Image reconstruction model generation and image reconstruction method, device, equipment and medium

Abstract

The embodiment of the present invention discloses an image reconstruction model generation and image reconstruction method, device, equipment and medium, wherein the image reconstruction model generation method includes: obtaining full-sampled magnetic resonance K-space data, and obtaining under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model; inputting the under-sampled K-space data into, optimizing and solving the inverse problem of the preset low-resolution model to obtain a sub-problem, establishing a super-resolution reconstruction generation network model, and then training the super-resolution reconstruction generation network model based on the discriminant network output result according to the under-sampled reconstructed image generated by the model and the full-sampled reconstructed image corresponding to the full-sampled magnetic resonance K-space data, and generating a target image reconstruction model. The technical solution of this embodiment realizes that while increasing the sampling acceleration multiple, a high-resolution image can be reconstructed to improve image quality.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of medical image processing, and in particular to an image reconstruction model generation and image reconstruction method, device, equipment and medium. Background Art

[0002] Magnetic Resonance Imaging (MRI) is an advanced imaging technology that is widely used in clinical diagnosis and medical research. However, due to the limitations of magnetic resonance physics and hardware, human activity, and tolerance, MRI often has limited spatial resolution and cannot accurately assess some cardiovascular and cerebrovascular diseases, such as thickening of the walls of some small cerebral blood vessels and plaques. Therefore, it is particularly important to improve the spatial resolution of MRI.

[0003] High-resolution magnetic resonance images correspond to high-density K-space sampling matrices, but it takes a lot of time to collect a large number of K-space points. Existing methods for obtaining high-resolution MRI images are mainly divided into two schemes. The first is to start from the perspective of compressed sensing (CS) reconstruction, using prior information such as the sparsity of MRI images to highly undersample the K-space to save scanning time, and then use the CS reconstruction algorithm to reconstruct high-resolution images. Another scheme is based on image super-resolution technology in computer vision, first collecting and reconstructing low-resolution MRI images, and then using image super-resolution algorithms to generate high-resolution images. This type of method has two steps. The first step is to collect low-resolution k-space information and use the reconstruction algorithm to obtain low-resolution images. The second step is to input the low-resolution MRI image into the image super-resolution algorithm model to obtain a high-resolution MRI image.

[0004] However, in existing methods, when the acceleration factor of the algorithm based on compressed sensing reconstruction is high, the original sampled K-space points are too sparse, and the sampled points are scattered in the high and low frequency parts, and the signal-to-noise ratio is relatively low, making it difficult to reconstruct a high-definition high-resolution image. In the image super-resolution technology based on computer vision, reconstruction and super-resolution are two separate steps without interaction. The original collected K-space information does not directly participate in the super-resolution step. The first step of the reconstruction algorithm cannot restore the information of the low-resolution image 100%, and there will be certain errors. This error will be accumulated and amplified in the second step of super-resolution, affecting the quality of the high-resolution image. Summary of the invention

[0005] The embodiments of the present invention provide an image reconstruction model generation and image reconstruction method, device, equipment and medium, so as to achieve the reconstruction of high-resolution images and improve image quality while increasing the sampling acceleration multiple.

[0006] In a first aspect, an embodiment of the present invention provides a method for generating an image reconstruction model, the method comprising:

[0007] Acquire full-sampled magnetic resonance K-space data, and obtain under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model;

[0008] The under-sampled K-space data is inputted into, a sub-problem is obtained by optimizing and solving the inverse problem of the preset low-resolution model, and a super-resolution reconstruction generation network model is established to obtain an under-sampled reconstructed image;

[0009] The under-sampled reconstructed image and the fully-sampled reconstructed image corresponding to the fully-sampled magnetic resonance K-space data are simultaneously input into a discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to an output result of the discriminant network model until a loss function of the discriminant network model meets a preset condition, thereby completing the training of the super-resolution reconstruction generation network model and generating a target image reconstruction model.

[0010] Optionally, the inverse problem of the preset low-resolution model is optimized and solved to obtain a sub-problem, and the process of establishing a super-resolution reconstruction generation network model includes:

[0011] Optimizing and solving the inverse problem of the preset low-resolution model under preset constraints to obtain an optimization problem;

[0012] Setting auxiliary variables of the inverse problem and determining a penalty function of the optimization problem;

[0013] Determine two sub-problems of the optimization problem according to the penalty function;

[0014] The two sub-problems are solved respectively and networked to obtain the super-resolution reconstruction generation network model.

