Super-resolution reconstruction method and apparatus for magnetic resonance image, device, and storage medium

Abstract

Embodiments of the present application belong to the technical field of image enhancement, and provide a super-resolution reconstruction method and apparatus for a magnetic resonance image, a device, and a storage medium. The method comprises: incorporating a total variation norm and a learnable regularization term into an image reconstruction model, inputting Gaussian random noise and a first low-resolution magnetic resonance image into the image reconstruction model, performing high-resolution image reconstruction on the Gaussian random noise, and obtaining a first high-resolution magnetic resonance image; by means of the image reconstruction model, degrading the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image; on the basis of the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image, optimizing network parameters of the image reconstruction model, and obtaining optimized network parameters; and, on the basis of the optimized network parameters, reconstructing the Gaussian random noise into a target high-resolution magnetic resonance image, so that the generalization ability and the reconstruction effect of the image reconstruction model are improved.

Description

Magnetic resonance image super-resolution reconstruction method, device, equipment and storage medium Technical Field

[0001] The present application relates to the field of image enhancement technology, and in particular to a method, apparatus, device and storage medium for super-resolution reconstruction of magnetic resonance images. Background Art

[0002] Magnetic resonance imaging (MRI) is currently widely used in the field of medical imaging. Compared with computed tomography (CT / PET), MRI does not cause radiation damage and can be used for multifaceted, multi-parameter imaging. However, due to factors such as hardware equipment, scanning time, signal-to-noise ratio, and patient physical fitness, it is often difficult to quickly obtain high-resolution MRI images. Furthermore, long scanning times can cause discomfort to some patients.

[0003] In related technologies, low-resolution MRI images are reconstructed into high-resolution MRI images using a trained image reconstruction model. However, training this model requires a large number of high- and low-resolution MRI images. However, obtaining a large number of high-resolution MRI images is difficult, and the lack of training data leads to overfitting of the image reconstruction model, resulting in poor reconstruction results. Furthermore, the diversity and complexity of MRI images require the image reconstruction model to have higher generalization capabilities and reliability. Therefore, improving the generalization capabilities and reconstruction results of image reconstruction models has become a pressing technical problem.

[0004] Summary of the Invention

[0005] The main purpose of the embodiments of the present application is to propose a method, apparatus, device and storage medium for super-resolution reconstruction of magnetic resonance images, aiming to improve the generalization ability and reconstruction effect of the image reconstruction model.

[0006] To achieve the above objectives, a first aspect of an embodiment of the present application provides a method for super-resolution reconstruction of magnetic resonance images, the method comprising:

[0007] acquiring a first low-resolution magnetic resonance image;

[0008] Inputting Gaussian random noise and the first low-resolution magnetic resonance image into an image reconstruction model, and performing high-resolution image reconstruction on the Gaussian random noise using the image reconstruction model to obtain a first high-resolution magnetic resonance image; wherein the image reconstruction model is obtained by combining a total variation norm and a learnable regularization term with a preset image reconstruction model;

[0009] degenerating the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image using the image reconstruction model;

[0010] optimizing the network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters;

[0011] The Gaussian random noise is reconstructed into a target high-resolution magnetic resonance image according to the optimized network parameters.

[0012] In some embodiments, optimizing the network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters includes:

[0013] Obtain a network parameter optimization formula for the image reconstruction model; wherein the network parameter optimization formula is characterized as:

[0014] Among them, θ * represents the optimal solution of θ, represents the value of θ to be minimized, A represents the downsampling operator, x represents the first low-resolution magnetic resonance image, G(θ; z) represents the first high-resolution magnetic resonance image generated with network parameters θ and Gaussian random noise z, AG(θ; z) represents the second low-resolution magnetic resonance image, ‖·‖ TV represents the total variation norm, represents the learnable regular term, λ1 and λ2 represent penalty parameters;

[0015] The network parameter optimization formula is iteratively solved according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters.

[0016] In some embodiments, iteratively solving the network parameter optimization formula according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters includes:

[0017] Preset auxiliary variables are introduced, and the variable splitting technique is used to convert the network parameter optimization formula into an equivalent constraint optimization formula; wherein the equivalent constraint optimization formula is characterized as follows: sty=G(θ;z),t=G(θ;z);

[0018] Among them, st represents the constraint condition, y and t represent auxiliary variables;

[0019] The equivalent constraint optimization formula is iteratively solved according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters.

