WO2021077557A1 - 一种磁共振图像重建方法、装置、设备和介质 - Google Patents

一种磁共振图像重建方法、装置、设备和介质 Download PDF

Info

Publication number
WO2021077557A1
WO2021077557A1 PCT/CN2019/123063 CN2019123063W WO2021077557A1 WO 2021077557 A1 WO2021077557 A1 WO 2021077557A1 CN 2019123063 W CN2019123063 W CN 2019123063W WO 2021077557 A1 WO2021077557 A1 WO 2021077557A1
Authority
WO
WIPO (PCT)
Prior art keywords
sub
image reconstruction
magnetic resonance
data
solution
Prior art date
Application number
PCT/CN2019/123063
Other languages
English (en)
French (fr)
Inventor
梁栋
程静
王海峰
朱燕杰
刘新
郑海荣
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2021077557A1 publication Critical patent/WO2021077557A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/30—Assessment of water resources

Definitions

  • the embodiments of the present invention relate to medical imaging technology, for example, to a magnetic resonance image reconstruction method, device, equipment, and medium.
  • Magnetic resonance uses static magnetic field and radio frequency magnetic field to image human tissues. It not only provides rich tissue contrast, but also has no side effects on the human body. Therefore, it has become a powerful tool for medical clinical diagnosis.
  • deep learning methods are often used for image reconstruction, such as the use of neural networks to learn the optimal parameters required for reconstruction from a large amount of training data or directly learn from under-collected data to full-scale image reconstruction. It adopts the mapping relationship between images to achieve better imaging quality and higher acceleration than traditional parallel imaging or compressed sensing methods.
  • the ADMM algorithm that is, the alternating direction multiplier method
  • the ADMM algorithm uses the Decomposition-Coordination process to decompose a large global problem into multiple smaller and easier to solve local sub-problems, and obtains the solution of the large global problem by coordinating the solutions of the sub-problems.
  • the ADMM-net method which combines deep learning and ADMM algorithm, uses deep neural network to learn the parameters in the algorithm, which solves the problem of difficult adjustment of parameters and long iteration time in the optimization problem.
  • the structure of the neural network structure model is relatively fixed, that is, the relationship between the parameters of the solution of each local sub-problem is fixed, and the learning ability of the neural network is not fully utilized, which leads to the reconstruction of the image
  • the imaging quality needs to be improved.
  • the embodiment of the present invention provides a method, device, equipment and medium for magnetic resonance image reconstruction, so as to improve the network freedom of the neural network, learn more prior information, and improve the image quality.
  • an embodiment of the present invention provides a magnetic resonance image reconstruction method, the method including:
  • the magnetic resonance data is input to an image reconstruction model based on an alternating direction multiplier algorithm to obtain a reconstructed target magnetic resonance image, wherein the image reconstruction model is an iterative relationship after decomposing the original image reconstruction model and solving iteratively The model obtained by solving the generalization of the formula.
  • the data fidelity item of the original image reconstruction model is a generalized indefinite item.
  • the process of training the image reconstruction model includes:
  • the original image reconstruction model is decomposed into a first sub-problem, a second sub-problem, and a third sub-problem, wherein the third sub-problem is the first sub-problem and the second sub-problem. Constraints on the solution of the sub-problems;
  • the determining each parameter value in the solution of the first sub-problem and the solution of the second sub-problem by using a convolutional neural network iterative calculation method includes:
  • the neural network structure includes four modules: a data layer, a reconstruction layer, an optimization layer, and a parameter update layer.
  • the loss function is the two-norm square of the difference between the reconstructed image obtained through the image reconstruction model and the reconstructed image corresponding to the full-sampled magnetic resonance data.
  • an embodiment of the present invention also provides a magnetic resonance image reconstruction device, which includes:
  • the data acquisition module is configured to acquire under-sampled magnetic resonance data
  • the image reconstruction module is configured to input the magnetic resonance data into an image reconstruction model based on an alternating direction multiplier algorithm to obtain a reconstructed target magnetic resonance image, wherein the image reconstruction model decomposes the original image reconstruction model And iteratively solve the iterative relationship after generalization to solve the model obtained.
  • the data fidelity item of the original image reconstruction model is a generalized indefinite item.
  • the device further includes a model training module for training the image reconstruction model;
  • the model training module includes:
  • the sample data acquisition sub-module is configured to acquire full-sampled magnetic resonance data, and extract at least one set of under-sampled data from the full-sampled magnetic resonance data to obtain at least one set of under-sampled data and full-sampled magnetic resonance Data pair
  • a sample input sub-module configured to input the under-sampled data into the original image reconstruction model
  • the decomposition calculation sub-module is configured to decompose the original image reconstruction model into a first sub-problem, a second sub-problem, and a third sub-problem based on an alternating direction multiplier algorithm, wherein the third sub-problem is the first sub-problem. Constraints on the solution of a sub-problem and the second sub-problem;
  • a sub-problem solving sub-module configured to use a gradient descent method to solve the first sub-problem and the second sub-problem;
  • the parameter solving sub-module is configured to determine the solution of the first sub-problem and the solution of the second sub-problem through a convolutional neural network iterative calculation method for the solution of the first sub-problem and the solution of the second sub-problem.
  • Each parameter value in the solution of the problem completes the training of the image reconstruction model.
  • the parameter solving submodule is configured as:
  • the neural network structure includes four modules: a data layer, a reconstruction layer, an optimization layer, and a parameter update layer.
  • the loss function is the two-norm square of the difference between the reconstructed image obtained through the image reconstruction model and the reconstructed image corresponding to the full-sampled magnetic resonance data.
  • an embodiment of the present invention also provides a computer device, and the computer device includes:
  • One or more processors are One or more processors;
  • Memory used to store one or more programs
  • the one or more processors When the one or more programs are executed by the one or more processors, the one or more processors implement the magnetic resonance image reconstruction method provided in any embodiment of the present application.
  • an embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the magnetic resonance image reconstruction method as provided in any embodiment of the present application is implemented.
  • the under-sampled magnetic resonance data is input to the image reconstruction model based on the alternating direction multiplier algorithm.
