Magnetic resonance image reconstruction method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of magnetic resonance image reconstruction, and discloses a magnetic resonance image reconstruction method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining a data set; training according to the first data, the second data and a preset student learning model to obtain Gaussian distribution and mapping; a Monte Carlo sampler is generated based on the Gaussian distribution and the mapping, and the student learning model is a model obtained based on Bayesian convolutional neural network training; and performing reconstruction according to the first data, the Gaussian distribution, the Monte Carlo sampler and a score matching algorithm to obtain a magnetic resonance image. According to the invention, through the student learning model and the score matching algorithm, the first under-sampling k space data can be rapidly reconstructed, and a high-quality magnetic resonance image is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic resonance image reconstruction, and in particular to a magnetic resonance image reconstruction method, device, computer equipment and storage medium. Background Art

[0002] With the development of artificial intelligence technology, score-based diffusion deep learning models, including denoising score matching models, denoising diffusion probability models and unified stochastic differential equation models, have become an attractive research field.

[0003] However, neither the denoising fractional matching model, the denoising diffusion probability model nor the unified stochastic differential equation model can quickly reconstruct high-quality magnetic resonance images. Summary of the invention

[0004] Based on this, it is necessary to propose a magnetic resonance image reconstruction method, device, computer equipment and storage medium to address the technical problem that the existing technology cannot quickly reconstruct high-quality magnetic resonance images.

[0005] In a first aspect, a magnetic resonance image reconstruction method is provided, the method comprising:

[0006] Acquire a data set, wherein the data set includes each first data and each second data corresponding to the first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data;

[0007] Performing training according to the first data, the second data, and a preset student learning model to obtain a Gaussian distribution and a mapping, and generating a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network;

[0008] Reconstruction is performed according to the first data, the Gaussian distribution, the Monte Carlo sampler, and the score matching algorithm to obtain a magnetic resonance image.

[0009] In a second aspect, a magnetic resonance image reconstruction device is provided, the device comprising:

[0010] an acquisition module, configured to acquire a data set, wherein the data set includes each first data and each second data corresponding to each first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data;

[0011] A generation module, configured to perform training according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and to generate a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network;

[0012] A reconstruction module is used to reconstruct according to the first data, the Gaussian distribution, the Monte Carlo sampler and the score matching algorithm to obtain a magnetic resonance image.

[0013] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned magnetic resonance image reconstruction method when executing the computer program.

[0014] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned magnetic resonance image reconstruction method are implemented.

[0015] The magnetic resonance image reconstruction method proposed in the present invention obtains a data set, wherein the data set includes each first data and each second data corresponding to the first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data. Then, training is performed based on the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and a Monte Carlo sampler is generated based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network. Finally, reconstruction is performed based on the first data, the Gaussian distribution, the Monte Carlo sampler and a score matching algorithm to obtain a magnetic resonance image. The first under-sampled k-space data can be quickly reconstructed through the student learning model and the score matching algorithm to obtain a high-quality magnetic resonance image. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] in:

[0018] Figure 1 A diagram showing an application environment of a magnetic resonance image reconstruction method in one embodiment;

[0019] Figure 2 is a flow chart of a magnetic resonance image reconstruction method in one embodiment;

[0020] Figure 3 is a structural block diagram of a magnetic resonance image reconstruction device in one embodiment;

[0021] Figure 4 is a structural block diagram of a computer device in one embodiment;

[0022] Figure 5 It is a structural block diagram of a computer device in another embodiment. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] The magnetic resonance image reconstruction method provided by the embodiment of the present invention can be applied in Figure 1 In an application environment, the client 110 communicates with the server 120 through a network. The server 120 can receive the acquisition data set proposed in this embodiment through the client 110, wherein the data set includes each first data and each second data corresponding to each first data, the first data is the first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data, then the first data, the second data and the preset student learning model are trained to obtain Gaussian distribution and mapping, and the server 120 generates a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network, and finally the server 120 reconstructs the magnetic resonance image according to the first data, the Gaussian distribution, the Monte Carlo sampler and the score matching algorithm, and can quickly reconstruct the first under-sampled k-space data through the student learning model and the score matching algorithm to obtain a high-quality magnetic resonance image. Among them, the client 110 can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server 120 can be implemented with an independent server or a server cluster consisting of multiple servers. The present invention is described in detail below through specific embodiments.