[0020] In some embodiments, iteratively solving the equivalent constraint optimization formula according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters includes:

[0021] The equivalent constrained optimization formula is equivalently replaced by an unconstrained augmented Lagrangian function; wherein the augmented Lagrangian function is characterized as:

[0022] Among them, μ1 and μ2 are penalty parameters;

[0023] The augmented Lagrangian function is solved according to the first low-resolution magnetic resonance image, the second low-resolution magnetic resonance image and a half-splitting algorithm to obtain optimized network parameters.

[0024] In some embodiments, acquiring a first low-resolution magnetic resonance image includes:

[0025] acquiring a second high-resolution magnetic resonance image;

[0026] The second high-resolution magnetic resonance image is downsampled to obtain the first low-resolution magnetic resonance image.

[0027] In some embodiments, after reconstructing the Gaussian random noise into a target high-resolution magnetic resonance image according to the optimized network parameters, the method further includes:

[0028] acquiring a peak signal-to-noise ratio of the second high-resolution magnetic resonance image and the target high-resolution magnetic resonance image;

[0029] acquiring a structural similarity index between the second high-resolution magnetic resonance image and the target high-resolution magnetic resonance image;

[0030] An evaluation result of the target high-resolution magnetic resonance image is obtained according to the peak signal-to-noise ratio and the structural similarity index.

[0031] In some embodiments, acquiring a second high-resolution magnetic resonance image comprises:

[0032] Acquire three-dimensional high-resolution magnetic resonance images;

[0033] The three-dimensional high-resolution magnetic resonance image is subjected to two-dimensional slicing processing to obtain the second two-dimensional high-resolution magnetic resonance image.

[0034] To achieve the above-mentioned object, a second aspect of an embodiment of the present application provides a magnetic resonance image super-resolution reconstruction device, the device comprising:

[0035] an acquisition module, configured to acquire a first low-resolution magnetic resonance image;

[0036] an input module, configured to input Gaussian random noise and the first low-resolution magnetic resonance image into the image reconstruction model, and perform high-resolution image reconstruction on the Gaussian random noise using the image reconstruction model to obtain a first high-resolution image; wherein the image reconstruction model is obtained by combining a total variation norm and a learnable regularization term with a preset image reconstruction model;

[0037] a degradation module, configured to degenerate the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image by using the image reconstruction model;

[0038] an optimization module, configured to optimize network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters;

[0039] A reconstruction module is used to reconstruct the Gaussian random noise into a target high-resolution magnetic resonance image according to the optimized network parameters.

[0040] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the method described in the first aspect when executing the computer program.

[0041] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method described in the first aspect.

[0042] The magnetic resonance image super-resolution reconstruction method, apparatus, device and storage medium proposed in this application combine the total variation norm and the learnable regularization term into a preset image reconstruction model, so that the training process of the image reconstruction model based on the hybrid regularization term deep image prior does not rely on a large number of high-low resolution magnetic resonance images, but captures the prior information of the magnetic resonance image through the network architecture itself, and further characterizes the complex structure of the magnetic resonance image through the hybrid regularization term, thereby improving the generalization ability of the image reconstruction model and achieving high-quality reconstruction of the magnetic resonance image. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] FIG1 is a flow chart of a method for super-resolution reconstruction of magnetic resonance images provided by an embodiment of the present application;

[0044] FIG2 is a schematic diagram of a neural network architecture of an image reconstruction model provided in an embodiment of the present application;

[0045] FIG3 is a flow chart of step S101 in FIG1 ;

[0046] FIG4 is a flowchart of another magnetic resonance image super-resolution reconstruction method provided by an embodiment of the present application;

[0047] FIG5 is a flow chart of step S301 in FIG3 ;

[0048] FIG6 is a schematic structural diagram of a magnetic resonance image super-resolution reconstruction apparatus provided in an embodiment of the present application;

[0049] FIG7 is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0051] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0053] First, let’s analyze some of the terms used in this application:

[0054] Magnetic resonance image: also known as MRI (Magnetic Resonance Imaging) image, is a medical image that uses magnetic fields, radio waves, and computer technology to combine imaging detectors to generate images of the internal structure of the human body.

[0055] Super-resolution: refers to the process of improving the resolution of low-resolution images (LR images) to convert them into high-resolution images (HR images) through image processing technology.

