WO2022193379A1

WO2022193379A1 - Image reconstruction model generation method and apparatus, image reconstruction method and apparatus, device, and medium

Info

Publication number: WO2022193379A1
Application number: PCT/CN2021/085552
Authority: WO
Inventors: 梁栋; 朱燕杰; 黄文麒; 柯子文; 崔卓须; 刘新; 郑海荣
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2021-03-17
Filing date: 2021-04-06
Publication date: 2022-09-22
Also published as: CN115115723A

Abstract

An image reconstruction model generation method and apparatus, an image reconstruction method and apparatus, a device, and a medium. The image reconstruction model generation method comprises: acquiring fully sampled k-space data of a dynamic magnetic resonance image sequence, and obtaining undersampled k-space data corresponding to the fully sampled k-space data on the basis of a preset undersampling model (S110); inputting the undersampled k-space data to an image reconstruction model established according to a sub-problem obtained by solving a magnetic resonance image reconstruction problem of the fully sampled k-space data under constraints having a low-rank property and a sparse property, to train the image reconstruction model (S120); and when a mean square error between a reconstructed image generated by the image reconstruction model and a reconstructed image corresponding to the fully sampled k-space data satisfies a preset condition, completing model training to obtain a target image reconstruction model (S130).

Description

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

This application claims the priority of the Chinese patent application with application number 202110287100.7 filed with the China Patent Office on March 17, 2021, the entire contents of which are incorporated herein by reference.

technical field

The embodiments of the present application relate to the technical field of medical image processing, for example, to a method and apparatus for generating an image reconstruction model, an image reconstruction method and apparatus, a device, and a medium.

Background technique

Magnetic resonance cardiac cine imaging is a non-invasive imaging technique that can be used to assess cardiac function, abnormal ventricular wall motion, etc., and provide rich information for clinical diagnosis of the heart. However, due to the physical characteristics of magnetic resonance, the hardware for realizing magnetic resonance and the limitation of the duration of the cardiac motion cycle, the temporal and spatial resolution of magnetic resonance cardiac cine imaging is often limited, and it is impossible to accurately assess some cardiac diseases, such as arrhythmia, etc. Cardiac function. Therefore, it is necessary to use fast imaging methods to improve the temporal and spatial resolution of magnetic resonance cardiac cine imaging under the premise of ensuring imaging quality.

Commonly used methods to accelerate magnetic resonance cardiac cine imaging include Parallel Imaging (PI), Compressed Sensing (CS) technology, low-rank matrix factorization and deep learning methods.

However, traditional parallel imaging or compressed sensing technologies do not utilize big data priors, and the parameters of this iterative optimization method are difficult to select and time-consuming. Although the neural network method based on deep learning can avoid the iterative solution step and speed up the reconstruction time, the deep learning neural network method only relies on the sparse prior of big data, which limits the improvement of image reconstruction effect.

SUMMARY OF THE INVENTION

The embodiments of the present application provide an image reconstruction model generation and device, an image reconstruction method, device, equipment, and medium, so as to realize the speed of image reconstruction while making full use of the low-rank and sparse characteristics of sampled data to create an image Rebuild the model to improve the quality of the reconstructed image.

An image reconstruction model generation method is provided, including:

Obtain full-sampled K-space data of a dynamic magnetic resonance image sequence, and obtain under-sampled K-space data corresponding to the full-sampled K-space data based on a preset under-sampling model; input the under-sampled K-space data to, according to the The sub-problems obtained by solving the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics of the fully sampled K-space data, establish an image reconstruction model, and train the image reconstruction model; in the image reconstruction model For the generated reconstructed image, when the mean square error between the reconstructed images corresponding to the fully sampled K-space data satisfies the preset condition, the training of the image reconstruction model is completed, and the trained image reconstruction model is used as the target image. Rebuild the model.

