Magnetic resonance imaging method and device, computer equipment and storage medium

Abstract

The invention relates to a magnetic resonance imaging method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring under-sampling K space data of an imaging object; determining an initial reconstruction image based on the under-sampled K space data; the initial reconstruction image is input into a trained correction model, an intermediate reconstruction image is obtained based on the output of the correction model, and the output of the correction model is the deviation between the initial reconstruction image and a full-sampling reconstruction image; and performing distortion correction on the intermediate reconstructed image based on the under-sampled K space data to obtain a target reconstructed image. According to the invention, the quality of the target reconstructed image is effectively improved while the image reconstruction speed is improved.

Description

Technical Field

[0001] The present application relates to the field of medical imaging technology, and in particular to a magnetic resonance imaging method, apparatus, computer equipment and storage medium. Background Art

[0002] Magnetic Resonance Imaging (MRI), as a medical imaging technology, can draw accurate three-dimensional images of different tissues, and is of great value in the diagnosis of diseases in various systems throughout the body, especially the diagnosis of early tumors. However, in clinical applications, MRI acquisition time is very long and the imaging speed is slow, which will affect the inspection efficiency and the patient's experience during the inspection. In addition, the long imaging wait increases the possibility of patient movement, which can easily lead to artifacts in magnetic resonance imaging and affect the image reconstruction effect. Therefore, while improving the reconstruction speed of magnetic resonance images, ensuring the quality of magnetic resonance images has become a major research issue in the industry.

[0003] At present, undersampling parallel imaging technology based on multi-channel coils is one of the main acceleration methods. However, the mainstream traditional analytical algorithms and their analytical extensions have limited ability to solve high-magnification undersampling (five times and above) parallel imaging. The fundamental reason is that the high-magnification undersampling signal is too sparse compared to the full-sampling signal, resulting in a very low signal-to-noise ratio in the parallel imaging results, which is insufficient to meet clinical needs.

[0004] Therefore, there is an urgent need in the related art for a method to improve the image reconstruction speed while ensuring the quality of magnetic resonance images. Summary of the invention

[0005] Based on this, it is necessary to provide a magnetic resonance imaging method, apparatus, computer equipment and storage medium that can improve the image reconstruction speed while ensuring the quality of the magnetic resonance image in order to solve the above technical problems.

[0006] In a first aspect, the present application provides a magnetic resonance imaging method. The method comprises:

[0007] Acquire undersampled K-space data of the imaging object;

[0008] determining an initial reconstructed image based on the undersampled K-space data;

[0009] Inputting the initial reconstructed image into a trained correction model, and obtaining an intermediate reconstructed image based on the output of the correction model, wherein the output of the correction model is the deviation between the initial reconstructed image and the fully sampled reconstructed image;

[0010] The intermediate reconstructed image is subjected to distortion correction based on the under-sampled K-space data to obtain a target reconstructed image.

[0011] In one embodiment, determining the initial reconstructed image based on the undersampled K-space data comprises:

[0012] Determining a first regularization term based on coil sensitivity and phase corresponding to the undersampled K-space data;

[0013] Based on the first regularization term, the under-sampled K-space data are reconstructed in parallel to determine an initial reconstructed image, and the first regularization term is used to achieve consistency between the under-sampled K-space data and the K-space data corresponding to the initial reconstructed image.

[0014] In one embodiment, performing distortion correction on the intermediate reconstructed image based on the undersampled K-space data to obtain a target reconstructed image includes:

[0015] determining a virtual coil sensitivity matrix based on coil sensitivities corresponding to the undersampled K-space data;

[0016] The intermediate reconstructed image is corrected based on the virtual coil sensitivity matrix to obtain a target reconstructed image.

[0017] In one embodiment, the performing distortion correction on the intermediate reconstructed image based on the undersampled K-space data to obtain a target reconstructed image further comprises:

[0018] determining a second regularization term based on coil sensitivity corresponding to the undersampled K-space data;

[0019] The intermediate reconstructed image is corrected based on the second regularization term to obtain a target reconstructed image.

[0020] In one of the embodiments, the second regularization term is a total variation regularization term.