  • the image reconstruction model is obtained by generalizing the iterative relationship after decomposing the original image reconstruction model and iteratively solving it. In the process of solving the original image reconstruction model, the relationship between each parameter in the solution of the sub-problem is broken, so that the network can learn the relationship between the parameters freely, which can increase the freedom of neural network learning, thereby improving the ADMM algorithm based Image quality after image reconstruction.
  • Fig. 1 is a flowchart of a magnetic resonance image reconstruction method in the first embodiment of the present invention
  • Fig. 2a is a flowchart of an image reconstruction model training method in the second embodiment of the present invention.
  • 2b is a schematic diagram of the structure of the convolutional neural network in the second embodiment of the present invention.
  • 2c is a comparison diagram of the reconstruction effect of image reconstruction performed by using the image reconstruction model obtained by training and image reconstruction performed by other algorithms in the second embodiment of the present invention
  • FIG. 3 is a schematic diagram of the structure of the magnetic resonance image reconstruction device in the third embodiment of the present invention.
  • Fig. 4 is a schematic structural diagram of a computer device in the fourth embodiment of the present invention.
  • FIG. 1 is a flowchart of a magnetic resonance image reconstruction method provided by Embodiment 1 of the present invention. This embodiment can be applied to the case of medical image reconstruction.
  • the magnetic resonance image reconstruction method may include the following steps:
  • the under-sampled magnetic resonance data is under-sampled magnetic resonance K-space data obtained by scanning by a magnetic resonance imaging device in a preset scanning manner.
  • K-space is also called Fourier space, which is the filling space of the original data of magnetic resonance signal with spatial positioning coding information.
  • Each magnetic resonance image has its corresponding K-space data lattice.
  • the under-sampled K-space data is the data of not all sampling points, which can reduce the time of data sampling.
  • the collected under-sampled K-space data is input to the pre-trained image reconstruction model, and the output of the model is the reconstructed target image.
  • the image reconstruction model is a model based on the alternating direction multiplier algorithm.
  • the under-sampled magnetic resonance data and its corresponding full-sampled magnetic resonance data are used as a data sample pair.
  • the fidelity and regular terms are mathematical models of indefinite terms.
  • the image reconstruction model parameters that can meet the image quality requirements are determined, so as to obtain the trained image reconstruction model.
  • the training sample includes multiple data sample pairs, and each sample pair can be rearranged from a set of full-sampled magnetic resonance data into multiple under-sampled magnetic resonance data according to a preset rule.
  • Each under-sampled magnetic resonance data and the group of full-sampled magnetic resonance data form a data sample pair.
  • multiple sets of full-sampled magnetic resonance data may correspond to one set of under-sampled magnetic resonance data, so as to obtain multiple pairs of sample data.
  • being able to meet the image quality requirement refers to the reconstructed image obtained after inputting the under-sampled magnetic resonance data into the image reconstruction model, and the reconstructed image is obtained after using the full-sampled magnetic resonance data corresponding to the under-sampled magnetic resonance data to perform image reconstruction. Compared with the reconstructed image, the difference between the two reconstructed images reaches the minimum value. After learning through the neural network, the image reconstruction model that meets the above conditions can be obtained, and the model training process can be completed.
  • the image reconstruction model in the embodiment of the present invention is a model obtained by decomposing the original image reconstruction model and generalizing the iterative relationship after iterative solution, where the data fidelity term of the original image reconstruction model can be It is an indeterminate term after generalization, which relieves the drawbacks of the original image reconstruction model that the data needs to be built on the premise of linear unbiased estimation, and more effectively guarantees the consistency of the data.
  • the technical solution of this embodiment is to input the under-sampled magnetic resonance data into the image reconstruction model based on the alternating direction multiplier algorithm.
  • the image reconstruction model generalizes the iterative relationship after the original image reconstruction model is decomposed and iteratively solved.
  • the model obtained after solving, and the data fidelity term in the original image reconstruction model is indefinite.
  • the relationship between the parameters in the solution of the sub-problem is broken, so that the network can learn the parameters freely. Relationship, thereby improving the image quality after image reconstruction based on the ADMM algorithm.
  • Fig. 2a is a flowchart of the image reconstruction model training method provided in the second embodiment of the invention. This embodiment describes the image reconstruction model training process on the basis of the above-mentioned embodiment.
  • the image reconstruction model training process may include the following steps:
  • S210 Acquire full-sampled magnetic resonance data, and extract at least one set of under-sampled data from the full-sampled magnetic resonance data to obtain at least one set of under-sampled data and full-sampled magnetic resonance data.
  • This step is the model training sample collection process.
  • the sampling data of the corresponding sampling line can be selected from the full-sampled data according to a preset rule to obtain the under-sampled data.
  • a preset rule to obtain the under-sampled data.
  • the sampling data of at least one set of 64 sampling lines can be extracted from the full sampling data of a set of 256 sampling lines as the under-sampling magnetic resonance data according to a preset rule, so that at least one set of under-sampling can be obtained.
  • a sample data pair composed of magnetic resonance data and corresponding full-sampling magnetic resonance data.
  • the original image reconstruction model can be expressed as: min m F(Am,f)+ ⁇ R(m) , Where m is the image to be reconstructed, f is the under-sampled k-space data, A represents the coding matrix, and represents the under-sampled Fourier transform operator in single-channel magnetic resonance imaging, ⁇ is the regular parameter, R(m ) Is the regular function, and F(Am,f) is the data fidelity function. Taking the F(Am,f) function as the data fidelity term function takes into account the general situation and is a more effective data consistency guarantee method. Unlike the model applicable to the ADMM-net method, the least squares constraint is based on linear Under the premise of unbiased estimation, the 2-norm between the reconstructed k-space and the sampling point is used to characterize the data fidelity term.
  • the process of decomposing the mathematical model is to introduce the z variable, which can be understood as the denoising image of m.
  • the original mathematical model is decomposed into three non-constrained sub-problems. among them, For the first sub-question, For the second sub-problem, argmax ⁇ ⁇ , mz> is the third sub-problem.