[0015] Optionally, solving the two sub-problems respectively and networking to obtain the super-resolution reconstruction generation network model includes:

[0016] The two sub-problems are solved respectively by using the approximate gradient method to obtain an iterative representation of the solution of the auxiliary variable and the solution of the inverse problem;

[0017] The iterative representation process is networked to obtain the super-resolution reconstruction generation network model.

[0018] Optionally, the super-resolution reconstruction generation network model includes a refinement module and a data consistency module, wherein the refinement module is a neural network module associated with the auxiliary variables.

[0019] Optionally, the discriminant network model is a pooling pyramid network; and the loss function is an adversarial generation network gradient penalty loss function.

[0020] Optionally, the preset low-resolution sampling model is determined by a preset low-resolution undersampling operator, a Fourier transform operator and coil sensitivity parameters.

[0021] In a second aspect, an embodiment of the present invention further provides an image reconstruction method, the method comprising:

[0022] Acquire undersampled K-space data sampled based on a preset low-resolution sampling model;

[0023] The K-space data is input into a target image reconstruction model obtained by the image reconstruction model generation method described in any embodiment to obtain a reconstructed image corresponding to the under-sampled K-space data.

[0024] In a third aspect, an embodiment of the present invention further provides an image reconstruction model generation device, the device comprising:

[0025] A sample processing module, used for acquiring full-sampled magnetic resonance K-space data, and obtaining under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model;

[0026] A sample input module is used to input the under-sampled K-space data into, optimize and solve the inverse problem of the preset low-resolution model to obtain a sub-problem, establish a super-resolution reconstruction generation network model, and obtain an under-sampled reconstructed image;

[0027] A model generation module is used to simultaneously input the under-sampled reconstructed image and the fully sampled reconstructed image corresponding to the fully sampled magnetic resonance K-space data into a discriminant network model associated with the super-resolution reconstruction generation network model, train the super-resolution reconstruction generation network model according to the output result of the discriminant network model, and complete the training of the super-resolution reconstruction generation network model until the loss function of the discriminant network model meets a preset condition, thereby generating a target image reconstruction model.

[0028] Optionally, the image reconstruction model generation device also includes a model construction module, which is used to: optimize and solve the inverse problem of the preset low-resolution model to obtain sub-problems, and establish a super-resolution reconstruction generation network model.

[0029] Optionally, the model construction module is specifically used to:

[0030] Optimizing and solving the inverse problem of the preset low-resolution model under preset constraints to obtain an optimization problem;

[0031] Setting auxiliary variables of the inverse problem and determining a penalty function of the optimization problem;

[0032] Determine two sub-problems of the optimization problem according to the penalty function;

[0033] The two sub-problems are solved respectively and networked to obtain the super-resolution reconstruction generation network model.

[0034] Optionally, the model construction module is further used to:

[0035] The two sub-problems are solved respectively by using the approximate gradient method to obtain an iterative representation of the solution of the auxiliary variable and the solution of the inverse problem;

[0036] The iterative representation process is networked to obtain the super-resolution reconstruction generation network model.

[0037] Optionally, the super-resolution reconstruction generation network model includes a refinement module and a data consistency module, wherein the refinement module is a neural network module associated with the auxiliary variables.

[0038] Optionally, the discriminant network model is a pooling pyramid network; and the loss function is an adversarial generation network gradient penalty loss function.

[0039] Optionally, the preset low-resolution sampling model is determined by a preset low-resolution undersampling operator, a Fourier transform operator and coil sensitivity parameters.

[0040] In a fourth aspect, an embodiment of the present invention further provides an image reconstruction device, the device comprising:

[0041] A data acquisition module, used for acquiring under-sampled K-space data sampled based on a preset low-resolution sampling model;

[0042] An image reconstruction module is used to input the K-space data into the target image reconstruction model obtained by the image reconstruction model generation method described in any embodiment to obtain a reconstructed image corresponding to the under-sampled K-space data.