In an implementation manner, an image reconstruction model is established according to a sub-problem obtained by solving the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics for the fully sampled K-space data, including:

According to the temporal periodicity of the motion of the magnetic resonance scanning object, the fully sampled K-space data is modeled to represent the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics; setting the magnetic resonance image reconstruction problem Auxiliary variables representing the model, and determining a penalty function representing the model based on the auxiliary variables; determining three sub-problems representing the model according to the penalty function; solving the three sub-problems respectively, and analyzing the solution results Networking obtains the image reconstruction model.

In an implementation manner, the three sub-problems are solved separately, and the solution results are networked to obtain the image reconstruction model, including:

Rewrite the three sub-problems respectively; use the approach gradient method to solve the rewritten three sub-problems respectively, and obtain the iterative representation of the solution of each sub-problem; network the process of the iterative representation of the solution of each sub-problem to obtain the image reconstruction model.

In an implementation manner, the rewriting the three sub-problems respectively includes:

The three sub-problems are respectively rewritten according to the result of the expansion of the penalty function at the preset value of the data fidelity item.

In an implementation manner, the image reconstruction model includes a low-rank network module, a sparse network module, and a data consistency module.

An image reconstruction method is also provided, including:

Obtain the under-sampled K-space data of the dynamic magnetic resonance image sequence obtained based on the preset under-sampling model; input the under-sampled K-space data into the target image reconstruction model obtained by the above-mentioned image reconstruction model generation method, and obtain the under-sampled K-space data. Sample the reconstructed image corresponding to the K-space data.

Also provided is an image reconstruction model generation device, comprising:

The data preprocessing module is configured to obtain the full-sampled K-space data of the dynamic magnetic resonance image sequence, and obtain the under-sampled K-space data corresponding to the full-sampled K-space data based on the preset under-sampling model; the data input module is configured to The under-sampled K-space data is input to the sub-problem obtained by solving the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics based on the fully-sampled K-space data, and the established image reconstruction model is used for all the sub-problems. The image reconstruction model is trained; the model generation module is configured to complete the process when the mean square error between the reconstructed images generated by the image reconstruction model and the reconstructed images corresponding to the fully sampled K-space data satisfies a preset condition. For the training of the image reconstruction model, the trained image reconstruction model is used as the target image reconstruction model.

Also provided is an image reconstruction device, comprising:

The data acquisition module is configured to acquire the under-sampled K-space data of the dynamic magnetic resonance image sequence obtained based on the preset under-sampling model; the image reconstruction module is configured to input the under-sampled K-space data into the data generated by the above-mentioned image reconstruction model. In the target image reconstruction model obtained by the method, the reconstructed image corresponding to the undersampled K-space data is obtained.

Also provided is a computer device comprising:

one or more processors; a memory arranged to store one or more programs; when said one or more programs are executed by said one or more processors, causing said one or more processors to implement the above-mentioned images Reconstruction model generation method or image reconstruction method.

A computer-readable storage medium is also provided, which stores a computer program, and when the program is executed by a processor, realizes the above-mentioned image reconstruction model generation method or image reconstruction method.

Description of drawings

1 is a flowchart of a method for generating an image reconstruction model provided in Embodiment 1 of the present application;

2 is a schematic diagram of a network structure of an image reconstruction network provided in Embodiment 1 of the present application;

3 is a flowchart of an image reconstruction method provided in Embodiment 2 of the present application;

4 is a schematic structural diagram of an apparatus for generating an image reconstruction model according to Embodiment 3 of the present application;

FIG. 5 is a schematic structural diagram of an image reconstruction apparatus according to Embodiment 4 of the present application;

FIG. 6 is a schematic structural diagram of a computer device according to Embodiment 5 of the present application.

Detailed ways

The present application will be described below with reference to the accompanying drawings and embodiments.

Example 1

FIG. 1 is a flowchart of a method for generating an image reconstruction model according to Embodiment 1 of the present application. This embodiment can be applied to a situation where a fully sampled image of a magnetic resonance dynamic image is used as a sample to train an image reconstruction model. The method may be executed by an image reconstruction model generating apparatus, which may be implemented in software and/or hardware, and integrated into an electronic device with an application development function.