[0021] In one embodiment, the correction model is a dual-channel network model, and the dual-channel network model is obtained by separately training the real part and the imaginary part of the under-sampled sample image in the training set.

[0022] In one of the embodiments, after obtaining the intermediate reconstructed image based on the output of the correction model, the method further includes:

[0023] The phase of the intermediate reconstructed image is iteratively corrected.

[0024] In a second aspect, the present application also provides a magnetic resonance imaging device. The device comprises:

[0025] A data acquisition module, used for acquiring undersampled K-space data of an imaging object;

[0026] An initial reconstruction module, used for determining an initial reconstructed image based on the undersampled K-space data;

[0027] A first correction module, used for inputting the initial reconstructed image into a trained correction model, and obtaining an intermediate reconstructed image based on an output of the correction model, wherein the output of the correction model is a deviation between the initial reconstructed image and a fully sampled reconstructed image;

[0028] The second correction module is used to perform distortion correction on the intermediate reconstructed image based on the under-sampled K-space data to obtain a target reconstructed image.

[0029] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the magnetic resonance imaging methods in the first aspect are implemented.

[0030] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of any one of the magnetic resonance imaging methods in the first aspect are implemented.

[0031] The magnetic resonance imaging method, device, computer equipment and storage medium described above acquire undersampled K-space data of the imaging object; determine the initial reconstructed image based on the undersampled K-space data; input the initial reconstructed image into a trained correction model, and obtain an intermediate reconstructed image based on the output of the correction model, wherein the input of the correction model is the undersampled reconstructed image, and the output of the correction model is the deviation between the undersampled reconstructed image and the fully sampled reconstructed image; the intermediate reconstructed image is corrected based on the coil sensitivity corresponding to the undersampled K-space data to obtain a target reconstructed image. The magnetic resonance imaging method provided by the present application, on the one hand, achieves high acceleration and greatly shortens the image scanning time by acquiring the undersampled K-space data of the imaging object; on the other hand, determines the initial reconstructed image based on the undersampled K-space data, and then obtains the intermediate reconstructed image according to the correction model, which further reduces the aliasing of the image while improving the image signal-to-noise ratio, and also makes the image smoother as a whole. Then, based on the undersampled K-space data, the intermediate reconstructed image is distorted and corrected to obtain the target reconstructed image, and the intermediate reconstructed image is constrained by the original real data, which further improves the authenticity of the target reconstructed image and makes the detailed features of the target reconstructed image closer to the real collected data. Therefore, the present application improves the image reconstruction speed while also effectively improving the quality of the target reconstructed image.

[0032] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0034] Figure 1 A diagram showing an application environment of a magnetic resonance imaging method in an embodiment;

[0035] Figure 2 is a schematic flow chart of a magnetic resonance imaging method in one embodiment;

[0036] Figure 3 It is a schematic diagram of under-sampled K-space data in a specific embodiment;

[0037] Figure 4 is a schematic diagram of a SENSE algorithm in one embodiment;

[0038] Figure 5 is a schematic diagram of a calibration model in one embodiment;

[0039] Figure 6 A schematic diagram of a Hank matrix in one embodiment;

[0040] Figure 7 A schematic diagram of image reconstruction test results in one embodiment;

[0041] Figure 8 A schematic diagram of image reconstruction test results in another embodiment;

[0042] Fig. 9 is a structural block diagram of a magnetic resonance imaging device in one embodiment;

[0043] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the", "these" and the like in this application do not represent quantitative restrictions, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether directly or indirectly. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. Usually, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0046] The terms "module", "unit", etc. used below are a combination of software and / or hardware that can implement a predetermined function. Although the devices described in the following embodiments are preferably implemented in hardware, the implementation of software, or a combination of software and hardware is also possible and conceivable.