[0043] In a fifth aspect, an embodiment of the present invention further provides a computer device, the computer device comprising:

[0044] one or more processors;

[0045] A memory for storing one or more programs;

[0046] When the one or more programs are executed by the one or more processors, the one or more processors implement the image reconstruction model generation method or image reconstruction method provided by any embodiment of the present invention.

[0047] In a sixth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an image reconstruction model generation method or an image reconstruction method as provided in any embodiment of the present invention.

[0048] The embodiments of the above invention have the following advantages or beneficial effects:

[0049] In the embodiment of the present invention, the full-sampled magnetic resonance K-space data is processed by using a preset low-resolution sampling model to obtain the corresponding under-sampled K-space data, and then the under-sampled K-space data is input into a super-resolution reconstruction generation network model established by optimizing and solving the inverse problem of the preset low-resolution model to obtain a sub-problem, and the under-sampled reconstructed image is obtained; the under-sampled reconstructed image and the full-sampled reconstructed image corresponding to the full-sampled magnetic resonance K-space data are simultaneously input into a discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to the output result of the discriminant network model until the loss function of the discriminant network model meets the preset conditions, and the super-resolution reconstruction generation network model training is completed to generate a target image reconstruction model for magnetic resonance image reconstruction. The problem that the acceleration multiple of magnetic resonance image reconstruction is limited and the quality of reconstructed images needs to be improved in the prior art is solved, and it is achieved that while increasing the sampling acceleration multiple, high-resolution images can be reconstructed to improve image quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a method for generating an image reconstruction model provided in Embodiment 1 of the present invention;

[0051] Figure 2 1 is a schematic diagram of a network structure of a super-resolution reconstruction network provided in Embodiment 1 of the present invention;

[0052] Figure 3 is a flow chart of an image reconstruction method provided by Embodiment 2 of the present invention;

[0053] Figure 4 is a structural schematic diagram of an image reconstruction model generation device provided by Embodiment 3 of the present invention;

[0054] Figure 5 is a structural schematic diagram of an image reconstruction device provided by Embodiment 4 of the present invention;

[0055] Figure 6 It is a structural diagram of a computer device provided in Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0056] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0057] Embodiment 1

[0058] Figure 1 This is a flowchart of an image reconstruction model generation method provided in Embodiment 1 of the present invention. This embodiment is applicable to the case where image reconstruction model training is performed based on full-sampled magnetic resonance K-space data. The method can be performed by an image reconstruction model generation device, which can be implemented by software and / or hardware and integrated into an electronic device with application development function.

[0059] like Figure 1 As shown, the image reconstruction model generation method includes the following steps:

[0060] S110 , acquiring full-sampled magnetic resonance K-space data, and obtaining under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model.

[0061] Among them, the full-sampled magnetic resonance K-space data is pre-collected sample data, and a high-resolution magnetic resonance image can be reconstructed based on the full-sampled magnetic resonance K-space data. The preset low-resolution sampling model can be determined by a preset low-resolution undersampling operator, a Fourier transform operator, and a coil sensitivity parameter according to the requirements of the acceleration multiple.

[0062] Furthermore, the acceleration factor refers to the acceleration factor of collecting the original data, that is, the full-sampled magnetic resonance K-space data, that is, the under-sampling factor. It can be determined by dividing the total number of data points of the full-sampled magnetic resonance K-space data by the number of collected data points. In this embodiment, the acceleration factor can reach more than 12 times, that is, the image reconstruction model generated by this embodiment can reconstruct a reconstructed image with a sufficiently high resolution at an acceleration factor of more than 12 times.

[0063] Specifically, the preset low-resolution sampling model can be expressed as y=MHFCx (1); where M is the undersampling operator, F is the Fourier transform operator, C is the coil sensitivity matrix (sensitivity maps), and x∈C N and y∈C M They are the K-space vector corresponding to the high-resolution MRI image and the under-sampled K-space vector corresponding to the low-resolution MRI image, H:C N →C MIt is a K-space downsampling operator. The specific operation can be to intercept the central part of the K-space information (low-frequency information) corresponding to the high-resolution image to make it correspond to the K-space of the low-resolution image. The resolution is determined by the size of the sampling matrix (sampling operator). The lower the resolution, the more concentrated the collected K-space points are in the central area, that is, the low-frequency area. When the resolution is improved, the sampling points spread to the higher-frequency area, and more high-frequency details can be obtained, which improves the image resolution. In this step, the full-sampled magnetic resonance K-space data can be used as x, C can be estimated by algorithms such as ESPIRiT or Walsh, F is the Fourier transform operator, and M is a vector composed of 0 and 1, which determines which points are collected and which points are not collected. The value of the points not collected is 0.