As shown in Figure 1, the image reconstruction model generation method includes the following steps:

S110. Acquire full-sampled K-space data of a dynamic magnetic resonance image sequence, and obtain under-sampled K-space data corresponding to the full-sampled K-space data based on a preset under-sampling model.

The full-sampled K-space data of the dynamic magnetic resonance image sequence is pre-collected sample data, and a high-resolution magnetic resonance dynamic image can be reconstructed according to the full-sampled K-space data of the dynamic magnetic resonance image sequence. For example, the dynamic magnetic resonance image sequence may be each frame of the magnetic resonance cardiac cine imaging. The preset undersampling model can be determined by the preset undersampling operator, the Fourier transform operator and the coil sensitivity parameter according to the requirement of the sampling acceleration multiple.

The preset undersampling model can be expressed as y=MFCX(1); where M is the undersampling operator, F is the Fourier transform operator, C is the coil sensitivity map, and X is the fully sampled Each frame of image in the dynamic magnetic resonance image sequence is stretched into a one-dimensional vector, and multiple one-dimensional vectors are spliced together in the order of image frames as columns to form an image matrix, where y is the undersampled K-space data. In step S110, first, in the fully sampled K-space data of the dynamic magnetic resonance image sequence, the multi-column data of the fully-sampled K-space data corresponding to each frame of image is spliced into a data with only one column in the order of the columns; then , splicing the column data corresponding to all image frames in the order of the image frames to obtain X. C can be estimated by algorithms such as ESPIRiT or Walsh, F is the operator of the Fourier transform, and M is a vector composed of 0 and 1, which determines which points are collected and which points are not collected, and the value of the points not collected is 0.

S120. Input the under-sampled K-space data to the sub-problem obtained by solving the magnetic resonance image reconstruction problem for the fully-sampled K-space data under the constraints of low-rank characteristics and sparse characteristics, and establish an image reconstruction model , to train the model.

The image reconstruction model is an image reconstruction model that is modeled and trained according to the sparse prior and low-rank prior characteristics of dynamic magnetic resonance data. The construction process of the image reconstruction model is as follows:

First, a low-rank matrix is constructed for the matrix X, and the matrix X can be decomposed into a background component L and a dynamic component S. Because there is a lot of related information between multiple frames in the background component L, the L matrix has a low rank. Meanwhile, in cardiac cine imaging, the beating area of the heart is small, and after the background component L is removed, the remaining S component itself has sparsity. In one embodiment, the dynamic component S may also be sparsely transformed to improve its sparsity. The stronger the S sparsity, the better the reconstructed image. The coefficient transformation can be realized by methods such as Fourier transform, wavelet transform or discrete cosine transform. For example, a one-dimensional Fourier transform is performed on S in the time direction, and the transformed matrix is more sparse than S. L, S, and X are data matrices of the same dimension.

According to the periodicity of the heart beat in the time direction, the MRI reconstruction problem can be written as:

Among them, λ _L and λ _S are the regularization factors for low-quality constraints and sparse constraints, respectively, A is the encoding matrix, y is the undermined k-space data, and D is the sparse transformation matrix. || ||* represents the matrix kernel norm, that is, the sum of the non-zero singular values of the matrix is calculated. In mathematics, the sum norm is often used to approximate the matrix rank.

In order to solve the MRI reconstruction problem, an auxiliary variable X is introduced in this embodiment to represent the accumulated sum of L and S. Then the above optimization problem can be written as:

The introduction of auxiliary variable X can enable L and S to perform inexact search in the initial iterative step after the model is successfully constructed, so that the algorithm can converge to the optimal L and S more quickly. Fast convergence is very important for iterative algorithm networking, because during networking, a small number of iterative network modules need to be used to replace the original more iterative steps. To understand the equation optimization problem in Equation (3), determine the penalty function for Equation (3):

The penalty function can be solved by an alternate minimization algorithm, and the following three sub-problems are obtained. Finally, the target image reconstruction model can be obtained by networking the following three sub-problems.