  • S240 Solve the first sub-problem and the second sub-problem by using a gradient descent method.
  • i and k are the number of inner loops of the first sub-problem and the second sub-problem respectively, and n is the number of iterations of the ADMM algorithm.
  • ⁇ 1 , ⁇ 2 , ⁇ 1 and ⁇ 2 are the parameters of the sub-problems, which will be given initial values during the calculation of the algorithm.
  • the initial value can be an empirical value.
  • F'and R' are the first-order partial derivatives of the functions F and R, that is, the first-order partial derivatives of the data fidelity function and the regular function.
  • the parameters in the solution of the first sub-problem and the solution of the second sub-problem are determined by a convolutional neural network iterative calculation method Value to complete the training of the image reconstruction model.
  • a convolutional neural network is used to fit the first-order partial derivative function of the data fidelity term function in the solution of the first sub-problem and the data regularization term function in the solution of the second sub-problem.
  • the first-order partial derivative function that is, the convolutional neural network CNN is used to replace the functions F′ and R′ in the formula in step S240, wherein each parameter ( ⁇ The initial values of 1 , ⁇ 2 , ⁇ 1 and ⁇ 2 ) are empirical values, which can be expressed as the following formula:
  • the relationship between the input items in the solution of the first sub-problem and the solution of the second sub-problem after function fitting is generalized. That is, the parameters ( ⁇ 1 , ⁇ 2 , ⁇ 1 and ⁇ 2 ) in the solutions of the first sub-problem and the second sub-problem no longer use empirical values as initial values, so that in the process of neural network learning, Break the parameter relationship between the input items in the solution of the first sub-problem and the solution of the second sub-problem, and determine better parameter values based on the training data.
  • the solution of each sub-problem after generalization can be expressed as: Can be named ADMM-net-ultimate.
  • ADMM-net-ultimate After a preset number of iterations, determine the value of each parameter in the solution of the first sub-problem after generalization and the solution of the second sub-problem, until the reconstructed image obtained by the image reconstruction model and the corresponding full-sampled magnetic resonance data The difference between the reconstructed images satisfies the loss function.
  • the network structure of ADMM-net-ultimate is shown in Fig. 2b, taking f as an input, and after n iterations of calculation, the output m that meets the demand is output.
  • ADMM-net-ultimate consists of four modules: data layer D, reconstruction layer M, optimization layer Z and parameter update layer P .
  • the size of the convolution kernel is 3x3
  • the activation function is the Relu function
  • the data layer has 2 convolution layers
  • the number of convolution kernels is (32, 2)
  • the reconstruction layer and optimization There are 3 convolutional layers in the layer
  • the number of convolution kernels is (32,32,2) and (8,8,2). Since the magnetic resonance signal is a complex signal, all data is divided into two channels, real part and imaginary part, for processing.
  • the loss function is defined as the mean square error: Where x is the reconstructed image output by the network, and x ref is the full-acquisition reconstructed image corresponding to the full-sampled magnetic resonance data corresponding to f.
  • the technical solution of this embodiment trains the mathematical model in which the data fidelity term and the regular term are indefinite terms through the learning process of the convolutional neural network, and breaks the difference between the parameters in the first sub-problem and the second sub-problem.
  • the relationship allows the network to learn the relationship between the parameters freely, so as to obtain an image reconstruction model that meets the requirements of the loss function, so that the image reconstruction model based on the ADMM algorithm has a wider application range and improves the quality of image reconstruction.
  • FIG 3 shows a schematic structural diagram of a magnetic resonance image reconstruction device in the third embodiment of the present invention.
  • This embodiment is suitable for medical image reconstruction.
  • the magnetic resonance image reconstruction device can be configured in medical equipment such as magnetic resonance imaging equipment and other medical equipment. Computer equipment.
  • the magnetic resonance image reconstruction apparatus may include: a data acquisition module 310 and an image reconstruction module 320.
  • the data acquisition module 310 is used to acquire under-sampled magnetic resonance data; the image reconstruction module 320 is used to input the magnetic resonance data into the image reconstruction model based on the alternating direction multiplier algorithm to obtain the reconstructed target magnetic resonance data.
  • a resonance image wherein the image reconstruction model is a model obtained by generalizing an iterative relationship after decomposing the original image reconstruction model and solving iteratively.
  • the technical solution of this embodiment inputs the under-sampled magnetic resonance data to the image reconstruction model based on the alternating direction multiplier algorithm.
  • the image reconstruction model generalizes the iterative relationship after decomposing the original image reconstruction model and solving iteratively.
  • the model obtained by the solution breaks the relationship between the parameters in the solution of the sub-problem in the process of solving the original image reconstruction model, allowing the network to learn the relationship between the parameters freely, solving the problem of decomposing the original image reconstruction model and iteratively solving the process
  • the relationship between the input items in the solution of each sub-problem is determined based on empirical values, and cannot be applied to all image reconstruction situations, thereby improving the image quality after image reconstruction based on the ADMM algorithm.
  • the data fidelity item of the original image reconstruction model is a generalized indefinite item.
  • the magnetic resonance image reconstruction device further includes a model training module for training the image reconstruction model;
  • the model training module may include:
  • the sample data acquisition sub-module is used to acquire full-sampled magnetic resonance data, and extract at least one set of under-sampled data from the full-sampled magnetic resonance data to obtain at least one set of under-sampled data and full-sampled magnetic resonance data Data pair;
  • the decomposition calculation sub-module is used to decompose the original image reconstruction model into a first sub-problem, a second sub-problem, and a third sub-problem based on the alternating direction multiplier algorithm, wherein the third sub-problem is the first sub-problem. Sub-problems and constraints on the solution of the second sub-problem;
  • the parameter solving sub-module is used to determine the solution of the first sub-problem and the second sub-problem through a convolutional neural network iterative calculation method for the solution of the first sub-problem and the solution of the second sub-problem The value of each parameter in the solution to complete the training of the image reconstruction model.
  • the parameter solving submodule is used to:
  • the neural network structure contains four modules: data layer, reconstruction layer, optimization layer and parameter update layer.