[0056] The Total Variation (TV) norm is a frequently used regularization method in image processing. It sums the gradients of an image and is often used for tasks such as image denoising and reconstruction. When used as a regularization term, the TV norm encourages smoothing of varying regions (contours, edges, etc.) in the image being processed, thereby achieving both denoising and reconstruction.

[0057] Learnable regularization: This involves setting the weight coefficients in the regularization term as learnable parameters. Using an iterative optimization algorithm, the regularization term learns the optimal weight value during model training. This approach adaptively controls the smoothness of the regularization term, achieving denoising and reconstruction effects at varying granularities and offering greater flexibility.

[0058] Due to factors such as hardware equipment, scanning time, signal-to-noise ratio, and patient physical fitness, it is often difficult to quickly obtain high-resolution MRI images. In related technologies, it is possible to obtain low-resolution MRI images and then reconstruct them into high-resolution MRI images using a trained deep learning model. However, the training process of deep learning models requires a large number of high- and low-resolution MRI images, which are difficult to obtain. The lack of training data makes deep learning models prone to overfitting and poor reconstruction results.

[0059] Based on this, the embodiments of the present application provide a method, apparatus, device and storage medium for super-resolution reconstruction of magnetic resonance images, which aims to combine the total variation norm and learnable regularization term into a preset image reconstruction model, and complete the iterative training of the model through a single low-resolution image, so that the training process of the image reconstruction model based on the hybrid regularization term deep image prior does not depend on a large number of high-low resolution magnetic resonance images, and further characterizes the complex structure of the magnetic resonance image through the hybrid regularization term, thereby improving the generalization ability of the image reconstruction model and achieving high-quality reconstruction of the magnetic resonance image.

[0060] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0061] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0062] The magnetic resonance image super-resolution reconstruction method provided in the embodiments of the present application can be applied to a terminal or a server, or can be software running on a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet computer, laptop computer, desktop computer, etc.; the server can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, or as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the magnetic resonance image super-resolution reconstruction method, etc., but is not limited to the above forms.

[0063] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0064] The magnetic resonance image super-resolution reconstruction method, apparatus, device and storage medium provided in the embodiments of the present application are specifically described through the following embodiments. First, the magnetic resonance image super-resolution reconstruction method in the embodiments of the present application is described.

[0065] Please refer to FIG. 1 , which is an optional flowchart of a magnetic resonance image super-resolution reconstruction method provided by an embodiment of the present application. The method in FIG. 1 may include but is not limited to steps S101 to S105 .

[0066] Step S101, acquiring a first low-resolution magnetic resonance image;

[0067] Step S102: inputting Gaussian random noise and the first low-resolution magnetic resonance image into an image reconstruction model, and performing high-resolution image reconstruction on the Gaussian random noise using the image reconstruction model to obtain a first high-resolution magnetic resonance image; wherein the image reconstruction model is obtained by combining a total variation norm and a learnable regularization term with a preset image reconstruction model;

[0068] Step S103, degenerating the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image using an image reconstruction model;

[0069] Step S104, optimizing the network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters;

[0070] Step S105 : reconstructing the Gaussian random noise into a target high-resolution magnetic resonance image according to the optimized network parameters.

[0071] In some embodiments, in steps S101 to S102, the image reconstruction model includes a skip connection network, an encoder and a decoder. After Gaussian random noise and the first low-resolution magnetic resonance image are input into the image reconstruction model, the Gaussian random noise extracts deep features through the encoder part, and the deep features are then used to generate the first high-resolution magnetic resonance image through the decoder.

[0072] In steps S103 to S104 of some embodiments, a downsampling operator is used to degenerate the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image. The first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image are then iteratively optimized for network parameters of the image reconstruction model, and finally the optimal network parameters are output.

[0073] Specifically, optimizing the network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain the optimized network parameters includes:

[0074] The first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image are made consistent under the constraints of the observation model, and the solution space is constrained by the total variation norm and the learnable regularization term until the optimal network parameters are obtained.

[0075] It's important to note that in image reconstruction, the observation model refers to the sampling model used in magnetic resonance imaging. Its purpose is to model the sampled signals, addressing the problem of how to better reconstruct images in situations such as incomplete information and noise. The observation model can be used to model and describe the relationship between the sampled signals and the image, enabling mathematical methods to generate a reconstruction that more closely resembles the true image. The solution space is the space consisting of all images that meet specific conditions.