In one embodiment, since it is time-consuming to solve (A ^* A+ρI) ^-1 in the X sub-problem, an inexact way can be used to solve the X sub-problem. Put the penalty function J in the data fidelity term

of

Linear expansion at X to get the X suboptimization problem:

in,

represents the result after J linearization,

In the representation of , the angle brackets indicate the inner product of the two matrices within the angle brackets. It is equivalent to the 0th-order term in the Taylor expansion of the function, but here is a matrix, so it must be written in the form of an inner product.

Then the subproblem in equation (5) can be rewritten as:

Each subproblem can be solved by the approaching gradient method, based on

The above subproblem can be written as the following iterative steps:

in,

is the singular value soft threshold operator.

in,

is the soft threshold operator, X=UΣV ^* is the singular value decomposition of X, Σ is the singular value matrix, U and V are the left and right matrices respectively, * represents the conjugate transpose;

is the approximation operator related to the sparse transform D, if D is an orthogonal transform,

It is a soft threshold operator. If D is a non-orthogonal transformation, the analytical solution cannot be directly written in the form of P; γ is the update step size, γ=1/(1+ηρ). In each iteration step,

By reducing the singular values less than the threshold λ _L , L has a low rank; in the X iteration step in formula (9),

is the residual in k-space between the reconstructed image and the input k-space data, which is used for data consistency correction.

Although the form of the iterative solution can be as shown in Equation 9, there are still three problems to be solved: first, the iteration takes a long time to converge, and the reconstruction speed is slow, especially after the calculation of singular value decomposition is added to the iterative process; third Second, the regularization parameter is difficult to adjust, and it takes a lot of time to try. Especially, L+S has two components. If λ _L is too large compared to λ _S , the S part of the reconstructed image will not be sparse enough and contain a large number of static components. If λ L _L is too small compared to λ _S , and the L part of the reconstructed image will contain a lot of dynamic information, which is contrary to the model; third, only when the sparse transformation D is orthogonal,

The explicit solution can only be written, and cannot be obtained when the sparse transformation D is not orthogonal

However, the orthogonal sparse transformation will greatly limit the selection range of sparse transformation, which may lead to limited reconstruction performance.

Therefore, the iterative process in formula (9) is networked and expanded into a low-rank + sparse network (L+S-Net). L+S-Net contains multiple network blocks, and each network block contains the Three network modules, namely the low-rank module L ^k , the sparse module S ^k , and the data consistency module X ^k :

The low-rank module L ^k+1 performs a learnable singular value soft threshold operation (Learned Singular Value Threshold, LSVT) on X _k -S _k , which is the same as that shown in formula (10), but the threshold is replaced by a learnable value Variables:

H _β (X)=UΛ _T(β,Σ) (Σ)V ^* (12)

Among them, the threshold value of the soft threshold operator is:

T(β,Σ)=β·σ ₁ (13)

where σ1 is the largest singular value and β is _a learnable threshold factor. In the sparse module in Eq. (11), a learnable convolutional neural network C is used to replace the approximation operator

In each network block, the threshold factor β, the convolutional neural network C, and the update step size γ are all independently learnable. During network training, X ₀ and L ₀ are initialized to zero-padded corresponding to the under-mined K-space data y In the figure, S ₀ is initialized with all 0s, and these learnable parameters are continuously optimized, and finally high-quality reconstruction results are obtained to generate an image reconstruction model.

S130. When the mean square error between the reconstructed images generated by the image reconstruction model and the reconstructed images corresponding to the fully sampled K-space data satisfies a preset condition, complete model training to obtain a target image reconstruction model.

The reconstructed image corresponding to the fully sampled K-space data is a standard image with high reconstruction quality. The process of image reconstruction model training is the process of learning and updating the parameters in the model. The optimization of parameters makes the image output by the model closer and closer to the standard image. . When the mean square error between the reconstructed images generated by the image reconstruction model and the reconstructed images corresponding to the fully sampled K-space data satisfies the preset condition, that is, when the model converges, the target image reconstruction model can be obtained, and the model training process can be completed.