[0064] S120, inputting the under-sampled K-space data into, optimizing and solving the inverse problem of the preset low-resolution model to obtain sub-problems, establishing a super-resolution reconstruction generation network model, and obtaining an under-sampled reconstructed image.

[0065] Among them, the super-resolution reconstruction generation network model is constructed based on the results of the inverse problem optimization solution of the preset low-resolution model. The specific process includes the following steps:

[0066] First, the inverse problem of the preset low-resolution model is optimized and solved under the preset constraints to obtain the optimization problem. Specifically, for the convenience of representation, each operator MHFC in formula (1) can be denoted as A. The inverse problem corresponding to formula (1) can be solved by the optimization problem to obtain formula (2),

[0067]

[0068] Here, R(x) is a generalized image domain constraint.

[0069] Then, the auxiliary variables for the optimization solution of the above inverse problem are set, and the penalty function of the optimization problem is determined, and formula (3) can be obtained.

[0070]

[0071] in, It is a coefficient of the penalty term. For the convenience of formula derivation, the denominator is set to 2, and the 2 obtained by taking the derivative of the second norm can be eliminated.

[0072] Furthermore, the alternating linearization minimization algorithm can be used to determine the two sub-problems of the optimization problem according to the penalty function, and obtain formula (4). The solution algorithm is called the alternating linearization minimization algorithm because After linearization, we can record in, is the linearized quadratic term coefficient. It should be noted that η can be replaced by another letter, but in this embodiment, the formula (5) derived later can be simplified in form, so the linearized quadratic term coefficient is taken as

[0073]

[0074] Furthermore, the approximate gradient method can be used to solve the two sub-problems separately, and the iterative steps of the two sub-problems are obtained, as shown in formula (5):

[0075]

[0076] Among them, Prox R is an approximation operator related to the image domain regularization operator R (i.e., image domain constraint); η is the update step size; It is the representation of the image K-space residual in the image domain, which can be specifically expressed by Formula 6:

[0077]

[0078] Finally, the iterative process in formula (5) is networked and expanded to obtain the super-resolution reconstruction generation network model (Super Resolution Reconstruction Network, SRR-Net), which is also a generator. The structure of the network can be referred to Figure 2 In the above network structure, SRR-Net contains multiple network blocks, each of which contains two network modules, namely, refinement module S k and data consistency module X k Specifically, networking can be expressed as

[0079]

[0080] Among them, the refinement module S k In formula (5), the approximation operator Prox R It is networked into a residual 3D convolutional neural network (3D residual CNN) C. C is a three-layer convolutional neural network, and α controls the degree of residual retention. α and step size γ are both learnable network parameters, which bring greater flexibility to the model. Further, by observing formula (7), it can be determined that only x0 needs to be initialized in this model. After x0 is initialized, s1 can be calculated based on x0, and then x1 can be calculated based on s1 to realize a cyclic iterative process. In this embodiment, the initialization value of x0 is taken as the zero-filled image data corresponding to the undersampled K-space data.

[0081] S130. The under-sampled reconstructed image and the fully-sampled reconstructed image corresponding to the fully-sampled magnetic resonance K-space data are simultaneously input into the discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to the output result of the discriminant network model until the loss function of the discriminant network model meets the preset conditions, thereby completing the training of the super-resolution reconstruction generation network model and generating a target image reconstruction model.