[0047] The magnetic resonance imaging method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The terminal 102 obtains the undersampled K space data of the imaging object and sends it to the server 102. The server 102 determines the initial reconstructed image based on the undersampled K space data; inputs the initial reconstructed image into the trained correction model, and obtains the intermediate reconstructed image based on the output of the correction model, the input of the correction model is the undersampled reconstructed image, and the output of the correction model is the deviation between the undersampled reconstructed image and the fully sampled reconstructed image; the intermediate reconstructed image is corrected based on the coil sensitivity corresponding to the undersampled K space data, and the target reconstructed image is obtained and sent to the terminal 102. In other embodiments, the steps performed by the server 102 can also be performed by the terminal 104, and this application does not make specific restrictions on this. Among them, the terminal 102 may include an image acquisition device, such as a magnetic resonance imaging device. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0048] In one embodiment, Figure 2 As shown, a magnetic resonance imaging method is provided, which is applied to Figure 1 The application environment in the example is used to illustrate the following steps:

[0049] S201: Acquire undersampled K-space data of an imaging object.

[0050] In the embodiment of the present application, K space refers to the dual space of the ordinary space under Fourier transformation. Acquiring undersampled K space data may include obtaining sampled data in a sampling method lower than the Nyquist rate in K space based on a preset downsampling ratio, wherein the preset downsampling ratio may be two times or more. The sampling method lower than the Nyquist rate may include acquiring multiple groups of sampling signals based on a preset number of radio frequency pulse excitation (shot), the preset number is not less than 2, and the imaging sequence corresponding to the sampling may include a diffusion weighted imaging (DWI) sequence, a susceptibility weighted imaging (SWI) sequence, a T1 weighted imaging sequence or a T2 weighted imaging sequence. In this embodiment, the imaging sequence selects a DWI sequence, and since multiple sampling images are acquired based on a preset number of radio frequency pulse excitation (shot), the phases of different sampling images are different. Each radio frequency pulse excitation is uniformly downsampled in K space based on a preset downsampling ratio.

[0051] In a specific embodiment, if the number of RF pulse excitations (shots) is 4, the undersampled K-space data is as follows: Figure 3 As shown. Due to the influence of the diffusion gradient, there is a phase difference or phase encoding line staggered between the sampling images corresponding to the four shots, so it is difficult to directly splice the sampling images corresponding to the four shots for image reconstruction. However, considering that the sampling images are real sampling data, in the embodiment of the present application, when reconstructing the sampling image of a certain phase, the reconstruction quality of the phase image can be improved by extracting the features of the sampling images of other phases.

[0052] S203: Determine an initial reconstructed image based on the under-sampled K-space data.

[0053] For each phase shot, it is essentially a uniformly downsampled K space, and the solution process of this K space is essentially a parallel imaging solution process. The embodiment of the present application is based on the SENSE algorithm and uses the information redundancy brought by the multi-channel acquisition of magnetic resonance to solve the image. The solution process of the SENSE algorithm is as follows: Figure 4 shown. Figure 4 As shown in the figure, there are four coils S1, S2, S3, and S4. Each coil corresponds to a sampled image with aliasing in the image domain (Aliased images in the figure, i.e., aliasing artifact images). The composition of the four aliased sampled images forms the original image (Original image) corresponding to the undersampled K-space data. These sampled images reflect that each point in the final image domain is given a corresponding weight under the "viewing angle" of each coil. The set of results after adding different weights to each point in the image domain corresponds to the image domain of different coil scanning results. The set of weights for each point in each coil is the coil sensitivity matrix, which is S in the figure. 1A -S 4B The formed matrix.

[0054] In order to improve the solution effect of the SENSE algorithm in the high-power downsampling scenario and thus improve the imaging quality, the embodiment of the present application introduces a SENSE algorithm with a regularization term added to solve the undersampled K space. Based on this, determining the initial reconstructed image based on the undersampled K space data may include reconstructing the undersampled K space data based on the SENSE algorithm with a regularization term added to determine the initial reconstructed image.

