027] The (Kx,Ky) domain data is transformed to the (x,Ky) mixed domain, and the phase error removal operation is performed on the mixed domain data using each of the first-order linear phase errors to obtain multiple corrected data.

[0028] The corrected data is transformed into the (x,y) image domain, the entropy value of the corresponding image is calculated, and the parameter combination with the smallest entropy value is obtained as the initial parameter.

[0029] Based on the initial parameters, the optimal parameter pair is searched and obtained. The information entropy is then corrected using the optimal parameter pair to obtain the second corrected data.

[0030] Furthermore, the optimal parameter pair is obtained by searching using the derivative-free method.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. This invention combines the traditional reference scan correction method and the information entropy correction method to perform two corrections on the Nyquist in the EPI image. This avoids the correction defects of the traditional information entropy correction method (half of the scan field along the phase coding direction of the corrected image) and can obtain an image with good correction effect.

[0033] 2. This invention segments the raw magnetic resonance data and combines reference scan correction and information entropy correction. The correction time is short, the robustness is good, and it can achieve stable correction in areas with severe eddy currents. It can handle EPI images under various conditions.

[0034] 3. In the information entropy correction of this invention, the parameter with the smallest entropy value is selected from the parameter grid as the initial parameter for iteration, which can quickly find the best solution and prevent random getting stuck in local optima, thereby further improving the correction effect and correction efficiency. Attached Figure Description

[0035] Figure 1 This is a flowchart of the method of the present invention;

[0036] Figure 2 This is a schematic diagram illustrating the phase error variation of the positive and negative polarity signals;

[0037] Figure 3 This is a schematic diagram of the information entropy method correction process;

[0038] Figure 4 A schematic diagram of GSR calculation;

[0039] Figure 5 GSR line graphs of different layers of the human brain;

[0040] Figure 6 Line graphs of GSR for different layers of water model. Detailed Implementation

[0041] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0042] like Figure 1 As shown, this invention provides a method for Nyquist ghosting correction in DWI-EPI images, comprising the following steps:

[0043] 1) Acquire raw magnetic resonance data, segment the raw magnetic resonance data, extract reference scan data for reference scan correction, and fill the remaining data into the corresponding k-space positions to form k-space data in the (Kx,Ky) domain;

[0044] 2) Transform both the k-space data and the reference scan data in the (Kx,Ky) domain into (x,Ky) mixed domain data;

[0045] 3) Perform reference scan correction on the (x,Ky) mixed domain data to correct the phase error between odd and even echoes, and obtain the first corrected data;

[0046] 4) Transform the first corrected data in the (x,Ky) mixed domain into (Kx,Ky) domain data;

[0047] 5) Perform a referenceless scan correction on the (Kx,Ky) domain data to correct the phase error between odd and even echoes, and obtain the second corrected data;

[0048] 6) Zero-padding is applied to the second correction data to obtain the third correction data, and a two-dimensional inverse Fourier transform is performed on the third correction data to obtain an artifact-free image.

[0049] The segmentation of the raw magnetic resonance data specifically involves: discarding noise points and signal data acquired during read gradient ramps based on sequence gradient information, acquisition frequency, and system delay information; retaining only data points within the stable read gradient segment during cyclic acquisition; and removing data without phase encoding, ensuring the data matrix size matches the matrix size set during scanning. The reference scan data refers to the data with Ky=0 in the scan-acquired data matrix.

[0050] The main technical features involved in step 1) above include:

[0051] (11) Segmentation matrix: Same as the matrix parameters set during scanning.

[0052] (12) Segmentation details: Based on the sequence gradient information, acquisition frequency and system delay information, discard the noise points and signal data acquired at the reading gradient ramp point, and only take the data points in the reading gradient stable segment during the cyclic acquisition.

[0053] (13) Copying the reference scan data for backup: The data matrix collected in the loop has two more rows of phase-uncoded data (Ky=0) than the data matrix used for reconstruction. Therefore, there are three rows of data with Ky=0 in the matrix. These three rows of data are copied as reference scan data for backup.

[0054] (14) Removal of redundant data: Remove the two redundant rows of phase-free encoded data from the data matrix obtained by cyclic acquisition, so that the size of the data matrix is ​​consistent with the size of the matrix set during scanning.

[0055] (15) k-space filling: k-space filling is the process of acquiring magnetic resonance signal data using different gradient modes and placing it in k-space. The position of the magnetic resonance signal data in k-space is determined by the gradient temporal structure. In conventional two-dimensional magnetic resonance imaging, the phase encoding gradient changes stepwise, and with each change in phase encoding, the magnetic resonance signal fills the k-space along the frequency encoding direction, forming a k-track in k-space. The length of each k-track depends on the frequency encoding gradient. In this embodiment, when filling the corresponding k-space position, the negative gradient data in the k-space data is reversed left and right to ensure that the k-space data in all directions are consistent.

[0056] In the above method, the conversion between the (x,Ky) mixed domain and the (Kx,Ky) domain is achieved through Fourier transform. Specifically, in step 2), the k-space data and the reference scan data are respectively subjected to inverse Fourier transform along the Kx direction to transform each signal with different phase encoding to the (x,Ky) mixed domain; in step 4), the first correction data in the (x,Ky) mixed domain is subjected to Fourier transform in the x direction to obtain the (Kx,Ky) domain data.