In the technical solution of this embodiment, by processing the obtained full-sampled K-space data samples of the dynamic magnetic resonance image sequence according to a preset under-sampling model, the under-sampled K-space data corresponding to the full-sampled K-space data is obtained; then, the The undersampled K-space data is input to, and the image reconstruction model is established according to the sub-problem obtained by solving the magnetic resonance image reconstruction problem under the constraints of low-rank and sparse characteristics on the full-sampled K-space data, and the model is trained; when When the mean square error between the reconstructed image generated by the image reconstruction model and the reconstructed image corresponding to the fully sampled K-space data satisfies the preset condition, the model training is completed, and the target image reconstruction model is obtained. It solves the problem that the image reconstruction time in the related art is long or the characteristics of the sampled data cannot be fully utilized for image reconstruction, and the image reconstruction model can be established by making full use of the low-rank and sparse characteristics of the sampled data while speeding up the image reconstruction speed. to improve the reconstructed image quality.

Embodiment 2

FIG. 3 is a flowchart of an image reconstruction method according to Embodiment 2 of the present application, and this embodiment is applicable to the case of medical image reconstruction. The method may be performed by an image reconstruction apparatus, and the apparatus may be implemented in software and/or hardware, and integrated into a computer device with an application development function.

As shown in Figure 3, the image reconstruction method includes the following steps:

S210. Acquire under-sampling K-space data of a dynamic magnetic resonance image sequence obtained based on a preset under-sampling model.

During the magnetic resonance scanning sampling process, the sampling model can be set according to the requirement of the sampling speed, and the preset undersampling model is determined by the preset undersampling operator, the Fourier transform operator and the coil sensitivity parameter. The preset undersampling operator determines how fast the samples are doubled.

The preset undersampling model can be expressed as y=MFCX; where M is the undersampling operator, F is the Fourier transform operator, C is the coil sensitivity maps, and X is the dynamic magnetic resonance that will be fully sampled Each frame of image in the image sequence is stretched into a one-dimensional vector, and multiple one-dimensional vectors are spliced together in the order of image frames as columns to form an image matrix, where y is the undersampled K-space data. In step S210, first, in the full-sampled K-space data of the dynamic magnetic resonance image sequence, the multi-column data of the full-sampled K-space data corresponding to each frame of image is spliced into a data with only one column according to the sequence of columns; then, The column data corresponding to all image frames are spliced in the order of the image frames to obtain X. C can be estimated by algorithms such as ESPIRiT or Walsh, F is the operator of the Fourier transform, and M is a vector composed of 0 and 1, which determines which points are collected and which points are not collected, and the value of the points not collected is 0.

During magnetic resonance scanning, under-sampled K-space data can be obtained through the above sampling model. For example, acquiring magnetic resonance sampling data of cardiac dynamics.

S220. Input the undersampled K-space data into the target image reconstruction model obtained by the image reconstruction model generation method described in any one of the embodiments, to obtain a reconstructed image corresponding to the undersampled K-space data.

The target image reconstruction model is a model for image reconstruction under the constraints of low-rank characteristics and sparse characteristics, which can reconstruct high-quality images.

In the technical solution of this embodiment, a reconstructed image is obtained by sampling according to a preset sampling model and inputting the sampling data into the trained image reconstruction model; it solves the problem that the image reconstruction time in the related art is long or the sampling data cannot be fully utilized. The problem of image reconstruction based on the characteristics of the image, realizes that while speeding up the image reconstruction speed, it can make full use of the low-rank and sparse characteristics of the sampled data to establish an image reconstruction model to improve the quality of the reconstructed image.

Embodiment 3

FIG. 4 is a schematic structural diagram of an image reconstruction model generating apparatus according to Embodiment 3 of the present application. This embodiment can be applied to a situation where an image reconstruction model is trained using a fully sampled image of a magnetic resonance dynamic image as a sample.

As shown in FIG. 4 , the image reconstruction model generation apparatus includes a data preprocessing module 310 , a data input module 320 and a model generation module 330 .