[0076] In step S105 of some embodiments, after obtaining the optimal network parameters, Gaussian random noise is reconstructed into a target high-resolution magnetic resonance image according to the optimal network parameters, thereby obtaining a high-resolution magnetic resonance image corresponding to the first low-resolution magnetic resonance image.

[0077] In steps S101 to S105, as shown in the embodiment of the present application, a first low-resolution magnetic resonance image is acquired; Gaussian random noise and the first low-resolution magnetic resonance image are input into an image reconstruction model, and a high-resolution image reconstruction is performed on the Gaussian random noise to obtain the first high-resolution magnetic resonance image; wherein the image reconstruction model is obtained by combining a total variation norm and a learnable regularization term with a preset image reconstruction model; the first high-resolution magnetic resonance image is degraded into a second low-resolution magnetic resonance image via the image reconstruction model; network parameters of the image reconstruction model are optimized based on the first and second low-resolution magnetic resonance images to obtain optimized network parameters; and the Gaussian random noise is reconstructed into a target high-resolution magnetic resonance image based on the optimized network parameters. In other words, in this embodiment, the training process of the image reconstruction model and the image reconstruction process are combined, and the optimal network parameters of the image reconstruction model are obtained from the iterations by iteratively performing training on a low-resolution magnetic resonance image, completing the training of the image reconstruction model, and then reconstructing the Gaussian random noise into a high-resolution magnetic resonance image of the low-resolution magnetic resonance image using the optimized network parameters. It solves the problem in related technologies that a large amount of training data is required to train deep learning models, and the lack of training data leads to overfitting and poor generalization ability. It also further characterizes the complex structure of the image through mixed regularization terms, thereby improving the reconstruction effect of high-resolution magnetic resonance images.

[0078] In some embodiments, step S104 may include the following steps:

[0079] The total variation norm and the learnable regularization term are combined into the preset image reconstruction model. The network parameter optimization formula of the image reconstruction model is shown in the following formula 1:

[0080] Among them, θ * represents the optimal solution of θ, represents the value of θ to be minimized, A represents the downsampling operator, x represents the first low-resolution magnetic resonance image, G(θ; z) represents the first high-resolution magnetic resonance image generated by the network parameter θ and the Gaussian random noise z, AG(θ; z) represents the downsampling operator A that degenerates the first high-resolution magnetic resonance image G(θ; z) into the second low-resolution magnetic resonance image, ‖·‖ TV represents the total variation norm, represents the learnable regular term, λ1 and λ2 represent penalty parameters;

[0081] The network parameter optimization formula is iteratively solved according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain the optimal network parameters.

[0082] In this embodiment, the complex structure of the image is further characterized by the total variation norm and a learnable regularization term, thereby improving the reconstruction effect of high-resolution images. Without the need for a large amount of training data, the optimal network parameters are obtained during the iteration by iterating only a low-resolution magnetic resonance image, and the training process of the image reconstruction model is completed, thus solving the problem of model overfitting and poor generalization ability caused by lack of training data.

[0083] Due to the introduction of the mixed regularization term, the optimization problem of Formula 1 becomes complicated. To solve Formula 1, in some embodiments, the network parameter optimization formula is iteratively solved according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain the optimized network parameters, including:

[0084] Preset auxiliary variables are introduced, and the variable splitting technique is used to convert the network parameter optimization formula into an equivalent constraint optimization formula. The equivalent constraint optimization formula is shown in the following formula 2: sty=G(θ;z),t=G(θ;z), (2)

[0085] Among them, st represents the constraint condition, y and t represent auxiliary variables;

[0086] An equivalent constraint optimization formula is iteratively solved according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimal network parameters.

[0087] In this embodiment, variable splitting refers to a technique in which the variable to be optimized is divided into two parts (or multiple parts) in the optimization algorithm and weighed according to different penalty coefficients to obtain a more optimal solution. Through this embodiment, the solution of Formula 1 is achieved.

[0088] In order to solve the minimization problem in Formula 2, in some embodiments, the equivalent constraint optimization formula is iteratively solved according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters, including:

[0089] The equivalent constrained optimization formula is equivalently replaced by the unconstrained augmented Lagrangian function; wherein the augmented Lagrangian function is shown in the following formula 3:

[0090] Among them, μ1 and μ2 are penalty parameters;

[0091] Based on the first and second low-resolution magnetic resonance images, the half-splitting algorithm is used to solve the augmented Lagrangian function and obtain the optimal network parameters. The half-splitting (HQS) algorithm is an efficient optimization algorithm for solving complex non-convex problems. The algorithm can solve the entire problem by breaking it down into several sub-problems through splitting and merging.