[0082] Specifically, a discriminant network is introduced in this step, and the super-resolution reconstruction generation network model is trained based on the idea of ​​generating adversarial networks. In this embodiment, the discriminant network model adopts a pooling pyramid network, and the adversarial generation network gradient penalty loss function (WGAN-GP loss) is used as the loss function of the pooling pyramid network. When the under-sampled reconstructed image and the fully sampled reconstructed image corresponding to the fully sampled magnetic resonance K-space data are simultaneously input into the discriminant network model, the discriminant network model will judge the difference between the under-sampled reconstructed image and the fully sampled reconstructed image. When the loss function value corresponding to the difference meets the preset convergence condition, the model training process is completed, and the target image reconstruction model is obtained.

[0083] The technical solution of this embodiment uses a preset low-resolution sampling model to process the full-sampled magnetic resonance K-space data to obtain the corresponding under-sampled K-space data, and then inputs the under-sampled K-space data into a super-resolution reconstruction generation network model established by optimizing and solving the inverse problem of the preset low-resolution model to obtain a sub-problem, and obtains an under-sampled reconstructed image; the under-sampled reconstructed image and the full-sampled reconstructed image corresponding to the full-sampled magnetic resonance K-space data are simultaneously input into a discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to the output result of the discriminant network model until the loss function of the discriminant network model meets the preset conditions, and the super-resolution reconstruction generation network model training is completed to generate a target image reconstruction model for magnetic resonance image reconstruction. The problem of limited acceleration multiples of magnetic resonance image reconstruction and the need to improve the quality of reconstructed images in the prior art is solved, and high-resolution images can be reconstructed while increasing the sampling acceleration multiple to improve image quality.

[0084] Embodiment 2

[0085] Figure 2 This is a flowchart of an image reconstruction method provided in Embodiment 2 of the present invention. This embodiment is applicable to the case of reconstructing medical images of undersampled magnetic resonance K-space data. The method can be performed by an image reconstruction device, which can be implemented by software and / or hardware and integrated into a computer device with application development function.

[0086] like Figure 2 As shown, the image reconstruction method includes the following steps:

[0087] S210 , obtaining under-sampled K-space data sampled based on a preset low-resolution sampling model.

[0088] Specifically, during the magnetic resonance scanning sampling process, the sampling model can be set according to the sampling speed requirement, and the preset low-resolution model is determined by the preset low-resolution undersampling operator, the Fourier transform operator and the coil sensitivity parameter. The preset low-resolution undersampling operator determines the doubling speed of the sampling.

[0089] The preset low-resolution sampling model can be expressed as y=MHFCx, where M is the undersampling operator, F is the Fourier transform operator, C is the coil sensitivity matrix (sensitivity maps), and x∈C N and y∈C M They are the K-space vector corresponding to the high-resolution MRI image and the under-sampled K-space vector corresponding to the low-resolution MRI image, H:C N →C M It is a K-space downsampling operator. The specific operation can be to intercept the central part of the K-space information (low-frequency information) corresponding to the high-resolution image to make it correspond to the K-space of the low-resolution image. The resolution is determined by the size of the sampling matrix (sampling operator). The lower the resolution, the more concentrated the collected K-space points are in the central area, that is, the low-frequency area. When the resolution is improved, the sampling points spread to the higher-frequency area, and more high-frequency details can be obtained, which improves the image resolution. In this step, the full-sampled magnetic resonance K-space data can be used as x, C can be estimated by algorithms such as ESPIRiT or Walsh, F is the Fourier transform operator, and M is a vector composed of 0 and 1, which determines which points are collected and which points are not collected. The value of the points not collected is 0.

[0090] S220: Input the K-space data into a target image reconstruction model obtained by the image reconstruction model generation method described in any embodiment to obtain a reconstructed image corresponding to the under-sampled K-space data.

[0091] The target image reconstruction model obtained by the image reconstruction model generation method described in any of the above embodiments can reconstruct the low-resolution K-space data to obtain a high-resolution image. Moreover, the acceleration factor of the undersampled low-resolution K-space data can reach more than 12 times, that is, the image reconstruction model generated by this embodiment can reconstruct a reconstructed image with a sufficiently high resolution at an acceleration factor of more than 12 times.

[0092] The technical solution of this embodiment performs sampling by using a preset low-resolution sampling model, and then inputs the sampled data into a trained image reconstruction model to obtain a high-resolution reconstructed image. This solves the problem in the prior art that the acceleration factor of magnetic resonance image reconstruction is limited and the quality of reconstructed images needs to be improved, and achieves the ability to reconstruct high-resolution images while increasing the sampling acceleration factor, thereby improving image quality.