The data preprocessing module 310 is configured to obtain full-sampled K-space data of the dynamic magnetic resonance image sequence, and obtain under-sampled K-space data corresponding to the full-sampled K-space data based on a preset under-sampling model; the data input module 320 is configured to set In order to input the under-sampled K-space data to the sub-problem obtained by solving the magnetic resonance image reconstruction problem under the constraints of the low-rank characteristic and sparse characteristic for the fully-sampled K-space data, the established image reconstruction model, The model is trained; the model generation module 330 is configured to complete the model training when the mean square error between the reconstructed images generated by the image reconstruction model and the reconstructed images corresponding to the fully sampled K-space data satisfies a preset condition, Get the target image reconstruction model.

Optionally, the image reconstruction model generation device further includes a model construction module, which is configured as a sub-problem obtained by solving the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics based on the fully sampled K-space data, Build an image reconstruction model.

Optionally, the model construction module is set to:

According to the temporal periodicity of the motion of the magnetic resonance scanning object, the fully sampled K-space data is modeled to represent the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics; setting the magnetic resonance image reconstruction problem Auxiliary variables representing the model, and determining a penalty function representing the model based on the auxiliary variables; determining three sub-problems representing the model according to the penalty function; solving the three sub-problems respectively, and obtaining a networked the image reconstruction model.

Optionally, the model construction module can also be set to:

The three sub-problems are optimized and rewritten respectively; the optimized and rewritten three sub-problems are solved respectively by the approaching gradient method, and an iterative representation of the solution of each sub-problem is obtained; the process of the iterative representation is networked to obtain the Image reconstruction model.

Optionally, the model construction module may also be configured to optimize and rewrite the three sub-problems in the following ways, including:

The three sub-problems are optimized and rewritten respectively according to the result of the expansion of the penalty function at the preset value of the data fidelity item.

Optionally, the image reconstruction model includes a low-rank network module, a sparse network module and a data consistency network module.

The image reconstruction model generation apparatus provided by the embodiment of the present application can execute the image reconstruction model generation method provided by any embodiment of the present application, and has functional modules and effects corresponding to the execution method.

Embodiment 4

FIG. 5 is a schematic structural diagram of an image reconstruction apparatus according to Embodiment 4 of the present application, and this embodiment is applicable to situations where this embodiment is applicable.

As shown in FIG. 5 , the image reconstruction apparatus includes a data acquisition module 410 and an image reconstruction module 420 .

The data acquisition module 410 is configured to acquire the under-sampled k-space data of the dynamic magnetic resonance image sequence obtained based on the preset under-sampling model; the image reconstruction module 420 is configured to input the under-sampled K-space data into the data obtained by any of the embodiments. In the target image reconstruction model obtained by the image reconstruction model generation method, a reconstructed image corresponding to the undersampled K-space data is obtained.

The image reconstruction apparatus provided by the embodiment of the present application can execute the image reconstruction method provided by any embodiment of the present application, and has functional modules and effects corresponding to the execution method.

Embodiment 5

FIG. 6 is a schematic structural diagram of a computer device according to Embodiment 5 of the present application. Figure 6 shows a block diagram of an exemplary computer device 12 suitable for use in implementing embodiments of the present application. The computer device 12 shown in FIG. 6 is only an example, and should not impose any limitations on the functions and scope of use of the embodiments of the present application. The computer device 12 may be any terminal device with computing capability, such as an intelligent controller, a server, a mobile phone and other terminal devices.

As shown in FIG. 6, computer device 12 takes the form of a general-purpose computing device. Components of computer device 12 may include, but are not limited to, one or more processors or processing units 16 , system memory 28 , and a bus 18 connecting various system components including system memory 28 and processing unit 16 .

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

Computer device 12 includes, for example, various computer system readable media. These media can be any available media that can be accessed by computer device 12, including both volatile and nonvolatile media, removable and non-removable media.