[0092] Specifically, the steps for solving Formula 3 using the HQS algorithm are as follows:

[0093] Input: Unconstrained augmented Lagrangian function: L; Iteration threshold number: T; CNN: G(θ k ; z); denoiser; parameters: λ1, λ2, μ1, μ2; regularization term: ‖y‖ TV , First low-resolution magnetic resonance image LR: x; downsampling operator: A; initialization variables: θ0, x0; step size of network parameter θ: β; step size optimizer of θ: Opt.

[0094] 1.For k=0,1,…,T-1do

[0095] 2. solved by y k+1 =prox TV (G(θ k ; z), λ1 / μ1)

[0096] 3. solved by t k+1 =denoiser(G(θ k ; z), λ2 / μ2)

[0097] 4.

[0098] Gradient descent solution

[0099] 5.End for outputting the target high-resolution magnetic resonance image HR:

[0100] Among them, prox TV It is a generalized projected gradient operator used to solve subproblems in sparse representation problems.

[0101] In practical applications, please refer to Figure 2, which shows the neural network architecture of the image reconstruction model. The backbone of this neural network architecture is a structure similar to a U-Net (convolutional neural network), consisting of a skip connection network, an encoder, and a decoder. Gaussian random noise is used to extract deep features through the encoder, and the decoder generates an estimated high-resolution magnetic resonance image. The downsampling operator then degenerates it into a second low-resolution magnetic resonance image. The augmented Lagrangian function L is minimized based on the second low-resolution magnetic resonance image and the input first low-resolution magnetic resonance image to obtain the optimal network parameter θ. * , and then according to the optimal network parameters θ * Reconstruct Gaussian random noise into a target high-resolution magnetic resonance image.

[0102] Please refer to FIG. 3 . In some embodiments, step S101 includes but is not limited to steps S301 to S302 :

[0103] Step S301, acquiring a second high-resolution magnetic resonance image;

[0104] Step S302 : downsampling the second high-resolution magnetic resonance image to obtain a first low-resolution magnetic resonance image.

[0105] In steps S301 to S302 shown in this embodiment, the first low-resolution magnetic resonance image can be obtained in a variety of ways. When the first low-resolution magnetic resonance image is obtained by downsampling the second high-resolution magnetic resonance image, when reconstructing the target-resolution magnetic resonance image of the first low-resolution magnetic resonance image, the reconstruction effect can be judged by comparing the second high-resolution magnetic resonance image with the target-resolution magnetic resonance image.

[0106] Referring to FIG. 4 , in some embodiments, after step S105 , the magnetic resonance image super-resolution reconstruction method may further include but is not limited to steps S401 to S403 :

[0107] Step S401, obtaining a peak signal-to-noise ratio of a second high-resolution magnetic resonance image and a target high-resolution magnetic resonance image;

[0108] Step S402 , obtaining a structural similarity index between the second high-resolution magnetic resonance image and the target high-resolution magnetic resonance image;

[0109] Step S403 : obtaining an evaluation result of the target high-resolution magnetic resonance image according to the peak signal-to-noise ratio and the structural similarity index.

[0110] In steps S401 to S403 of this embodiment, the peak signal-to-noise ratio (PSNR) is a metric used to measure digital image quality. It is obtained by measuring the difference between the original image and the processed image, namely, the average of the squared differences between the pixel values ​​of the original image and the processed image. A larger PSNR value indicates that the difference between the processed image and the original image is smaller. The structural similarity index (SSIM) is a metric used to measure the similarity between two images. It evaluates the similarity between the two images by calculating the similarity in brightness, contrast, and structure. A larger SSIM value indicates a higher similarity between the processed image and the original image. By calculating the peak signal-to-noise ratio and structural similarity index of the second high-resolution MRI image and the target high-resolution MRI image, an evaluation result of the target high-resolution MRI image is obtained. The evaluation result can be used to determine the reconstruction effect of the image reconstruction model.

[0111] In practical applications, the proposed magnetic resonance image super-resolution reconstruction method was compared with bicubic interpolation (Bicubic), deep image prior (DIP), and total variation (TV). Peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) were selected as metrics for estimating effectiveness. The numerical results are shown in Table 1 below.