[0025] See also Figure 2 As shown, Figure 2A flowchart of a magnetic resonance image reconstruction method provided by an embodiment of the present invention includes the following steps:

[0026] Step S101: Acquire a data set, wherein the data set includes each first data and each second data corresponding to the first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data;

[0027] As an example, construct a data set where y i represents the undersampled k-space data, y′ i Represents y i The under-sampled data is obtained by under-sampling, where N is the maximum value of i, for example, N is 100.

[0028] Step S102: training according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and generating a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network;

[0029] It should be noted that, in one implementation, the first data is represented by y, the second data is represented by y′, and there is a mapping f for the first data y θ , maps y to the full k-space data, θ represents the mapping f θ According to the interpolation and translation invariance, the mapping f θ The following equation is satisfied:

[0030] f θ (y)=k+n1,f θ (y′)=y+n2 (1)

[0031] k represents the full k-space data, n1 and n2 represent the disturbance noise. If the mapping f θ It is known that from formula (1), according to the Bayesian formula, p(x|y,θ) can be inferred, that is, the probability distribution of x under the conditions of parameters θ and y, where x represents the full-sampled MRI image. Then, the probability distribution of the full-sampled MRI image can be inferred again by the Bayesian formula:

[0032] p(x)=∫p(x|y,θ)p(y)p(θ|D0)dydθ (2)

[0033] Among them, p(θ|D0) represents the probability distribution of parameter θ under the condition of data set D0. Therefore, the focus of the student learning model is to learn p(θ|D0) and the interpolation mapping f θFurthermore, in one embodiment, the step of training according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping includes:

[0034] Step S1021, performing model training on the student learning model according to the first data, the second data and the preset KL divergence to obtain a trained student learning model;

[0035] In this embodiment, the parameters in the student learning model can be updated by minimizing the preset KL divergence to obtain a trained student learning model.

[0036] Furthermore, in one embodiment, the KL divergence is expressed as follows:

[0037]

[0038] Wherein, θ represents the parameter of the mapping, D0 represents the data set, p(θ|D0) represents the probability distribution of the parameter θ under the condition of D0, and μ θ represents the mean under the parameter θ, the σ θ The table represents the standard deviation of the parameter θ, the q(θ|μ θ ,σ θ ) represents a Gaussian distribution.

[0039] It should be noted that according to q(θ|μ θ ,σ θ ) can get a parameter μ θ ,σ θ The Monte Carlo sampler can be denoted as sampler μ θ ,σ θ .

[0040] Furthermore, in one embodiment, the Monte Carlo sampler is expressed as follows:

[0041]

[0042] Wherein, y represents the first data, S represents coil sensitivity, F represents Fourier transform, and θ i is based on the θ ,σ θ ) is obtained by sampling, wherein i represents the number of samples, is the output of the Monte Carlo sampler.

[0043] Step S1022, input the first data and the second data into the trained student learning model to obtain Gaussian distribution and mapping.

[0044] Step S103: Reconstruct according to the first data, the Gaussian distribution, the Monte Carlo sampler and the score matching algorithm to obtain a magnetic resonance image.

[0045] Specifically, determine whether the Monte Carlo sampler meets the preset end condition; if so, determine the magnetic resonance image according to the Gaussian distribution, the first data and the Monte Carlo sampler. For example, sample the Gaussian distribution to obtain various sampling results; input the sampling results and the first data into the Monte Carlo sampler to obtain the target data output by the Monte Carlo sampler; calculate the average value of each target data to obtain the magnetic resonance image. It should be noted that the preset end condition is met when the Monte Carlo sampler reaches equilibrium, or the preset end condition is met when the number of times the Monte Carlo sampler updates the parameters reaches the preset number.