[0055] In some specific embodiments, the SENSE algorithm with the regularization term added can be expressed as formula (1):

[0056]

[0057] In formula (1), Ns refers to the number of shots, specifically, it means that each shot is first split and solved separately, and then the results of the solution are merged in the image domain. dt refers to the undersampled K space corresponding to the shot, which is usually solved by expanding the matrix into a one-dimensional vector. F t is the undersampled discrete Fourier transform (DFT) corresponding to the t-th phase RF pulse excitation, C is the coil sensitivity matrix, x t is the two-dimensional complex image (RO_PE) corresponding to each shot to be solved, d t is the actual undersampled K space corresponding to the shot. regularization term is the added regularization term, which can include TV regularization term (Total Variation Regularization), bi-norm square weighted regularization term, etc., or a combination of multiple regularization terms. The regularization term includes the image domain x to be solved. When equation (1) converges to the minimum, x is the image domain to be solved, and then the initial reconstructed image can be obtained based on the image domain x.

[0058] S205: Inputting the initial reconstructed image into a trained correction model, and obtaining an intermediate reconstructed image based on the output of the correction model, wherein the output of the correction model is the deviation between the initial reconstructed image and the fully sampled reconstructed image.

[0059] In an embodiment of the present application, the correction model may include a residual network, and the residual network includes a deep learning network of residual mapping, and the network structure of the correction model includes but is not limited to a Unet network structure. During the training of the correction model, the input samples of the training set are two-dimensional images reconstructed based on undersampled K space, and the gold standard of the validation set is the residual of the two-dimensional image reconstructed based on undersampled K space and the two-dimensional image reconstructed based on full sampled K space. By comparing the difference between the residual output of the correction model during the training process and the residual of the gold standard, the parameters of the correction model are iteratively adjusted until the difference meets the preset requirements. In the embodiment of the present application, the structure, number of channels, and loss function of the correction model can be set according to actual needs, and this application does not impose any restrictions on this.

[0060] In a specific embodiment, the calibration model is as follows Figure 5 As shown, the network structure of the correction model is as follows Figure 5 The network structure is shown in . Figure 5The image corresponding to Under sampled represents the training sample of the two-dimensional image reconstructed based on the undersampled K-space input of the correction model, the image corresponding to True Error of Under sampled represents the gold standard residual image of the validation set of the correction network, and the image corresponding to Learned Error of Under sampled represents the output residual of the correction network.

[0061] After the correction model training is completed, the initial reconstructed image is input into the correction model. It can be understood that the initial reconstructed image is an undersampled reconstructed image, and the output of the correction model is the deviation between the initial reconstructed image and the fully sampled reconstructed image. Based on the output of the correction model, the intermediate reconstructed image can be obtained according to formula (2).

[0062] ρ net =ρ+f(ρ) (2)

[0063] In formula (2), ρ net is an intermediate reconstructed image with a high signal-to-noise ratio, ρ is an initial reconstructed image with a low signal-to-noise ratio, and f(ρ) is the deviation between the initial reconstructed image output by the correction model and the fully sampled reconstructed image. The present application corrects the initial reconstructed image by using the residual output of the correction model, so that the undersampled initial reconstructed image approaches the fully sampled image, which greatly improves the signal-to-noise ratio of the image, maintains a relatively clear contrast in the case of high b-value dispersion, and can further reduce the aliasing of the image and improve the smoothness of the image.

[0064] S207: Perform distortion correction on the intermediate reconstructed image based on the under-sampled K-space data to obtain a target reconstructed image.

[0065] In an embodiment of the present application, the intermediate reconstructed image is obtained based on a SENSE algorithm and a correction model with a regularization term added. Solving the optimal solution may cause the position of the image domain corresponding to the intermediate reconstructed image in the K space to deviate from the position of the image domain corresponding to the real collected data in the K space. Based on this, in order to make the detailed features of the target reconstructed image closer to the real collected data, the intermediate reconstructed image can be subjected to distortion correction. In some embodiments, the distortion correction of the intermediate reconstructed image based on the undersampled K space data may include determining distortion correction data based on the undersampled K space data, and performing distortion correction on the intermediate reconstructed image based on the distortion correction data. In some embodiments, the distortion correction data may include the coil sensitivity corresponding to the undersampled K space data, and may also include a virtual coil sensitivity matrix determined based on the coil sensitivity. Specifically, the distortion correction may refer to the SENSE algorithm with a regularization term added in the above embodiment, introduce distortion correction data based on the original limited matrix of the SENSE algorithm, and set the regularization term. In other embodiments, the distortion correction may also include backfilling the center of the undersampled K space. Since the undersampled K-space center can reflect the image contour and overall contrast, distortion correction can be performed by backfilling the undersampled K-space center. In other embodiments, the distortion correction can also include distortion correction based on the MUSE (GE) principle.