  • the loss function is the two-norm square of the difference between the reconstructed image obtained through the image reconstruction model and the reconstructed image corresponding to the full-sampled magnetic resonance data.
  • the magnetic resonance image reconstruction device provided by the embodiment of the present invention can execute the magnetic resonance image reconstruction method provided in any embodiment of the present application, and has the corresponding functional modules and beneficial effects for executing the magnetic resonance image reconstruction method.
  • FIG. 4 is a schematic structural diagram of a computer device according to Embodiment 5 of the present invention.
  • FIG. 4 shows a block diagram of an exemplary computer device 12 suitable for implementing the embodiments of the present application.
  • the computer device 12 shown in FIG. 4 is only an example.
  • the computer device 12 is represented in the form of a general-purpose computing device.
  • the components of the computer device 12 may include: one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components (including the system memory 28 and the processing unit 16).
  • the bus 18 represents one or more of several types of bus structures, including a memory bus or a memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any bus structure among multiple bus structures.
  • these architectures can include industry standard architecture (ISA) bus, microchannel architecture (MAC) bus, enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.
  • ISA industry standard architecture
  • MAC microchannel architecture
  • VESA Video Electronics Standards Association
  • PCI Peripheral Component Interconnect
  • the computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the device computer 12, including volatile and nonvolatile media, removable and non-removable media.
  • the system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and/or cache memory 32.
  • the computer device 12 may include other removable/non-removable, volatile/non-volatile computer system storage media.
  • the storage system 34 may be used to read and write non-removable, non-volatile magnetic media (not shown in FIG. 4, usually referred to as a "hard drive").
  • a disk drive for reading and writing to removable non-volatile disks such as "floppy disks”
  • a removable non-volatile optical disk such as CD-ROM, DVD-ROM
  • each drive can be connected to the bus 18 through one or more data media interfaces.
  • the system memory 28 may include at least one program product.
  • the program product has a set of (for example, at least one) program modules, which are configured to perform the functions of the various embodiments of the present application.
  • a program/utility tool 40 having a set of (at least one) program module 42 may be stored in, for example, the system memory 28.
  • Such program module 42 may include an operating system, one or more application programs, other program modules, and program data, Each of these examples or some combination may include the implementation of a network environment.
  • the program module 42 generally executes the functions and/or methods in the embodiments described in this application.
  • the computer device 12 may also communicate with one or more external devices 14 (such as keyboards, pointing devices, displays 24, etc.), and may also communicate with one or more devices that enable users to interact with the computer device 12, and/or communicate with Any device (such as a network card, modem, etc.) that enables the computer device 12 to communicate with one or more other computing devices. This communication can be performed through an input/output (I/O) interface 22.
  • the device 12 may also communicate with one or more networks (for example, a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through the network adapter 20. As shown in the figure, the network adapter 20 communicates with other modules of the computer device 12 through the bus 18.
  • LAN local area network
  • WAN wide area network
  • public network such as the Internet
  • the processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, for example, to implement the steps of a magnetic resonance image reconstruction method provided by the embodiment of the present invention, the method includes:
  • the magnetic resonance data is input to an image reconstruction model based on an alternating direction multiplier algorithm to obtain a reconstructed target magnetic resonance image, wherein the image reconstruction model is an iterative relationship after decomposing the original image reconstruction model and solving iteratively The model obtained by solving the generalization of the formula.
  • processor can also implement the technical solution of the magnetic resonance image reconstruction method provided by any embodiment of the present application.
  • the fifth embodiment provides a computer-readable storage medium on which a computer program is stored.
  • the program is executed by a processor, the steps of the magnetic resonance image reconstruction method as provided in any embodiment of the present application are realized, and the method includes:
  • the magnetic resonance data is input to an image reconstruction model based on an alternating direction multiplier algorithm to obtain a reconstructed target magnetic resonance image, wherein the image reconstruction model is an iterative relationship after decomposing the original image reconstruction model and solving iteratively The model obtained by solving the generalization of the formula.
  • the computer storage medium of the embodiment of the present invention 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, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or a combination of any of the above.
  • Examples of computer-readable storage media may include: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • the computer-readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device.
  • the computer-readable signal medium may include a data signal propagated in baseband or as a part of a carrier wave, and computer-readable program code is carried therein. This propagated data signal can take many forms, and can include electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium.
  • the computer-readable medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device .
  • the program code contained on the computer-readable medium can be transmitted by any suitable medium, which may include: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.
  • the computer program code used to perform the operations of this application can be written in one or more programming languages or a combination thereof.
  • the programming languages include object-oriented programming languages, such as Java, Smalltalk, C++, and also conventional Procedural programming language-such as "C" language or similar programming language.
  • the program code can be executed entirely on the user's computer, partly on the user's computer, executed as an independent software package, partly on the user's computer and partly executed on a remote computer, or entirely executed on the remote computer or server.
  • the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to access the Internet). connection).
  • LAN local area network
  • WAN wide area network
  • modules or steps of this application can be implemented by a general computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Alternatively, they can be implemented with program codes 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 made into individual integrated circuit modules, or multiple modules of them Or the steps can be implemented as a single integrated circuit module. In this way, this application is not limited to any specific combination of hardware and software.