[0093] Embodiment 3

[0094] Figure 4 This is a schematic diagram of the structure of an image reconstruction model generation device provided in Embodiment 3 of the present invention. This embodiment is applicable to the case where image reconstruction model training is performed based on full-sampled magnetic resonance K-space data.

[0095] like Figure 4 As shown, the image reconstruction model generation device includes a sample processing module 310 , a sample input module 320 and a model generation module 330 .

[0096] Among them, the sample processing module 310 is used to obtain full-sampled magnetic resonance K-space data, and obtain under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model; the sample input module 320 is used to input the under-sampled K-space data into a super-resolution reconstruction generation network model established by optimizing and solving the inverse problem of the preset low-resolution model to obtain a sub-problem, and obtain an under-sampled reconstructed image; the model generation module 330 is used to simultaneously input the under-sampled reconstructed image and the full-sampled reconstructed image corresponding to the full-sampled magnetic resonance K-space data into a discriminant network model associated with the super-resolution reconstruction generation network model, train the super-resolution reconstruction generation network model according to the output result of the discriminant network model, until the loss function of the discriminant network model meets the preset conditions, complete the training of the super-resolution reconstruction generation network model, and generate a target image reconstruction model.

[0097] The technical solution of this embodiment uses a preset low-resolution sampling model to process the full-sampled magnetic resonance K-space data to obtain the corresponding under-sampled K-space data, and then inputs the under-sampled K-space data into a super-resolution reconstruction generation network model established by optimizing and solving the inverse problem of the preset low-resolution model to obtain a sub-problem, and obtains an under-sampled reconstructed image; the under-sampled reconstructed image and the full-sampled reconstructed image corresponding to the full-sampled magnetic resonance K-space data are simultaneously input into a discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to the output result of the discriminant network model until the loss function of the discriminant network model meets the preset conditions, and the super-resolution reconstruction generation network model training is completed to generate a target image reconstruction model for magnetic resonance image reconstruction. The problem of limited acceleration multiples of magnetic resonance image reconstruction and the need to improve the quality of reconstructed images in the prior art is solved, and high-resolution images can be reconstructed while increasing the sampling acceleration multiple to improve image quality.

[0098] Optionally, the image reconstruction model generation device also includes a model construction module, which is used to: optimize and solve the inverse problem of the preset low-resolution model to obtain sub-problems, and establish a super-resolution reconstruction generation network model.

[0099] Optionally, the model construction module is specifically used to:

[0100] Optimizing and solving the inverse problem of the preset low-resolution model under preset constraints to obtain an optimization problem;

[0101] Setting auxiliary variables of the inverse problem and determining a penalty function of the optimization problem;

[0102] Determine two sub-problems of the optimization problem according to the penalty function;

[0103] The two sub-problems are solved respectively and networked to obtain the super-resolution reconstruction generation network model.

[0104] Optionally, the model construction module is further used to:

[0105] The two sub-problems are solved respectively by using the approximate gradient method to obtain an iterative representation of the solution of the auxiliary variable and the solution of the inverse problem;

[0106] The iterative representation process is networked to obtain the super-resolution reconstruction generation network model.

[0107] Optionally, the super-resolution reconstruction generation network model includes a refinement module and a data consistency module, wherein the refinement module is a neural network module associated with the auxiliary variables.

[0108] Optionally, the discriminant network model is a pooling pyramid network; and the loss function is an adversarial generation network gradient penalty loss function.

[0109] Optionally, the preset low-resolution sampling model is determined by a preset low-resolution undersampling operator, a Fourier transform operator and coil sensitivity parameters.

[0110] The image reconstruction model generation device provided in the embodiment of the present invention can execute the image reconstruction model generation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0111] Embodiment 4

[0112] Figure 5 This is a schematic diagram of the structure of an image reconstruction device provided in Embodiment 4 of the present invention. This embodiment is applicable to the case of performing medical image reconstruction on under-sampled magnetic resonance K-space data.

[0113] like Figure 5 As shown, the image reconstruction device includes a data acquisition module 410 and an image reconstruction module 420 .