[0066] The magnetic resonance imaging method provided by the present application, on the one hand, achieves high acceleration and greatly shortens the image scanning time by acquiring the undersampled K-space data of the imaging object; on the other hand, the initial reconstructed image is determined based on the undersampled K-space data, and then the intermediate reconstructed image is obtained according to the correction model, which further reduces the aliasing of the image while improving the image signal-to-noise ratio, and also makes the image smoother as a whole. Then, the intermediate reconstructed image is subjected to distortion correction based on the undersampled K-space data to obtain the target reconstructed image, and the intermediate reconstructed image is constrained by the original real data, which further improves the authenticity of the target reconstructed image and makes the detailed features of the target reconstructed image closer to the real collected data. Therefore, while improving the image reconstruction speed, the present application also effectively improves the quality of the target reconstructed image.

[0067] In an embodiment of the present application, determining the initial reconstructed image based on the undersampled K-space data includes:

[0068] S301: Determine a first regularization term based on coil sensitivity and phase corresponding to the under-sampled K-space data.

[0069] S303: Based on the first regularization term, the under-sampled K-space data is reconstructed in parallel to determine an initial reconstructed image, wherein the first regularization term is used to achieve consistency between the under-sampled K-space data and the K-space data corresponding to the initial reconstructed image.

[0070] In the embodiment of the present application, according to the relationship between the under-sampled K-space data acquired by different RF pulse excitations (shots), formula (3) can be determined:

[0071] m I (x)=p(x)Φ I (x)

[0072] m i (x)=p(x)Φ i (x) (3)

[0073] In formula (3), / represents the under-sampled spatial data obtained by RF pulse excitation shot I, i represents the under-sampled spatial data obtained by RF pulse excitation shot i, Φ I (x), Φ i (x) is the phase corresponding to the undersampled spatial data obtained for shotI and shot i, respectively, m I (x), m i (x) The coil sensitivity corresponding to the undersampled spatial data acquired for shotI and shot i, respectively.

[0074] Based on formula (3), formula (4) can be determined:

[0075]

[0076] Transforming equation (4) to the frequency domain can determine equation (5):

[0077]

[0078] In formula (5), x represents the time domain signal, and in formula (4), k represents the frequency domain signal. After the frequency conversion, x is represented as k, and the two represent the same physical information. Formula (5) also reflects the annihilation relationship in the frequency domain. The convolution of the annihilation filter shown in formula (5) can be transformed by introducing the block-Hankel matrices to obtain the first regularization term. The first regularization term is shown in formula (6):

[0079] H(m I )Φ i -H(m i )Φ I =0 (6)

[0080] Formula (6) represents the first regularization term in a single scenario. Extending Formula (6) to cover all RF pulse excitation shots yields Formula (7):

[0081]

[0082] The Hank matrix shown in formula (7) is constructed as follows Figure 6 As shown, for Figure 6 The Hank matrix shown in FIG. 7 shows a matrix whose rank is defined as the number of non-zero singular values ​​in the matrix, which reflects the richness of information contained in the matrix. Usually, a low-rank matrix contains less information, and the lower the rank, the less information there is. Therefore, adding the rank of the matrix to the objective function can minimize the objective function while minimizing the rank of the matrix, which helps to prevent overfitting. Therefore, the undersampled K-space data can be reconstructed in parallel based on the rank of the Hank matrix as shown in FIG. (7).