Abstract

一种磁共振图像重建方法、装置、设备和介质。其中，方法包括：获取欠采样的磁共振数据（S110）；将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型（S120）。该方法解决了对原始图像重建模型进行分解并迭代求解过程中，分解后各子问题的解中的输入项的参数基于经验数值确定，并不能适用于所有的图像重建情况，从而提高基于ADMM算法的图像重建后的图像质量。

Description

一种磁共振图像重建方法、装置、设备和介质
本公开要求在2019年10月23日提交中国专利局、申请号为201911011740.4的中国专利申请的优先权，以上申请的全部内容通过引用结合在本公开中。
技术领域
本发明实施例涉及医学成像技术，例如涉及一种磁共振图像重建方法、装置、设备和介质。
背景技术
磁共振利用静磁场和射频磁场对人体组织成像，它不仅提供了丰富的组织对比度，且对人体无副作用，因此成为医学临床诊断的一种强有力的工具。
为了提高磁共振成像速度以及成像质量，相关技术中，多采用深度学习方法进行图像重建，如利用神经网络，从大量训练数据中学习重建所需的最优参数或者直接学习从欠采数据到全采图像之间的映射关系，从而取得比传统并行成像或者压缩感知方法更好的成像质量和更高的加速倍数。
其中，ADMM算法，即交替方向乘子方法，是一种求解优化问题的计算框架，适用于求解分布式凸优化问题。ADMM算法通过分解协调(Decomposition-Coordination)过程，将大的全局问题分解为多个较小、较容易求解的局部子问题，并通过协调子问题的解而得到大的全局问题的解。将深度学习与ADMM算法结合的ADMM-net方法采用深度神经网络来学习算法中的参数，解决了优化问题中参数难以调节、迭代时间长的问题。
但是，在利用神经网络学习算法确定局部子问题的解过程中，神经网络结构模型结构较为固定，即各局部子问题的解的参数间关系固定，没有充分利用神经网络的学习能力，导致重建图像的成像质量还有待提高。
发明内容
本发明实施例提供一种磁共振图像重建方法、装置、设备和介质，以实现提高神经网络的网络自由度，学习到更多的先验信息，提高图像质量。
第一方面，本发明实施例提供了一种磁共振图像重建方法，该方法包括：
获取欠采样的磁共振数据；
将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到 重建后的目标磁共振图像，其中，所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
可选的，所述原始图像重建模型的数据保真项为经过泛化的不定项。
可选的，所述图像重建模型训练的过程包括：
获取全采样的磁共振数据，并从所述全采样的磁共振数据中提取出至少一组欠采样数据，得到至少一组欠采样数据与全采样的磁共振数据的数据对；
将所述欠采样数据输入至所述原始图像重建模型；
基于交替方向乘子算法将所述原始图像重建模型分解为第一子问题、第二子问题和第三子问题，其中，所述第三子问题为所述第一子问题和所述第二子问题的解的约束条件；
采用梯度下降法求解所述第一子问题和所述第二子问题；
针对所述第一子问题的解和所述第二子问题的解，通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中每个参数值，完成图像重建模型的训练。
可选的，所述通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中每个参数值，包括：
采用卷积神经网络拟合所述第一子问题的解中的数据保真项函数的一阶偏导函数和所述第二子问题的解中的数据正则项函数的一阶偏导函数；
对经过函数拟合后的所述第一子问题的解和所述第二子问题的解中的每个输入项之间的关系进行泛化；
经过预设迭代次数，确定经过泛化处理的第一子问题的解和所述第二子问题的解中每个参数的数值，直到通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间的差值满足损失函数。
可选的，在每次迭代计算中，神经网络结构中包含有数据层、重建层、优化层和参数更新层四个模块。
可选的，所述损失函数为所述通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间差值的二范数的平方。
第二方面，本发明实施例还提供了一种磁共振图像重建装置，该装置包括：
数据获取模块，被配置为获取欠采样的磁共振数据；
图像重建模块，被配置为将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型 是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
可选的，所述原始图像重建模型的数据保真项为经过泛化的不定项。
可选的，所述装置还包括模型训练模块，用于对所述图像重建模型进行训练；所述模型训练模块包括：
样本数据获取子模块，被配置为获取全采样的磁共振数据，并从所述全采样的磁共振数据中提取出至少一组欠采样数据，得到至少一组欠采样数据与全采样的磁共振数据的数据对；
样本输入子模块，被配置为将所述欠采样数据输入至所述原始图像重建模型；
分解计算子模块，被配置为基于交替方向乘子算法将所述原始图像重建模型分解为第一子问题、第二子问题和第三子问题，其中，所述第三子问题为所述第一子问题和所述第二子问题的解的约束条件；
子问题求解子模块，被配置为采用梯度下降法求解所述第一子问题和所述第二子问题；
参数求解子模块，被配置为针对所述第一子问题的解和所述第二子问题的解，通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中每个参数值，完成图像重建模型的训练。
可选的，参数求解子模块被配置为：
采用卷积神经网络拟合所述第一子问题的解中的数据保真项函数的一阶偏导函数和所述第二子问题的解中的数据正则项函数的一阶偏导函数；
对经过函数拟合后的所述第一子问题的解和所述第二子问题的解中的每个输入项之间的关系进行泛化；
经过预设迭代次数，确定经过泛化处理的第一子问题的解和所述第二子问题的解中每个参数的数值，直到通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间的差值满足损失函数。
可选的，在每次迭代计算中，神经网络结构中包含有数据层、重建层、优化层和参数更新层四个模块。
可选的，所述损失函数为所述通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间差值的二范数的平方。
第三方面，本发明实施例还提供了一种计算机设备，所述计算机设备包括：
一个或多个处理器；
存储器，用于存储一个或多个程序；
当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如本申请任意实施例所提供的磁共振图像重建方法。
第四方面，本发明实施例还提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如本申请任意实施例所提供的磁共振图像重建方法。
本发明实施例通过将欠采样的磁共振数据输入至基于交替方向乘子算法的图像重建模型，该图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型，在对原始图像重建模型求解过程中打破子问题的解中每个参数之间的关系，使网络自由学习参数间的关系，可以提高神经网络学习的自由度，从而提高基于ADMM算法的图像重建后的图像质量。
附图说明
图1是本发明实施例一中的磁共振图像重建方法的流程图；
图2a是本发明实施例二中的图像重建模型训练方法的流程图；
图2b是本发明实施例二中的卷积神经网络结构示意图；
图2c是本发明实施例二中利用训练得到的图像重建模型进行图像重建与其他算法进行图像重建的重建效果对比图；
图3是本发明实施例三中的磁共振图像重建装置的结构示意图；
图4是本发明实施例四中的计算机设备的结构示意图。
具体实施方式
下面结合附图和实施例对本申请作进一步的详细说明。可以理解的是，此处所描述的具体实施例仅仅用于解释本申请，而非对本申请的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与本申请相关的部分而非全部结构。
实施例一
图1为本发明实施例一提供的磁共振图像重建方法的流程图，本实施例可适用于医学图像重建的情况。
如图1所示，磁共振图像重建方法可以包括如下步骤：
S110、获取欠采样的磁共振数据。
在一些实施例中，欠采样的磁共振数据是通过磁共振成像设备按照预设的扫描方式扫描获得的欠采样的磁共振K空间数据。
K空间也叫傅里叶空间，是带有空间定位编码信息的磁共振信号原始数据的填充空间，每一幅磁共振图像都有其相应的K空间数据点阵。欠采样的K空间数据则是非全部采样点的数据，这样可以减少数据采样的时间。