[0114] Among them, the data acquisition module 410 is used to obtain under-sampled K-space data sampled based on a preset low-resolution sampling model; the image reconstruction module 420 is used to input the K-space data into the target image reconstruction model obtained by the image reconstruction model generation method described in any embodiment, and obtain a reconstructed image corresponding to the under-sampled K-space data.

[0115] The technical solution of this embodiment performs sampling by using a preset low-resolution sampling model, and then inputs the sampled data into a trained image reconstruction model to obtain a high-resolution reconstructed image. This solves the problem in the prior art that the acceleration factor of magnetic resonance image reconstruction is limited and the quality of reconstructed images needs to be improved, and achieves the ability to reconstruct high-resolution images while increasing the sampling acceleration factor, thereby improving image quality.

[0116] The image reconstruction device provided in the embodiment of the present invention can execute the image reconstruction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0117] Embodiment 5

[0118] Figure 6 A schematic diagram of the structure of a computer device provided in Embodiment 5 of the present invention. Figure 6 A block diagram of an exemplary computer device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 6The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention. The computer device 12 can be any terminal device with computing capabilities, such as an intelligent controller and server, a mobile phone and other terminal devices.

[0119] like Figure 6 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0120] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0121] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0122] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 6 not shown, usually called a "hard drive"). Although Figure 6 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0123] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0124] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the computer device 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with the other modules of the computer device 12 via the bus 18. It should be understood that although Figure 6 Not shown, other hardware and / or software modules may be used in conjunction with computer device 12, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0125] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing the steps of an image reconstruction model generation method provided in the embodiment of the present invention, the method comprising:

[0126] Acquire full-sampled magnetic resonance K-space data, and obtain under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model;

[0127] The under-sampled K-space data is inputted into, a sub-problem is obtained by optimizing and solving the inverse problem of the preset low-resolution model, and a super-resolution reconstruction generation network model is established to obtain an under-sampled reconstructed image;

[0128] The under-sampled reconstructed image and the fully-sampled reconstructed image corresponding to the fully-sampled magnetic resonance K-space data are simultaneously input into a discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to an output result of the discriminant network model until a loss function of the discriminant network model meets a preset condition, thereby completing the training of the super-resolution reconstruction generation network model and generating a target image reconstruction model.

[0129] Alternatively, an image reconstruction method provided by an embodiment of the present invention may be implemented, the method comprising:

[0130] Acquire undersampled K-space data sampled based on a preset low-resolution sampling model;

[0131] The K-space data is input into a target image reconstruction model obtained by the image reconstruction model generation method described in any embodiment to obtain a reconstructed image corresponding to the under-sampled K-space data.

[0132] Embodiment 6

[0133] This embodiment 6 provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the image reconstruction model generation method provided in any embodiment of the present invention is implemented, including:

[0134] Acquire full-sampled magnetic resonance K-space data, and obtain under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model;

[0135] The under-sampled K-space data is inputted into, a sub-problem is obtained by optimizing and solving the inverse problem of the preset low-resolution model, and a super-resolution reconstruction generation network model is established to obtain an under-sampled reconstructed image;

[0136] The under-sampled reconstructed image and the fully-sampled reconstructed image corresponding to the fully-sampled magnetic resonance K-space data are simultaneously input into a discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to an output result of the discriminant network model until a loss function of the discriminant network model meets a preset condition, thereby completing the training of the super-resolution reconstruction generation network model and generating a target image reconstruction model.

[0137] Alternatively, an image reconstruction method provided by an embodiment of the present invention may be implemented, the method comprising:

[0138] Acquire undersampled K-space data sampled based on a preset low-resolution sampling model;

[0139] The K-space data is input into a target image reconstruction model obtained by the image reconstruction model generation method described in any embodiment to obtain a reconstructed image corresponding to the under-sampled K-space data.