[0046] If the Monte Carlo sampler does not meet the preset termination condition, the Gaussian noise and the disturbance noise scale are obtained, and the score matching algorithm (Score-Matching) is used to train the score function according to the first data, the Gaussian noise, and the disturbance noise scale to obtain a trained score function. For example, the score matching algorithm is used to make the score function approximate the prior distribution on the disturbance data to obtain a trained score function, wherein the disturbance data is the sum of the third data and the fourth data, the third data is obtained according to the Gaussian distribution, the first data and the Monte Carlo sampler, and the fourth data is the product of the Gaussian noise and the disturbance noise scale. It should also be noted that the Gaussian distribution is sampled to obtain various sampling results; the sampling results and the first data are input into the Monte Carlo sampler to obtain the target data output by the Monte Carlo sampler, which can be used as the third data.

[0047] Next, according to the trained score function, Langevin MCMC sampling is performed to obtain sampling data, the sampling data is used as the second data, and the step of training according to the first data, the second data and the preset student learning model to obtain Gaussian distribution and mapping is returned to execute. The Langevin MCMC sampling is expressed as follows:

[0048]

[0049] Wherein, the y represents the first data, S φ refers to the trained score function, φ is obtained by training on the perturbation data through the score matching method, A represents the magnetic resonance signal encoding acquisition matrix, γ is the data noise scale collected by the magnetic resonance system, η i is the step size, ε iis the disturbance noise scale, To initialize Gaussian noise z k To preset Gaussian noise, i and k are the preset numbers used for iterative loops. represents the k-th sampling data, Represents the k+1th sampling data.

[0050] The present embodiment proposes acquiring a data set, wherein the data set includes each first data and each second data corresponding to the first data, wherein the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data. Then, training is performed according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and based on the Gaussian distribution and the mapping, a Monte Carlo sampler is generated, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network. Finally, reconstruction is performed according to the first data, the Gaussian distribution, the Monte Carlo sampler and a score matching algorithm to obtain a magnetic resonance image. The first under-sampled k-space data can be quickly reconstructed through the student learning model and the score matching algorithm to obtain a high-quality magnetic resonance image.

[0051] See also Figure 3 As shown, in one embodiment, a magnetic resonance image reconstruction device is provided, the device comprising: an acquisition module 10, used to acquire a data set, wherein the data set comprises each first data and each second data corresponding to each first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data;

[0052] A generation module 20 is used to perform training according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and to generate a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network;

[0053] The reconstruction module 30 is used to reconstruct according to the first data, the Gaussian distribution, the Monte Carlo sampler and the score matching algorithm to obtain a magnetic resonance image.

[0054] Furthermore, in one embodiment, the generating module 20 is used to:

[0055] Performing model training on the student learning model according to the first data, the second data and a preset KL divergence to obtain a trained student learning model;

[0056] The first data and the second data are input into a trained student learning model to obtain Gaussian distribution and mapping.

[0057] The KL divergence is expressed as follows:

[0058]

[0059] Wherein, θ represents the parameter of the mapping, D0 represents the data set, p(θ|D0) represents the probability distribution of the parameter θ under the condition of D0, and μ θ represents the mean under the parameter θ, the σ θ represents the standard deviation under the parameter θ, and the q(θ|μ θ ,σ θ ) represents the Gaussian distribution.

[0060] The Monte Carlo sampler is expressed as follows:

[0061]

[0062] Wherein, y represents the first data, S represents coil sensitivity, F represents Fourier transform, and θ i is based on the θ ,σ θ ) is obtained by sampling, wherein i represents the number of samples, is the output of the Monte Carlo sampler.

[0063] Furthermore, in one embodiment, the reconstruction module 30 is used to:

[0064] Determining whether the Monte Carlo sampler satisfies a preset termination condition;

[0065] If satisfied, determining a magnetic resonance image according to the Gaussian distribution, the first data and the Monte Carlo sampler;

[0066] If not satisfied, then obtain Gaussian noise and disturbance noise scale;

[0067] Using a score matching algorithm, training a score function according to the first data, the Gaussian noise, and the disturbance noise scale to obtain a trained score function;

[0068] According to the trained score function, Langevin MCMC sampling is performed to obtain sampling data;

[0069] The sampled data is used as the second data, and the step of performing training according to the first data, the second data and the preset student learning model to obtain Gaussian distribution and mapping is returned.