[0057] For reference scan data without phase encoding (Ky=0), theoretically, reference scan data in the (x,Ky) mixed domain would have the same phase change if there were no phase error. However, due to the presence of phase error, reference scan data in opposite directions should have phase error changes in opposite directions. Figure 2 This can illustrate the phase error change of the positive and negative angle signals. The dashed line and the dotted line represent the change of the phase error of the positive and negative angle signals with x, respectively.

[0058] k-space data containing phase error is expressed as follows:

[0059] ,

[0060] in The phase error indicates that there is a difference between the phase errors of odd and even rows. Therefore, there is a phase error between odd and even rows, which causes Nyquist ghosting. The solution is to eliminate the phase error between odd and even rows. This can eliminate ghosting, and the reference scan can be used to estimate the image. .

[0061] Based on the above principles, step 3) of reference scan correction specifically includes the following steps:

[0062] (31) The reference scan data for the (x,Ky) mixed domain comprises three rows of data, with the first, second, and third rows labeled as 1+, 2-, and 3+, respectively.

[0063] The interpolation row (2+) is obtained by averaging the corresponding positions of the first row (1+) and the third row (3+). The interpolation row (2+) is multiplied by the conjugate of the second row (2-) of the reference scan data in the (x,Ky) mixing domain to obtain complex row data that includes both the echo phase difference and the signal mode.

[0064] (32) Phase difference information row data are obtained by fitting the complex row data using a first-order linear fitting method, which is the least squares method;

[0065] (33) Remove the phase difference information row data from the positive values ​​in the image data in the (x,Ky) mixed domain to obtain the first correction data.

[0066] The referenceless scanning correction method used in this approach is an improved information entropy correction method. The information entropy E of a source is considered from the perspective of the statistical characteristics of the entire source. It characterizes the overall characteristics of the source in an average sense. For a specific source, there is only one information entropy. Different sources have different entropies due to their different statistical characteristics. The greater the uncertainty of the variable, the greater the entropy. Magnetic resonance imaging is a form of information, and since it is a grayscale image, it means that it carries one-dimensional information entropy. The calculation of image information entropy E is expressed by the following formula:

[0067] ,

[0068] The formula for the reconstructed image with ghosting is:

[0069]

[0070] As can be seen from the formula ,when (Phase error between odd and even rows) is When the image reaches its maximum entropy value, for At that time, the image has the minimum entropy value.

[0071] In this improved information entropy correction method, in order to find the optimal solution as soon as possible and to prevent random getting stuck in local optima, the parameter with the smallest entropy value is selected from the parameter grid as the initial parameter for iteration.

[0072] Based on the above principles, such as Figure 3 As shown, the improved information entropy correction process includes:

[0073] (51) Input the k-space data after reference scan correction;

[0074] (52) Generate the range of two parameters, namely constant parameter b and first-order linear parameter k, to form a parameter grid, and randomly generate multiple parameter combinations within the parameter range, each parameter combination corresponding to a first-order linear phase error;

[0075] In this embodiment, the ranges of the two parameters are set to [-π / 3, -π / 3] and [-0.1, 0.1], respectively. Both ranges are quantized into 50 parameters, for a total of 2500 parameter combinations. These 2500 parameter combinations are used to generate 2500 first-order linear phase errors.

[0076] (53) Transform the data obtained in step 4) to the (x,Ky) mixed domain, and then use the 2500 first-order linear errors in step (52) to perform the same phase error removal operation as in step (33) on the (x,Ky) mixed domain data to obtain 2500 corrected data. Transform the obtained 2500 corrected data to the (x,y) image domain and calculate the entropy value of each image to obtain the entropy matrix.

[0077] (54) Find the pair of parameters with the smallest entropy value as the initial parameters for iteration, and search for the parameter pair that minimizes the entropy value of the corrected image. This search is implemented using the derivative-free method. In specific implementations, the fminsearch function built into MATLAB can be used.

[0078] If the above methods are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0079] In another embodiment, an electronic device may be provided, including one or more processors, a memory, and one or more programs stored in the memory, the one or more programs including instructions for performing the DWI-EPI image Nyquist ghosting correction method as described above.

[0080] The correction effect of the DWI-EPI image Nyquist ghosting correction method of the present invention can be further illustrated by the following experiments.

[0081] 1. Simulation Experiment:

[0082] The experiments were conducted using DWI-EPI imaging on a water phantom and in the human brain, respectively, obtaining 8-channel data through a 1.5T magnetic resonance imaging (MRI) scanner. This is because when the b-value (diffusion sensitivity factor, affecting the diffusion sensitivity of diffusion-weighted imaging) is too high (e.g., 800 or 1000),... When the image has a relatively low amplitude, noise will overwhelm most of the residual artifacts, resulting in artifacts that can be completely corrected by various correction methods. Therefore, experimental testing was conducted at b=0. Water model and b=300 The procedure was performed on a human brain using DWI. Finally, a pre-written program was used for processing and correction.