System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and/or cache memory 32 . Computer device 12 may include other removable/non-removable, volatile/non-volatile computer system storage media. For example only, storage system 34 may be configured to read and write to non-removable, non-volatile magnetic media (not shown in FIG. 6, commonly referred to as a "hard drive"). Although not shown in Figure 6, a magnetic disk drive configured to read and write to removable non-volatile magnetic disks (eg "floppy disks") and removable non-volatile optical disks (eg Compact Disc Read-Only Memory) may be provided Read-Only Memory, CD-ROM), digital versatile disc read-only memory (Digital Versatile Disc Read-Only Memory, DVD-ROM) or other optical media) optical disk drive for reading and writing. In these cases, each drive may be connected to bus 18 through one or more data media interfaces. System memory 28 may include at least one program product having a set (eg, at least one) of program modules configured to perform the functions of embodiments of the present application.

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

Computer device 12 may also communicate with one or more external devices 14 (eg, keyboard, pointing device, display 24, etc.), may also communicate with one or more devices that enable a user to interact with computer device 12, and/or communicate with Any device (eg, network card, modem, etc.) that enables the computer device 12 to communicate with one or more other computing devices. Such communication may take place through an input/output (I/O) interface 22 . Also, computer device 12 may communicate with one or more networks (eg, Local Area Network (LAN), Wide Area Network (WAN), and/or public networks such as the Internet) through network adapter 20. As shown, network adapter 20 communicates with other modules of computer device 12 via bus 18 . It should be understood that, although not shown in FIG. 6, other hardware and/or software modules may be used in conjunction with computer device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, disk arrays (Redundant Arrays of Independent Disks, RAID) systems, tape drives, and data backup storage systems, etc.

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

Obtain full-sampled K-space data of a dynamic magnetic resonance image sequence, and obtain under-sampled K-space data corresponding to the full-sampled K-space data based on a preset under-sampling model; input the under-sampled K-space data to, according to the The sub-problem obtained by solving the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics of the fully sampled K-space data, establishes an image reconstruction model, and trains the model; when the reconstructed image generated by the image reconstruction model , when the mean square error between the reconstructed images corresponding to the fully sampled K-space data satisfies the preset condition, the model training is completed, and the target image reconstruction model is obtained.

Alternatively, the steps of an image reconstruction method provided by the embodiment of the present invention can also be implemented, and the method includes:

Obtaining the under-sampled K-space data of the dynamic magnetic resonance image sequence obtained based on the preset under-sampling model; inputting the under-sampled K-space data into the target image reconstruction model obtained by the image reconstruction model generation method described in any one of the embodiments , the reconstructed image corresponding to the undersampled K-space data is obtained.

Embodiment 6

The sixth embodiment provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the image reconstruction model generation method provided by any embodiment of the present application, including:

The computer storage medium of the embodiments of the present application may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination of the above. Examples (non-exhaustive list) of computer readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, RAM, ROM, Erasable Programmable Read-Only Memory (Erasable Programmable Read-Only Memory) Memory, EPROM or flash memory), optical fiber, CD-ROM, optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

A computer-readable signal medium may include a propagated data signal in baseband or as part of a carrier wave, with computer-readable program code embodied thereon. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device .

The program code embodied on the computer readable medium may be transmitted by any suitable medium, including but not limited to: wireless, wire, optical fiber cable, radio frequency (RF), etc., or any suitable combination of the above.

Computer program code for carrying out the operations of the present application may be written in one or more programming languages, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional A procedural programming language, such as the "C" language or similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. Where a remote computer is involved, the remote computer may be connected to the user's computer through any kind of network, including a LAN or WAN, or may be connected to an external computer (eg, using an Internet service provider to connect through the Internet).

The above-mentioned multiple modules or multiple steps of the present application can be implemented by a general-purpose computing device, and they can be centralized on a single computing device, or distributed on a network composed of multiple computing devices. implemented by program code executable by a computer device so that they can be stored in a storage device and executed by a computing device, or they can be separately made into a plurality of integrated circuit modules, or a plurality of modules or steps in them can be made into a single integrated circuit modules. As such, the present application is not limited to any particular combination of hardware and software.