[0070] The Langevin MCMC sampling is expressed as follows:

[0071]

[0072] Wherein, the y represents the first data, S φ refers to the trained score function, φ is obtained by training on the perturbation data through the score matching method, A represents the magnetic resonance signal encoding acquisition matrix, γ is the data noise scale collected by the magnetic resonance system, η i is the step size, ε i is the disturbance noise scale, To initialize Gaussian noise z k To preset Gaussian noise, i and k are the preset numbers used for iterative loops. represents the k-th sampling data, Represents the k+1th sampling data.

[0073] Furthermore, in one embodiment, the reconstruction module 30 is used to:

[0074] Sampling the Gaussian distribution to obtain various sampling results;

[0075] Input the sampling result and the first data into the Monte Carlo sampler to obtain the target data output by the Monte Carlo sampler;

[0076] The target data are averaged to obtain a magnetic resonance image.

[0077] Furthermore, in one embodiment, the reconstruction module 30 is used to:

[0078] A score matching algorithm is used to make the score function approximate the prior distribution on the perturbation data to obtain a trained score function, wherein the perturbation data is the sum of the third data and the fourth data, the third data is obtained according to the Gaussian distribution, the first data and the Monte Carlo sampler, and the fourth data is the product of the Gaussian noise and the perturbation noise scale.

[0079] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, it realizes the functions or steps of a magnetic resonance image reconstruction method server side.

[0080] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, it implements the functions or steps of a client side of a magnetic resonance image reconstruction method.

[0081] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are implemented:

[0082] Acquire a data set, wherein the data set includes each first data and each second data corresponding to the first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data;

[0083] Performing training according to the first data, the second data, and a preset student learning model to obtain a Gaussian distribution and a mapping, and generating a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network;

[0084] Reconstruction is performed according to the first data, the Gaussian distribution, the Monte Carlo sampler, and the score matching algorithm to obtain a magnetic resonance image.

[0085] The present embodiment proposes acquiring a data set, wherein the data set includes each first data and each second data corresponding to the first data, wherein the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data. Then, training is performed according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and based on the Gaussian distribution and the mapping, a Monte Carlo sampler is generated, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network. Finally, reconstruction is performed according to the first data, the Gaussian distribution, the Monte Carlo sampler and a score matching algorithm to obtain a magnetic resonance image. The first under-sampled k-space data can be quickly reconstructed through the student learning model and the score matching algorithm to obtain a high-quality magnetic resonance image.

[0086] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0087] Acquire a data set, wherein the data set includes each first data and each second data corresponding to the first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data;

[0088] Performing training according to the first data, the second data, and a preset student learning model to obtain a Gaussian distribution and a mapping, and generating a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network;

[0089] Reconstruction is performed according to the first data, the Gaussian distribution, the Monte Carlo sampler, and the score matching algorithm to obtain a magnetic resonance image.

[0090] The present embodiment proposes acquiring a data set, wherein the data set includes each first data and each second data corresponding to the first data, wherein the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data. Then, training is performed according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and based on the Gaussian distribution and the mapping, a Monte Carlo sampler is generated, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network. Finally, reconstruction is performed according to the first data, the Gaussian distribution, the Monte Carlo sampler and a score matching algorithm to obtain a magnetic resonance image. The first under-sampled k-space data can be quickly reconstructed through the student learning model and the score matching algorithm to obtain a high-quality magnetic resonance image.

[0091] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0092] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0093] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0094] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A magnetic resonance image reconstruction method, characterized in that: The magnetic resonance image reconstruction method comprises: Acquire a data set, wherein the data set includes each first data and each second data corresponding to the first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data; Performing training according to the first data, the second data, and a preset student learning model to obtain a Gaussian distribution and a mapping, and generating a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network; Reconstruction is performed according to the first data, the Gaussian distribution, the Monte Carlo sampler, and the score matching algorithm to obtain a magnetic resonance image.