S120、将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型是所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
将采集的欠采样K空间数据输入至预先训练好的图像重建模型，该模型的输出则为重建后的目标图像。
其中，图像重建模型是基于交替方向乘子算法的一个模型，在该模型的训练过程中，通过将欠采样磁共振数据与其相应的全采样磁共振数据作为一个数据样本对，将欠采样磁共振数据输入至待训练的数据保真项和正则项均为不定项的数学模型，通过卷积神经网络的迭代计算，确定能够满足图像质量要求的图像重建模型参数，从而得到经过训练的图像重建模型。可以理解的是，在模型训练的过程中，训练样本包括多个数据样本对，各样本对可以是从一组全采样磁共振数据中按照预设规则重新排列为多个欠采样磁共振数据，各欠采样磁共振数据分别与该组全采样磁共振数据组成数据样本对。或者，还可以是多组全采样磁共振数据分别对应一组欠采样磁共振数据，从而得到多个样本数据对。
在一些实施例中，能够满足图像质量要求是指将欠采样磁共振数据输入至图像重建模型后得到的重建图像，与利用该欠采样磁共振数据对应的全采样磁共振数据进行图像重建后得到的重建图像相比，使两个重建图像间的差值达到最小值。当通过神经网络学习后，即可得到满足上述条件的图像重建模型，完成模型训练过程。
需要说明的是，本发明实施例中的图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型，其中，原始图像重建模型的数据保真项可以是经过泛化后的不定项，这样解除了原始图像重建模型中数据需要建立在线性无偏估计的前提下的弊端，更有效的保障数据的一致 性，此外，在建立图像重建模型的过程中，对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解，打破分解后得到的子问题的解中各参数之间的关系，使网络自由学习参数间的关系，使得目标图像重建模型的适用性更广，提高了重建图像的质量。
本实施例的技术方案，通过将欠采样的磁共振数据输入至基于交替方向乘子算法的图像重建模型，该图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型，而且原始图像重建模型中的数据保真项为不定项，在对原始图像重建模型求解过程中打破子问题的解中各参数之间的关系，使网络自由学习参数间的关系，从而提高基于ADMM算法的图像重建后的图像质量。
实施例二
图2a为发明实施例二提供的图像重建模型训练方法的流程图，本实施例在上述实施例的基础上说明图像重建模型训练的过程。
如图2a所示，图像重建模型训练的过程可以包括如下步骤：
S210、获取全采样的磁共振数据，并从所述全采样的磁共振数据中提取出至少一组欠采样数据，得到至少一组欠采样数据与全采样的磁共振数据的数据对。
该步骤即模型训练样本采集过程，针对全采样K空间数据，按照预设的规则从全采样数据中选取相应的采样线的采样数据可得到欠采样数据。。示例性的，在全采样过程中，有256条采样线，如果需要4倍加速采样，即需要采集64条线，那么欠采是指从这256条线中选取64条线进行采样。在一些实施例中，可以按照预设规则从一组256条采样线的全采样数据中提取出至少一组64条采样线的采样数据作为欠采样磁共振数据，从而可以得到至少一组欠采样磁共振数据与对应的全采样磁共振数据组成的样本数据对。
S220、将所述欠采样数据输入至所述原始图像重建模型。
在一些实施例中，由于图像重建模型是基于ADMM算法的模型，且模型中的数据保真项为不定项，原始图像重建模型可表示为：min mF(Am,f)+λR(m)，其中，m为所需重建的图像，f为欠采样的k空间数据，A表示编码矩阵，在单通道磁共振成像中表示欠采样傅里叶变换算子，λ为正则参数，R(m)为正则函数，F(Am,f)为数据保真项函数。以F(Am,f)函数作为数据保真项函数考虑到了一般 情况，是更为有效的数据一致性保障方法，而不同于ADMM-net方法适用的模型中，最小二乘约束是建立在线性无偏估计的前提下，采用重建的k空间与采样点之间的2范数来表征数据保真项。
S230、基于交替方向乘子算法将所述原始图像重建模型分解为第一子问题、第二子问题和第三子问题，其中，所述第三子问题为所述第一子问题和所述第二子问题的解的约束条件。
Figure PCTCN2019123063-appb-000001
将数学模型分解的过程是引入了z变量，其可以理解为m的去噪图像。在令m＝z的前提下，将原数学模型分解为非约束的三个子问题。其中，
Figure PCTCN2019123063-appb-000002
为第一子问题，
Figure PCTCN2019123063-appb-000003
Figure PCTCN2019123063-appb-000004
为第二子问题，argmax β<β,m-z>为第三子问题。
S240、采用梯度下降法求解所述第一子问题和所述第二子问题。
通过梯度下降法对第一子问题和第二子问题求解之后，各解的公式可表示为：
Figure PCTCN2019123063-appb-000005
其中，i和k分别为第一子问题和第二子问题的内循环次数，n为ADMM算法迭代次数。γ 1、γ 2、μ 1和μ 2为子问题中各项的参数，在算法计算的过程中，会被赋予初始值。初始值可以是经验值。F′和R′为函数F和R的一阶偏导，即数据保真项函数和正则函数的一阶偏导函数。
S250、针对所述第一子问题的解和所述第二子问题的解，通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中各参数值，完成图像重建模型的训练。
在一些实施例中，采用卷积神经网络拟合所述第一子问题的解中的数据保真项函数的一阶偏导函数和所述第二子问题的解中的数据正则项函数的一阶偏导函数，即用卷积神经网络CNN代替步骤S240中的公式中的函数F′和R′，其中，所述第一子问题和所述第二子问题的解中各参数(γ 1、γ 2、μ 1和μ 2)的初始值为经验值，可表示为如下公式：
Figure PCTCN2019123063-appb-000006
然后，对经过函数拟合后的所述第一子问题的解和所述第二子问题的解中的各输入项之间的关系进行泛化。即所述第一子问题和所述第二子问题的解中各参数(γ 1、γ 2、μ 1和μ 2)不再采用经验值作为初始值，使在神经网络学习的过程中，打破第一子问题的解与第二子问题的解中的各输入项的参数关系，基于训练数据去确定更优的参数值。经过泛化处理的各子问题的解可表示为:
Figure PCTCN2019123063-appb-000007
可命名为ADMM-net-ultimate。
经过预设迭代次数，确定经过泛化处理的第一子问题的解和所述第二子问题的解中各参数的数值，直到通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间的差值满足损失函数。在一些实施例中，ADMM-net-ultimate的网络结构如图2b所示，以f作为输入，经过n次迭代计算，输出满足需求的m。以第二次迭代(iter-2)计算的过程为例进行说明：每次迭代中，ADMM-net-ultimate由四个模块组成：数据层D，重建层M，优化层Z和参数更新层P。在一个实施例中可设置为迭代15次，卷积核大小为3x3，激 活函数为Relu函数，数据层有2个卷积层，卷积核个数为(32，2)，重建层和优化层有3个卷积层，卷积核个数为(32,32,2)和(8,8,2)。由于磁共振信号为复数信号，因此所有数据分为实部和虚部两个通道进行处理。训练过程中，损失函数定义为均方误差：
Figure PCTCN2019123063-appb-000008
其中x为网络输出的重建图像，x ref为与f相应全采样磁共振数据对应的全采重建图像。
利用经过该训练过程得到的图像重建模型进行图像重建，图像重建的效果与利用其他算法得到的重建图像的效果对比可参考图2c所示的对比图，其中，ref是指利用与输入的欠采样数据对应的全采样磁共振数据进行图像重建得到的重建图像；ADMM-net-ultimate是指通过本实施例的技术方案，利用输入的欠采样数据进行图像重建得到的重建图像；ADMM-net是指利用相关技术中的结合神经网络算法的ADMM-net算法进行图像重建得到的重建图像；Zero-filling是指用0来填充非采样点数据之后得到的重建图像。
本实施例的技术方案，通过卷积神经网络的学习过程，对数据保真项和正则项均为不定项的数学模型进行训练，打破第一子问题和第二子问题中各参数之间的关系，使网络自由学习参数间的关系，从而得到满足损失函数要求的图像重建模型，使基于ADMM算法的图像重建模型适用范围更广，提高了图像重建的质量。
实施例三