[0083] According to formula (8), based on the first regularization term, the undersampled K-space data is reconstructed in parallel to determine an initial reconstructed image:

[0084]

[0085] The parameters of the first term in equation (8) can refer to the description in equation (1). x is a three-dimensional matrix (RO_PE_Shot). H(x) corresponds to the H function in equation (7). Both are matrices H(x) constructed together in the actual scenario based on the actual data of multiple excitations. ‖H(x)‖ * represents the singular values ​​and constraints of the Hank matrix, which is the low rank perior of the undersampled K-space data of multi-shot excitation. λ is a preset coefficient, which is used as a weight coefficient to adjust the weight of the constraint term. The regularization term helps to prevent overfitting, and the weight coefficient of the regularization term reflects the degree of prevention of overfitting. The larger the λ, the higher the signal-to-noise ratio of the initial reconstructed image and the smoother it is. However, excessive smoothness may cause the image to lose details, so an appropriate λ should be selected for constraint. When equation (8) converges to the minimum, x is the desired image domain, and then the initial reconstructed image can be obtained based on the image domain x.

[0086] In an embodiment of the present application, a first regularization term is determined based on the coil sensitivity and phase corresponding to the under-sampled K-space data, and based on the first regularization term, the under-sampled K-space data is reconstructed in parallel to determine an initial reconstructed image, which can further reduce the signal-to-noise ratio of the initial reconstructed image and improve image quality.

[0087] In order to further improve image quality, reduce image aliasing artifacts, and improve image realism, in an embodiment of the present application, the distortion correction of the intermediate reconstructed image based on the undersampled K-space data to obtain the target reconstructed image includes:

[0088] S401: Determine a virtual coil sensitivity matrix based on the coil sensitivity corresponding to the under-sampled K-space data.

[0089] S403: Correcting the intermediate reconstructed image based on the virtual coil sensitivity matrix to obtain a target reconstructed image.

[0090] In the embodiment of the present application, the coil sensitivity matrix can be determined based on the coil sensitivity corresponding to the undersampled K-space data, and then the virtual coil sensitivity matrix can be determined based on the sensitivity matrix. Based on the virtual coil sensitivity matrix, the intermediate reconstructed image can be corrected according to formula (9):

[0091]

[0092] In formula (9), φ t represents the phase corresponding to the undersampled spatial data obtained by shot t, F t is the undersampled discrete Fourier transform (DFT) corresponding to the t-th phase RF pulse excitation, C is the coil sensitivity matrix (coil sensitivity map or coil sensitivity operator), is the virtual coil sensitivity matrix, and Corresponding to and d t R(m) is the regularization term. β and λ in formula (8) have similar meanings and are both weight coefficients, but the constraints they correspond to are different. Therefore, the actual effects of adjusting these two parameters are also significantly different.

[0093] In some embodiments, performing distortion correction on the intermediate reconstructed image based on the undersampled K-space data to obtain a target reconstructed image further comprises:

[0094] S501: Determine a second regularization term based on the coil sensitivity corresponding to the under-sampled K-space data.

[0095] S503: Correcting the intermediate reconstructed image based on the second regularization term to obtain a target reconstructed image.

[0096] In the embodiment of the present application, R(m) in formula (9) is the second regularization term, and the second regularization term can be a total variation regularization term (Total Variation Regularization, TV regularization). The TV regularization term uses the total variation of the signal or image, that is, the sum of the absolute values ​​of the differences between adjacent areas of the signal or image for regularization. Specifically, the total variation of the image refers to the sum of the absolute values ​​of the grayscale differences between adjacent pixels of the image. The TV regularization term can be determined according to formula (10):

[0097] TV(I)=∑i,j|I(i+1,j)-I(i,j)|+|I(i,j+1)-I(i,j)| (10)

[0098] In formula (10), I represents the intermediate reconstructed image, and i and j represent pixel indices. The constraint objective of the TV regularization term is to minimize the total variation of the image, that is, to make the image smoother while keeping the image edge unchanged. Based on the principle of TV regularization, the grayscale of the noise point is different from that of the surrounding points, so the smaller the total variation, the smaller the grayscale difference between the noise point and the surrounding points, while the structural edge is almost unaffected due to the obvious grayscale change.