Claims

An image reconstruction model generation method, comprising:

Acquiring full-sampled K-space data of a dynamic magnetic resonance image sequence, and obtaining under-sampled K-space data corresponding to the full-sampled K-space data based on a preset under-sampling model;

The under-sampled K-space data is input into the sub-problem obtained by solving the magnetic resonance image reconstruction problem under the constraints of the low-rank characteristic and sparse characteristic for the fully-sampled K-space data, the established image reconstruction model, The image reconstruction model is trained;

In the case where the mean square error between the reconstructed images generated by the image reconstruction model and the reconstructed images corresponding to the fully sampled K-space data satisfies a preset condition, the training of the image reconstruction model is completed, and the trained image The image reconstruction model is used as the target image reconstruction model.
The method according to claim 1, wherein the image reconstruction model is established according to a sub-problem obtained by solving a magnetic resonance image reconstruction problem on the fully sampled K-space data under the constraints of low-rank characteristics and sparse characteristics, comprising: :

According to the temporal periodicity of the motion of the magnetic resonance scanning object, the fully sampled K-space data is modeled to represent the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics;

setting auxiliary variables representing a model of the magnetic resonance image reconstruction problem, and determining a penalty function representing the model based on the auxiliary variables;

determine three sub-problems of the representation model according to the penalty function;

The three sub-problems are solved respectively, and the solution results are networked to obtain the image reconstruction model.
The method according to claim 2, wherein said solving the three sub-problems respectively, and networking the solving results to obtain the image reconstruction model, comprising:

Rewrite the three sub-problems respectively;

The three rewritten sub-problems are solved respectively by the approaching gradient method, and the iterative representation of the solution of each sub-problem is obtained;

The image reconstruction model is obtained by networking the process of iterative representation of the solution to each subproblem.
The method according to claim 3, wherein the rewriting the three sub-problems respectively comprises:

The three sub-problems are respectively rewritten according to the result of the expansion of the penalty function at the preset value of the data fidelity item.
The method of any one of claims 1-4, wherein the image reconstruction model includes a low-rank network module, a sparse network module, and a data-consistent network module.
An image reconstruction method, comprising:

acquiring under-sampling K-space data of the dynamic magnetic resonance image sequence obtained based on the preset under-sampling model;

Inputting the under-sampled K-space data into a target image reconstruction model obtained by the image reconstruction model generation method according to any one of claims 1-5, to obtain a reconstructed image corresponding to the under-sampled K-space data.
An image reconstruction model generation device, comprising:

a data preprocessing module, configured to obtain fully sampled K-space data of the dynamic magnetic resonance image sequence, and obtain under-sampled K-space data corresponding to the fully-sampled K-space data based on a preset under-sampling model;

The data input module is configured to input the under-sampled K-space data to the sub-problems obtained by solving the magnetic resonance image reconstruction problem under the constraints of low-rank characteristics and sparse characteristics for the fully-sampled K-space data, establishes The image reconstruction model, the image reconstruction model is trained;

The model generation module is configured to complete the image reconstruction model when the mean square error between the reconstructed images generated by the image reconstruction model and the reconstructed images corresponding to the fully sampled K-space data satisfies a preset condition. For training, use the trained image reconstruction model as the target image reconstruction model.
An image reconstruction device, comprising:

a data acquisition module, configured to acquire the under-sampled K-space data of the dynamic magnetic resonance image sequence obtained based on the preset under-sampling model;

An image reconstruction module, configured to input the undersampled K-space data into a target image reconstruction model obtained by the image reconstruction model generation method according to any one of claims 1-5, to obtain the undersampled K-space data corresponding reconstructed images.
A computer device comprising:

at least one processor;

a memory, arranged to store at least one program;

The at least one program, when executed by the at least one processor, causes the at least one processor to implement the method of any one of claims 1-6.
A computer-readable storage medium storing a computer program, wherein the program, when executed by a processor, implements the method according to any one of claims 1-6.