[0112] Table 1

[0113] In Table 1, T1W represents T1-weighted MRI images, and T2W represents T2-weighted MRI images. T1-weighted and T2-weighted MRI images are two commonly used imaging sequences in MRI imaging. It can be seen that the magnetic resonance image super-resolution reconstruction method of the present invention achieves higher PSNR and SSIM values ​​for both T1W and T2W images, demonstrating that the magnetic resonance image super-resolution reconstruction method of the present invention achieves better reconstruction results.

[0114] Referring to FIG. 5 , in some embodiments, step S301 includes but is not limited to steps S501 to S502:

[0115] Step S501, acquiring a three-dimensional high-resolution magnetic resonance image;

[0116] Step S502 : performing two-dimensional slicing processing on the three-dimensional high-resolution magnetic resonance image to obtain a two-dimensional second high-resolution magnetic resonance image.

[0117] In steps S501 and S502 of this embodiment, a 5T MRI scanner can be used to obtain T1-weighted and T2-weighted magnetic resonance images of a volunteer, i.e., three-dimensional high-resolution magnetic resonance images. These images are then subjected to two-dimensional slicing to obtain a two-dimensional second-high-resolution magnetic resonance image. If reconstruction of the three-dimensional low-resolution magnetic resonance image is required, the three-dimensional low-resolution magnetic resonance image can be subjected to two-dimensional slicing to obtain multiple two-dimensional low-resolution magnetic resonance images. These multiple two-dimensional low-resolution magnetic resonance images are then reconstructed using the magnetic resonance image super-resolution reconstruction method of the present invention to obtain multiple two-dimensional high-resolution magnetic resonance images. The multiple two-dimensional high-resolution magnetic resonance images are then three-dimensionally spliced ​​to obtain a reconstructed three-dimensional high-resolution magnetic resonance image.

[0118] Referring to FIG. 6 , an embodiment of the present application further provides a magnetic resonance image super-resolution reconstruction device, which can implement the above-mentioned magnetic resonance image super-resolution reconstruction method. The device includes:

[0119] An acquisition module 601 is configured to acquire a first low-resolution magnetic resonance image;

[0120] An input module 602 is configured to input Gaussian random noise and a first low-resolution magnetic resonance image into an image reconstruction model, and perform high-resolution image reconstruction on the Gaussian random noise to obtain a first high-resolution image; wherein the image reconstruction model is obtained by combining a total variation norm and a learnable regularization term with a preset image reconstruction model;

[0121] a degradation module 603, configured to degrade the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image using an image reconstruction model;

[0122] An optimization module 604 is configured to optimize network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters;

[0123] The reconstruction module 605 is configured to reconstruct the Gaussian random noise into a target high-resolution magnetic resonance image according to the optimized network parameters.

[0124] The specific implementation of the magnetic resonance image super-resolution reconstruction device is basically the same as the specific embodiment of the magnetic resonance image super-resolution reconstruction method described above, and will not be repeated here.

[0125] The present application also provides an electronic device comprising a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described magnetic resonance image super-resolution reconstruction method. The electronic device can be any smart terminal, such as a tablet computer or an in-vehicle computer.

[0126] Please refer to FIG7 , which illustrates a hardware structure of an electronic device according to another embodiment. The electronic device includes:

[0127] The processor 301 may be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0128] The memory 302 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 302 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 302 and is called by the processor 301 to execute the magnetic resonance image super-resolution reconstruction method of the embodiments of this application.

[0129] Input / output interface 303, used to implement information input and output;

[0130] Communication interface 304, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0131] bus 305 , which transmits information between the various components of the device (e.g., processor 301 , memory 302 , input / output interface 303 , and communication interface 304 );

[0132] The processor 301 , the memory 302 , the input / output interface 303 and the communication interface 304 are connected to each other in communication within the device via the bus 305 .

[0133] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, which implements the above-mentioned magnetic resonance image super-resolution reconstruction method when executed by a processor.