[0099] In the embodiment of the present application, by introducing the virtual coil sensitivity matrix and the second regularization term, the intermediate reconstructed image is corrected, so that the position of the image domain of the target reconstructed image in the K space can be closer to the position of the image domain of the actual data in the K space, and the texture details of the target reconstructed image can be closer to the actual image, and the human tissue structure details in the image can be restored to improve the authenticity of the target reconstructed image. In addition, the edge information of the image can be retained while suppressing noise, a relatively sparse solution can be obtained, and the uniqueness problem of degenerate or underdetermined linear systems can be solved. Furthermore, the phase of the intermediate reconstructed image can be corrected, effectively reducing aliasing artifacts and improving the quality of magnetic resonance images.

[0100] The following is a specific example to illustrate the test results of the reconstructed image in this application. Figure 7 As shown in the figure, the image corresponding to Undersampled is the initial reconstructed image determined based on the undersampled K-space data. Figure 7 The middle left 2) is the intermediate reconstructed image obtained based on the output of the correction model. It can be seen that the smoothness of the intermediate reconstructed image is improved and the clarity is significantly improved. Furthermore, the image corresponding to Result from Mix (i.e. Figure 7(3) in the middle left is the target reconstructed image after the introduction of the virtual coil sensitivity matrix and the second regularization term. It can be seen that the texture details in the image are restored and the image realism is improved. The image corresponding to Full sampled is the fully sampled reconstructed image, which is also used to calculate the image residual (gold standard) during the training process of the correction model.

[0101] Figure 8 The images corresponding to Under sampled, Result after UNET, Result of Mix, and Full sampled are shown in Figure 1. Figure 7 The meaning is the same. By comparing the intermediate reconstructed image corresponding to Result after UNET (i.e. Figure 8 2) and the target reconstructed image after introducing the virtual coil sensitivity matrix and the second regularization term correction corresponding to Result of Mix (i.e. Figure 8 In the middle left 2), it can be found that the phase of the intermediate reconstructed image is corrected, and the aliasing artifact disappears on the image. By comparing the two image domains of left 3 and left 4, it can also be found that the upper limit of the solution method provided by this application is higher than the traditional network solution, because the fully sampled reconstructed image corresponding to left 4 is used in this application to calculate the gold standard residual of the correction model.

[0102] In an embodiment of the present application, the correction model can be further optimized. In some embodiments, the correction model is a dual-channel network model, and the dual-channel network model is obtained by training the real part and the imaginary part of the under-sampled sample image in the training set respectively. After the intermediate reconstructed image is obtained based on the output of the correction model, it also includes: iteratively correcting the phase of the intermediate reconstructed image.

[0103] In the embodiment of the present application, φ in formula (9) can be t Further optimization can be performed. The correction model can be set as a dual-channel network model, and the real and imaginary parts of the sample images in the training set can be trained separately. Based on the test results of the dual-channel residual network, additional iterative correction can be performed on the phase between different RF pulse excitations (shots) according to formula (11).

[0104]

[0105] In formula (11), x net is the modified dual-channel residual network test result, ‖Wφ e ‖1 is a one-norm constraint term based on wavelet transform.

[0106] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0107] Based on the same inventive concept, the embodiment of the present application also provides a magnetic resonance imaging device for implementing the magnetic resonance imaging method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more magnetic resonance imaging device embodiments provided below can refer to the limitations on the magnetic resonance imaging method above, and will not be repeated here.

[0108] In one embodiment, Fig. 9 As shown, a magnetic resonance imaging device 900 is provided, comprising:

[0109] A data acquisition module 901 is used to acquire undersampled K-space data of an imaging object;

[0110] An initial reconstruction module 902, configured to determine an initial reconstructed image based on the undersampled K-space data;

[0111] A first correction module 903, configured to input the initial reconstructed image into a trained correction model, and obtain an intermediate reconstructed image based on an output of the correction model, wherein the output of the correction model is a deviation between the initial reconstructed image and a fully sampled reconstructed image;

[0112] The second correction module 904 is configured to perform distortion correction on the intermediate reconstructed image based on the under-sampled K-space data to obtain a target reconstructed image.

[0113] In one embodiment, the initial reconstruction module 902 is also used to determine a first regularization term based on the coil sensitivity and phase corresponding to the undersampled K-space data; based on the first regularization term, the undersampled K-space data is reconstructed in parallel to determine an initial reconstructed image, and the first regularization term is used to achieve consistency between the undersampled K-space data and the K-space data corresponding to the initial reconstructed image.