图3示出了本发明实施例三中的磁共振图像重建装置的结构示意图，该实施例适用于医学图像重建的情况中，磁共振图像重建装置可配置于磁共振成像设备等医学设备及其他计算机设备中。
如图3所示，磁共振图像重建装置可以包括：数据获取模块310和图像重建模块320。
其中，数据获取模块310，用于获取欠采样的磁共振数据；图像重建模块320，用于将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
本实施例技术方案，通过将欠采样的磁共振数据输入至基于交替方向乘子算法的图像重建模型，该图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型，在对原始图像重建模型求解过程中打破子问题的解中各参数之间的关系，使网络自由学习参数间的关系，解决了对原始图像重建模型进行分解并迭代求解过程中，分解后各子问题的解中的输入项之间的关系基于经验数值确定，并不能适用于所有的图像重建情况，从而提高基于ADMM算法的图像重建后的图像质量。
可选的，所述原始图像重建模型的数据保真项为经过泛化的不定项。
可选的，磁共振图像重建装置还包括模型训练模块，用于对所述图像重建模型进行训练；所述模型训练模块可以包括：
样本数据获取子模块，用于获取全采样的磁共振数据，并从所述全采样的磁共振数据中提取出至少一组欠采样数据，得到至少一组欠采样数据与全采样的磁共振数据的数据对；
样本输入子模块，用于将所述欠采样数据输入至所述原始图像重建模型；
分解计算子模块，用于基于交替方向乘子算法将所述原始图像重建模型分解为第一子问题、第二子问题和第三子问题，其中，所述第三子问题为所述第一子问题和所述第二子问题的解的约束条件；
子问题求解子模块，用于采用梯度下降法求解所述第一子问题和所述第二子问题；
参数求解子模块，用于针对所述第一子问题的解和所述第二子问题的解，通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中各参数值，完成图像重建模型的训练。
可选的，参数求解子模块用于：
采用卷积神经网络拟合所述第一子问题的解中的数据保真项函数的一阶偏导函数和所述第二子问题的解中的数据正则项函数的一阶偏导函数；
对经过函数拟合后的所述第一子问题的解和所述第二子问题的解中的各输入项之间的关系进行泛化；
经过预设迭代次数，确定经过泛化处理的第一子问题的解和所述第二子问题的解中各参数的数值，直到通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间的差值满足损失函数。
可选的，在每次迭代计算中，神经网络结构中包含有数据层、重建层、优 化层和参数更新层四个模块。
可选的，所述损失函数为所述通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间差值的二范数的平方。
本发明实施例所提供的磁共振图像重建装置可执行本申请任意实施例所提供的磁共振图像重建方法，具备执行磁共振图像重建方法相应的功能模块和有益效果。
实施例四
图4为本发明实施例五提供的一种计算机设备的结构示意图。图4示出了适于用来实现本申请实施方式的示例性计算机设备12的框图。图4显示的计算机设备12仅仅是一个示例。
如图4所示，计算机设备12以通用计算设备的形式表现。计算机设备12的组件可以包括：一个或者多个处理器或者处理单元16，系统存储器28，连接不同系统组件(包括系统存储器28和处理单元16)的总线18。
总线18表示几类总线结构中的一种或多种，包括存储器总线或者存储器控制器，外围总线，图形加速端口，处理器或者使用多种总线结构中的任意总线结构的局域总线。举例来说，这些体系结构可以包括工业标准体系结构(ISA)总线，微通道体系结构(MAC)总线，增强型ISA总线、视频电子标准协会(VESA)局域总线以及外围组件互连(PCI)总线。
计算机设备12典型地包括多种计算机系统可读介质。这些介质可以是任何能够被设备计算机12访问的可用介质，包括易失性和非易失性介质，可移动的和不可移动的介质。
系统存储器28可以包括易失性存储器形式的计算机系统可读介质，例如随机存取存储器(RAM)30和/或高速缓存存储器32。计算机设备12可以包括其它可移动/不可移动的、易失性/非易失性计算机系统存储介质。仅作为举例，存储系统34可以用于读写不可移动的、非易失性磁介质(图4未显示，通常称为“硬盘驱动器”)。尽管图4中未示出，可以提供用于对可移动非易失性磁盘(例如“软盘”)读写的磁盘驱动器，以及对可移动非易失性光盘(例如CD-ROM,DVD-ROM或者其它光介质)读写的光盘驱动器。在这些情况下，每个驱动器可以通过一个或者多个数据介质接口与总线18相连。系统存储器28可以包括至少一个程序产品，该程序产品具有一组(例如至少一个)程序模块，这些程 序模块被配置以执行本申请各实施例的功能。
具有一组(至少一个)程序模块42的程序/实用工具40，可以存储在例如系统存储器28中，这样的程序模块42可以包括操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。程序模块42通常执行本申请所描述的实施例中的功能和/或方法。
计算机设备12也可以与一个或多个外部设备14(例如键盘、指向设备、显示器24等)通信，还可与一个或者多个使得用户能与该计算机设备12交互的设备通信，和/或与使得该计算机设备12能与一个或多个其它计算设备进行通信的任何设备(例如网卡，调制解调器等等)通信。这种通信可以通过输入/输出(I/O)接口22进行。并且，设备12还可以通过网络适配器20与一个或者多个网络(例如局域网(LAN)，广域网(WAN)和/或公共网络，例如因特网)通信。如图所示，网络适配器20通过总线18与计算机设备12的其它模块通信。应当明白，尽管图中未示出，可以结合计算机设备12使用其它硬件和/或软件模块，可以包括：微代码、设备驱动器、冗余处理单元、外部磁盘驱动阵列、RAID系统、磁带驱动器以及数据备份存储系统等。
处理单元16通过运行存储在系统存储器28中的程序，从而执行各种功能应用以及数据处理，例如实现本发实施例所提供的一种磁共振图像重建方法步骤，该方法包括：
获取欠采样的磁共振数据；
将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
当然，本领域技术人员可以理解，处理器还可以实现本申请任意实施例所提供的磁共振图像重建方法的技术方案。
实施例五
本实施例五提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如本申请任意实施例所提供的磁共振图像重建方法步骤，该方法包括：
获取欠采样的磁共振数据；
将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
本发明实施例的计算机存储介质，可以采用一个或多个计算机可读的介质的任意组合。计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质。计算机可读存储介质例如可以是：电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质例子(非穷举的列表)可以包括：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本文件中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。
计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，可以包括电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
计算机可读介质上包含的程序代码可以用任何适当的介质传输，可以包括：无线、电线、光缆、RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言或其组合来编写用于执行本申请操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言，诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络，包括局域网(LAN)或广域网(WAN)，连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。
本领域普通技术人员应该明白，上述的本申请的各模块或各步骤可以用通 用的计算装置来实现，它们可以集中在单个计算装置上，或者分布在多个计算装置所组成的网络上，可选地，他们可以用计算机装置可执行的程序代码来实现，从而可以将它们存储在存储装置中由计算装置来执行，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本申请不限制于任何特定的硬件和软件的结合。