[0134] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0135] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0136] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0137] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0138] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0139] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0140] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0141] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0142] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0143] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0144] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: various media that can store programs, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0145] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A method for super-resolution reconstruction of magnetic resonance images, characterized in that, the method includes: obtaining a first low-resolution magnetic resonance image; inputting Gaussian random noise and the first low-resolution magnetic resonance image into an image reconstruction model, and performing high-resolution image reconstruction on the Gaussian random noise through the image reconstruction model to obtain a first high-resolution magnetic resonance image; wherein, the image reconstruction model is obtained by combining the total variation norm and a learnable regularization term into a preset image reconstruction model; degrading the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image through the image reconstruction model; optimizing the network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters; reconstructing the Gaussian random noise into a target high-resolution magnetic resonance image according to the optimized network parameters.

2. The method according to claim 1, characterized in that, the optimizing the network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters includes: Obtain the network parameter optimization formula of the image reconstruction model; wherein, the network parameter optimization formula is characterized as: where, θ * represents the optimal solution of θ, denotes the θ value to be minimized, A denotes the downsampling operator, x denotes the first low-resolution magnetic resonance image, G(θ; z) denotes the first high-resolution magnetic resonance image generated by the network with parameters θ and Gaussian random noise z, AG(θ; z) is the second low-resolution magnetic resonance image, and ‖·‖ TV denotes the total variation norm, Denote learnable regularization terms, λ 1 and λ 2 denote penalty parameters; iteratively solving the network parameter optimization formula according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters.

3. The method according to claim 2, characterized in that, the iteratively solving the network parameter optimization formula according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters includes: Introduce a preset auxiliary variable, and use the variable splitting technique to transform the network parameter optimization formula into an equivalent constrained optimization formula; wherein, the equivalent constrained optimization formula is characterized as: s.t. y = G(θ; z), t = G(θ; z); wherein, s.t. represents a constraint condition, and y and t represent auxiliary variables; iteratively solving the equivalent constraint optimization formula according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters.

4. The method according to claim 3, characterized in that, the iteratively solving the equivalent constraint optimization formula according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters includes: Equivalently replace the equivalent constraint optimization formula with an unconstrained augmented Lagrangian function; wherein, the augmented Lagrangian function is characterized as: where, μ 1 and μ 2 are penalty parameters; solving the augmented Lagrangian function according to the first low-resolution magnetic resonance image, the second low-resolution magnetic resonance image and the semi-splitting algorithm to obtain optimized network parameters.

5. The method according to claim 1, characterized in that, the obtaining the first low-resolution magnetic resonance image includes: obtaining a second high-resolution magnetic resonance image; performing downsampling on the second high-resolution magnetic resonance image to obtain the first low-resolution magnetic resonance image.

6. The method according to claim 5, characterized in that, after reconstructing the Gaussian random noise into a target high-resolution magnetic resonance image according to the optimized network parameters, the method further includes: obtaining the peak signal-to-noise ratio of the second high-resolution magnetic resonance image and the target high-resolution magnetic resonance image; obtaining the structural similarity index of the second high-resolution magnetic resonance image and the target high-resolution magnetic resonance image; Based on the peak signal-to-noise ratio and the structural similarity index, an evaluation result of the target high-resolution magnetic resonance image is obtained.

7. The method according to claim 5, wherein, the obtaining of the second high-resolution magnetic resonance image includes: obtaining a three-dimensional high-resolution magnetic resonance image; performing two-dimensional slicing on the three-dimensional high-resolution magnetic resonance image to obtain the two-dimensional second high-resolution magnetic resonance image.

8. A magnetic resonance image super-resolution reconstruction device, wherein, the device includes: an obtaining module, configured to obtain a first low-resolution magnetic resonance image; an input module, configured to input Gaussian random noise and the first low-resolution magnetic resonance image into the image reconstruction model, and perform high-resolution image reconstruction on the Gaussian random noise through the image reconstruction model to obtain a first high-resolution image; wherein, the image reconstruction model is obtained by combining a total variation norm and a learnable regularization term into a preset image reconstruction model; a degradation module, configured to degrade the first high-resolution magnetic resonance image into a second low-resolution magnetic resonance image through the image reconstruction model; an optimization module, configured to optimize network parameters of the image reconstruction model according to the first low-resolution magnetic resonance image and the second low-resolution magnetic resonance image to obtain optimized network parameters; a reconstruction module, configured to reconstruct the Gaussian random noise into a target high-resolution magnetic resonance image according to the optimized network parameters.

9. An electronic device, wherein, the electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the magnetic resonance image super-resolution reconstruction method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, wherein, when the computer program is executed by a processor, the magnetic resonance image super-resolution reconstruction method according to any one of claims 1 to 7 is implemented. ​

Images