[0114] In one embodiment, the second correction module 904 is further used to determine a virtual coil sensitivity matrix based on the coil sensitivity corresponding to the undersampled K-space data; and to correct the intermediate reconstructed image based on the virtual coil sensitivity matrix to obtain a target reconstructed image.

[0115] In one embodiment, the second correction module 904 is further used to determine a second regularization term based on the coil sensitivity corresponding to the undersampled K-space data; and to correct the intermediate reconstructed image based on the second regularization term to obtain a target reconstructed image.

[0116] In one embodiment, the second regularization term is a total variation regularization term.

[0117] In one embodiment, the correction model is a dual-channel network model, and the dual-channel network model is obtained by respectively training the real part and the imaginary part of the under-sampled sample image in the training set.

[0118] In one embodiment, the first correction module 903 is further configured to iteratively correct the phase of the intermediate reconstructed image.

[0119] Each module in the magnetic resonance imaging device 900 can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0120] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a magnetic resonance imaging method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0121] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0122] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the magnetic resonance imaging method in any of the above embodiments are implemented.

[0123] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the magnetic resonance imaging method in any of the above embodiments are implemented.

[0124] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the magnetic resonance imaging method in any of the above embodiments.

[0125] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0126] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0127] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0128] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A magnetic resonance imaging method, characterized in that: The method comprises: Acquire undersampled K-space data of the imaging object; determining an initial reconstructed image based on the undersampled K-space data; Inputting the initial reconstructed image into a trained correction model, and obtaining an intermediate reconstructed image based on the output of the correction model, wherein the output of the correction model is an image residual between the initial reconstructed image and a fully sampled reconstructed image; The intermediate reconstructed image is subjected to distortion correction based on the under-sampled K-space data to obtain a target reconstructed image.

2. The method according to claim 1, characterized in that Determining an initial reconstructed image based on the undersampled K-space data includes: Determining a first regularization term based on coil sensitivity and phase corresponding to the undersampled K-space data; Based on the first regularization term, the under-sampled K-space data are reconstructed in parallel to determine an initial reconstructed image, and the first regularization term is used to achieve consistency between the under-sampled K-space data and the K-space data corresponding to the initial reconstructed image.

3. The method according to claim 1, characterized in that The performing distortion correction on the intermediate reconstructed image based on the under-sampled K-space data to obtain a target reconstructed image comprises: determining a virtual coil sensitivity matrix based on coil sensitivities corresponding to the undersampled K-space data; The intermediate reconstructed image is corrected based on the virtual coil sensitivity matrix to obtain a target reconstructed image.

4. The method according to claim 1 or 3, characterized in that: The step of performing distortion correction on the intermediate reconstructed image based on the under-sampled K-space data to obtain a target reconstructed image further comprises: determining a second regularization term based on coil sensitivity corresponding to the undersampled K-space data; The intermediate reconstructed image is corrected based on the second regularization term to obtain a target reconstructed image.

5. The method according to claim 4, characterized in that The second regularization term is a total variation regularization term.

6. The method according to claim 1, characterized in that The correction model is a dual-channel network model, which is obtained by respectively training the real part and the imaginary part of the under-sampled sample image in the training set.

7. The method according to claim 6, characterized in that After obtaining the intermediate reconstructed image based on the output of the correction model, the method further includes: The phase of the intermediate reconstructed image is iteratively corrected.

8. A magnetic resonance imaging device, characterized in that: The device comprises: A data acquisition module, used for acquiring undersampled K-space data of an imaging object; An initial reconstruction module, used for determining an initial reconstructed image based on the undersampled K-space data; A first correction module, used for inputting the initial reconstructed image into a trained correction model, and obtaining an intermediate reconstructed image based on an output of the correction model, wherein the output of the correction model is a deviation between the initial reconstructed image and a fully sampled reconstructed image; The second correction module is used to perform distortion correction on the intermediate reconstructed image based on the under-sampled K-space data to obtain a target reconstructed image.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images