[0083] 2. Evaluation Indicators and Results of Water Pyramid Imaging and Human Brain Imaging Correction Experiments

[0084] GSR is used as an evaluation index for correction effectiveness, such as Figure 4 As shown, the formula for calculating GSR is as follows:

[0085]

[0086] The correction and evaluation results of the water model images are shown in Table 1 and Figure 5 The correction results for the human brain images are given in Table 2 and Figure 6 Note: ① When using the information entropy correction method alone, the correction result is often that the original image is shifted by N / 2, resulting in a GSR greater than 1. This makes it inconvenient to compare other GSRs, so the entropy method was not included in the comparison; ② When the GSR is negative, it means that the ghosting designed by the algorithm is almost 0.

[0087] Table 1

[0088]

[0089] Table 2

[0090]

[0091] 3. Simulation Experiment Analysis

[0092] Through Table 1, Table 2, Figure 5 and Figure 6 We can draw the following conclusions: This invention combines traditional reference scan correction with a modified entropy correction method, demonstrating good correction results in both water phantom and human brain DWI-EPI imaging, and its accuracy is superior to both traditional reference scan and entropy. Furthermore, this invention features essentially identical correction parameters across different repetitions, effectively optimizing the code, significantly reducing computational overhead, and shortening correction time.

[0093] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for Nyquist ghosting correction in DWI-EPI images, characterized in that, Includes the following steps: Acquire raw magnetic resonance data, segment the raw magnetic resonance data, extract reference scan data for reference scan correction, and fill the remaining data into the corresponding k-space positions to form k-space data in the (Kx,Ky) domain. Both the k-space data and the reference scan data in the (Kx,Ky) domain are transformed into (x,Ky) mixed domain data; Perform reference scan correction on the (x,Ky) mixed domain data to obtain the first corrected data; Transform the first corrected data in the (x,Ky) mixed domain into (Kx,Ky) domain data; The (Kx,Ky) domain data is corrected using a no-reference scan correction method to obtain the second corrected data; The second correction data is zero-padding to obtain the third correction data, and the third correction data is subjected to a two-dimensional inverse Fourier transform to obtain an artifact-free image. The specific steps for segmenting the raw magnetic resonance data are as follows: Based on the sequence gradient information, acquisition frequency, and system delay information, noise points and signal data acquired during gradient ramps are discarded. Only data points in the steady gradient segment during cyclic acquisition are taken, and data without phase encoding are removed, so that the data matrix size is consistent with the matrix size set during scanning.

2. The DWI-EPI image Nyquist ghosting correction method according to claim 1, characterized in that, The reference scan data is the data with Ky=0 in the scanned data matrix.

3. The DWI-EPI image Nyquist ghosting correction method according to claim 1, characterized in that, When filling the corresponding k-space positions, the negative gradient data in the k-space data are reversed left and right so that the degree directions of the k-space data are consistent.

4. The DWI-EPI image Nyquist ghosting correction method according to claim 1, characterized in that, The k-space data and the reference scan data are respectively subjected to inverse Fourier transform along the Kx direction to transform each signal with different phase encoding to the (x,Ky) mixed domain.

5. The DWI-EPI image Nyquist ghosting correction method according to claim 1, characterized in that, The reference scan correction specifically refers to: The reference scan data of the (x,Ky) mixed domain contains three rows of data. The interpolation row is obtained by averaging the corresponding positions of the first and third rows. The interpolation row is multiplied by the conjugate of the second row of the reference scan data of the (x,Ky) mixed domain to obtain a complex number of rows of data. Phase difference information row data is obtained by fitting the complex row data using a first-order linear fitting method. The phase difference information row data is removed from the positive values ​​in the image data in the (x,Ky) mixed domain to obtain the first correction data.

6. The DWI-EPI image Nyquist ghosting correction method according to claim 5, characterized in that, The first-order linear fitting method is the least squares method.

7. The DWI-EPI image Nyquist ghosting correction method according to claim 1, characterized in that, Apply a Fourier transform to the x-direction of the first corrected data in the (x,Ky) mixed domain to obtain (Kx,Ky) domain data.

8. The DWI-EPI image Nyquist ghosting correction method according to claim 1, characterized in that, The referenceless scanning correction method is an improved information entropy correction method, and the correction process includes: The parameter range of the constant parameter b and the first-order linear parameter k of the information entropy correction method is generated. Multiple parameter combinations are randomly generated within the parameter range, and each parameter combination corresponds to a first-order linear phase error. The (Kx,Ky) domain data is transformed to the (x,Ky) mixed domain, and the phase error removal operation is performed on the mixed domain data using each of the first-order linear phase errors to obtain multiple corrected data. The corrected data is transformed into the (x,y) image domain, the entropy value of the corresponding image is calculated, and the parameter combination with the smallest entropy value is obtained as the initial parameter. Based on the initial parameters, the optimal parameter pair is searched and obtained. The information entropy is then corrected using the optimal parameter pair to obtain the second corrected data.

9. The DWI-EPI image Nyquist ghosting correction method according to claim 8, characterized in that, The optimal parameter pair is obtained by searching using the derivative-free method.

Images