[0140] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0141] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0142] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0143] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0144] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0145] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for generating an image reconstruction model, characterized in that: include: Acquire full-sampled magnetic resonance K-space data, and obtain under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model; The under-sampled K-space data is input into a super-resolution reconstruction generation network model to obtain an under-sampled reconstructed image; wherein the super-resolution reconstruction generation network model is a model established by optimizing and solving the inverse problem of the preset low-resolution model to obtain a sub-problem; the inverse problem of the preset low-resolution model is optimized and solved to obtain a sub-problem, and the process of establishing the super-resolution reconstruction generation network model includes: optimizing and solving the inverse problem of the preset low-resolution model under preset constraints to obtain an optimization problem; setting auxiliary variables of the inverse problem and determining a penalty function of the optimization problem; determining two sub-problems of the optimization problem according to the penalty function; solving the two sub-problems respectively, and networking to obtain the super-resolution reconstruction generation network model; The under-sampled reconstructed image and the fully-sampled reconstructed image corresponding to the fully-sampled magnetic resonance K-space data are simultaneously input into a discriminant network model associated with the super-resolution reconstruction generation network model, and the super-resolution reconstruction generation network model is trained according to an output result of the discriminant network model until a loss function of the discriminant network model meets a preset condition, thereby completing the training of the super-resolution reconstruction generation network model and generating a target image reconstruction model.

2. The method according to claim 1, characterized in that The two sub-problems are solved respectively and the super-resolution reconstruction generation network model is obtained through networking, including: The two sub-problems are solved respectively by using the approximate gradient method to obtain an iterative representation of the solution of the auxiliary variable and the solution of the inverse problem; The iterative representation process is networked to obtain the super-resolution reconstruction generation network model.

3. The method according to claim 1 or 2, characterized in that: The super-resolution reconstruction generation network model includes a refinement module and a data consistency module, wherein the refinement module is a neural network module associated with the auxiliary variables.

4. The method according to claim 1, characterized in that The discriminant network model is a pooling pyramid network; and the loss function is an adversarial generation network gradient penalty loss function.

5. The method according to claim 1, characterized in that The preset low-resolution sampling model is determined by a preset low-resolution undersampling operator, a Fourier transform operator and coil sensitivity parameters.

6. An image reconstruction method, characterized in that: include: Acquire undersampled K-space data sampled based on a preset low-resolution sampling model; The K-space data is input into a target image reconstruction model obtained by the image reconstruction model generation method according to any one of claims 1 to 5 to obtain a reconstructed image corresponding to the under-sampled K-space data.

7. An image reconstruction model generation device, characterized in that: include: A sample processing module, used for acquiring full-sampled magnetic resonance K-space data, and obtaining under-sampled K-space data corresponding to the full-sampled magnetic resonance K-space data based on a preset low-resolution sampling model; A sample input module is used to input the under-sampled K-space data into a super-resolution reconstruction generation network model to obtain an under-sampled reconstructed image; wherein the super-resolution reconstruction generation network model is a model established by optimizing and solving the inverse problem of the preset low-resolution model to obtain a sub-problem; the inverse problem of the preset low-resolution model is optimized and solved to obtain a sub-problem, and the process of establishing the super-resolution reconstruction generation network model includes: optimizing and solving the inverse problem of the preset low-resolution model under preset constraints to obtain an optimization problem; setting auxiliary variables of the inverse problem and determining a penalty function of the optimization problem; determining two sub-problems of the optimization problem according to the penalty function; solving the two sub-problems respectively, and networking to obtain the super-resolution reconstruction generation network model; A model generation module is used to simultaneously input the under-sampled reconstructed image and the fully sampled reconstructed image corresponding to the fully sampled magnetic resonance K-space data into a discriminant network model associated with the super-resolution reconstruction generation network model, train the super-resolution reconstruction generation network model according to the output result of the discriminant network model, and complete the training of the super-resolution reconstruction generation network model until the loss function of the discriminant network model meets a preset condition, thereby generating a target image reconstruction model.

8. An image reconstruction device, characterized in that: include: A data acquisition module, used for acquiring under-sampled K-space data sampled based on a preset low-resolution sampling model; An image reconstruction module is used to input the K-space data into the target image reconstruction model obtained by any image reconstruction model generation method in claims 1-5 to obtain a reconstructed image corresponding to the undersampled K-space data.

9. A computer device, characterized in that: The computer device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the image reconstruction model generation method or the image reconstruction method as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the image reconstruction model generation method or the image reconstruction method as described in any one of claims 1 to 6 is implemented.

Images