2. The magnetic resonance image reconstruction method according to claim 1, characterized in that: The step of training according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping includes: Performing model training on the student learning model according to the first data, the second data and a preset KL divergence to obtain a trained student learning model; The first data and the second data are input into a trained student learning model to obtain Gaussian distribution and mapping.

3. The magnetic resonance image reconstruction method according to claim 2, characterized in that: The KL divergence is expressed as follows: Wherein, θ represents the parameter of the mapping, D0 represents the data set, p(θ|D0) represents the probability distribution of the parameter θ under the condition of D0, and μ θ represents the mean under the parameter θ, the σ θ represents the standard deviation under the parameter θ, and the q(θ|μ θ ,σ θ ) represents the Gaussian distribution.

4. The magnetic resonance image reconstruction method according to claim 3, characterized in that: The Monte Carlo sampler is expressed as follows: Wherein, y represents the first data, S represents coil sensitivity, F represents Fourier transform, and θ i is based on the θ ,σ θ ) is obtained by sampling, wherein i represents the number of samples, is the output of the Monte Carlo sampler.

5. The magnetic resonance image reconstruction method according to claim 1, characterized in that: The step of reconstructing according to the first data, the Gaussian distribution, the Monte Carlo sampler and the score matching algorithm to obtain a magnetic resonance image comprises: Determining whether the Monte Carlo sampler satisfies a preset termination condition; If satisfied, determining a magnetic resonance image according to the Gaussian distribution, the first data and the Monte Carlo sampler; If not satisfied, then obtain Gaussian noise and disturbance noise scale; Using a score matching algorithm, training a score function according to the first data, the Gaussian noise, and the disturbance noise scale to obtain a trained score function; According to the trained score function, Langevin MCMC sampling is performed to obtain sampling data; The sampled data is used as the second data, and the step of performing training according to the first data, the second data and the preset student learning model to obtain Gaussian distribution and mapping is returned. The Langevin MCMC sampling is expressed as follows: Wherein, the y represents the first data, S φ refers to the trained score function, φ is obtained by training on the perturbation data through the score matching method, A represents the magnetic resonance signal encoding acquisition matrix, γ is the data noise scale collected by the magnetic resonance system, η i is the step size, ε i is the disturbance noise scale, To initialize Gaussian noise z k is the preset Gaussian noise, i and k are the preset numbers used for iterative loops, represents the k-th sampling data, Represents the k+1th sampling data.

6. The magnetic resonance image reconstruction method according to claim 5, characterized in that: The step of determining a magnetic resonance image according to the Gaussian distribution, the first data and the Monte Carlo sampler comprises: Sampling the Gaussian distribution to obtain various sampling results; Input the sampling result and the first data into the Monte Carlo sampler to obtain the target data output by the Monte Carlo sampler; The target data are averaged to obtain a magnetic resonance image.

7. The magnetic resonance image reconstruction method according to claim 5, characterized in that: The step of using a score matching algorithm to train a score function according to the first data, the probability distribution of the first data, the Gaussian noise, and the disturbance noise scale to obtain a trained score function includes: A score matching algorithm is used to make the score function approximate the prior distribution on the perturbation data to obtain a trained score function, wherein the perturbation data is the sum of the third data and the fourth data, the third data is obtained according to the Gaussian distribution, the first data and the Monte Carlo sampler, and the fourth data is the product of the Gaussian noise and the perturbation noise scale.

8. A magnetic resonance image reconstruction device, characterized in that: The magnetic resonance image reconstruction device comprises: an acquisition module, configured to acquire a data set, wherein the data set includes each first data and each second data corresponding to each first data, the first data is first under-sampled k-space data, and the second data is obtained by under-sampling the first under-sampled k-space data; A generation module, configured to perform training according to the first data, the second data and a preset student learning model to obtain a Gaussian distribution and a mapping, and to generate a Monte Carlo sampler based on the Gaussian distribution and the mapping, wherein the student learning model is a model obtained by training a Bayesian convolutional neural network; A reconstruction module is used to reconstruct according to the first data, the Gaussian distribution, the Monte Carlo sampler and the score matching algorithm to obtain a magnetic resonance image.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the magnetic resonance image reconstruction method according to any one of claims 1 to 7 are implemented.

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

Images