Claims (10)

  1. 一种磁共振图像重建方法，包括：
    获取欠采样的磁共振数据；
    将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
  2. 根据权利要求1所述的方法，其中，所述原始图像重建模型的数据保真项为经过泛化的不定项。
  3. 根据权利要求1或2所述的方法，其中，所述图像重建模型训练的过程包括：
    获取全采样的磁共振数据，并从所述全采样的磁共振数据中提取出至少一组欠采样数据，得到至少一组欠采样数据与全采样的磁共振数据的数据对；
    将所述欠采样数据输入至所述原始图像重建模型；
    基于交替方向乘子算法将所述原始图像重建模型分解为第一子问题、第二子问题和第三子问题，其中，所述第三子问题为所述第一子问题和所述第二子问题的解的约束条件；
    采用梯度下降法求解所述第一子问题和所述第二子问题；
    针对所述第一子问题的解和所述第二子问题的解，通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中每个参数值，完成图像重建模型的训练。
  4. 根据权利要求3所述的方法，其中，所述通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中每个参数值，包括：
    采用卷积神经网络拟合所述第一子问题的解中的数据保真项函数的一阶偏导函数和所述第二子问题的解中的数据正则项函数的一阶偏导函数；
    对经过函数拟合后的所述第一子问题的解和所述第二子问题的解中的每个输入项之间的关系进行泛化；
    经过预设迭代次数，确定经过泛化处理的第一子问题的解和所述第二子问题的解中每个参数的数值，直到通过图像重建模型得到的重建图像与对应全采样磁共振数据的重建图像之间的差值满足损失函数。
  5. 根据权利要求4所述的方法，其中，在每次迭代计算中，神经网络结构中包含有数据层、重建层、优化层和参数更新层四个模块。
  6. 根据权利要求4所述的方法，其中，所述损失函数为所述通过图像重建 模型得到的重建图像与对应全采样磁共振数据的重建图像之间差值的二范数的平方。
  7. 一种磁共振图像重建装置，包括：
    数据获取模块，被配置为获取欠采样的磁共振数据；
    图像重建模块，被配置为将所述磁共振数据输入至基于交替方向乘子算法的图像重建模型，以得到重建后的目标磁共振图像，其中，所述图像重建模型是对原始图像重建模型分解并迭代求解后的迭代关系式进行泛化后求解得到的模型。
  8. 根据权利要求7所述的装置，所述装置还包括模型训练模块，用于对所述图像重建模型进行训练；所述模型训练模块包括：
    样本数据获取子模块，被配置为获取全采样的磁共振数据，并从所述全采样的磁共振数据中提取出至少一组欠采样数据，得到至少一组欠采样数据与全采样的磁共振数据的数据对；
    样本输入子模块，被配置为将所述欠采样数据输入至所述原始图像重建模型；
    分解计算子模块，被配置为基于交替方向乘子算法将所述原始图像重建模型分解为第一子问题、第二子问题和第三子问题，其中，所述第三子问题为所述第一子问题和所述第二子问题的解的约束条件；
    子问题求解子模块，被配置为采用梯度下降法求解所述第一子问题和所述第二子问题；
    参数求解子模块，被配置为针对所述第一子问题的解和所述第二子问题的解，通过卷积神经网络迭代计算方法确定所述第一子问题的解和所述第二子问题的解中每个参数值，完成图像重建模型的训练。
  9. 一种计算机设备，所述计算机设备包括：
    一个或多个处理器；
    存储器，用于存储一个或多个程序；
    当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如权利要求1-6中任一所述的磁共振图像重建方法。
  10. 一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如权利要求1-6中任一所述的磁共振图像重建方法。
PCT/CN2019/123063 2019-10-23 2019-12-04 一种磁共振图像重建方法、装置、设备和介质 WO2021077557A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911011740.4 2019-10-23
CN201911011740.4A CN110766769B (zh) 2019-10-23 2019-10-23 一种磁共振图像重建方法、装置、设备和介质

Publications (1)

Publication Number Publication Date
WO2021077557A1 true WO2021077557A1 (zh) 2021-04-29

Family

ID=69333185

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/123063 WO2021077557A1 (zh) 2019-10-23 2019-12-04 一种磁共振图像重建方法、装置、设备和介质

Country Status (2)

Country Link
CN (1) CN110766769B (zh)
WO (1) WO2021077557A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113592973A (zh) * 2021-07-30 2021-11-02 哈尔滨工业大学（深圳） 一种基于多频复数卷积的磁共振图像重建方法及装置
CN113592972A (zh) * 2021-07-30 2021-11-02 哈尔滨工业大学（深圳） 一种基于多模态聚合的磁共振图像重建方法及装置
CN113763501A (zh) * 2021-09-08 2021-12-07 上海壁仞智能科技有限公司 图像重建模型的迭代方法和图像重建方法

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111127575A (zh) * 2019-12-12 2020-05-08 深圳先进技术研究院 图像重建方法、计算机可读介质及计算机设备
CN113359076B (zh) * 2020-03-06 2022-09-27 上海联影医疗科技股份有限公司 磁共振成像方法、装置、计算机设备和存储介质
CN111612865B (zh) * 2020-05-18 2023-04-18 中山大学 一种基于条件生成对抗网络的mri成像方法及装置
CN111598966B (zh) * 2020-05-18 2023-04-18 中山大学 一种基于生成对抗网络的磁共振成像方法及装置
CN111714124B (zh) * 2020-06-18 2023-11-03 中国科学院深圳先进技术研究院 磁共振电影成像方法、装置、成像设备及存储介质
CN112213674B (zh) * 2020-09-11 2023-03-21 上海东软医疗科技有限公司 磁共振压缩感知重建方法及装置
CN112865748B (zh) * 2021-01-13 2022-05-10 西南大学 基于递归最小二乘的在线分布式多任务图滤波器构建方法
CN115115722A (zh) * 2021-03-17 2022-09-27 中国科学院深圳先进技术研究院 图像重建模型生成及图像重建方法、装置、设备和介质
CN116203484A (zh) * 2021-11-30 2023-06-02 中国科学院深圳先进技术研究院 一种图像生成方法、装置、设备和介质
CN114663544A (zh) * 2022-05-26 2022-06-24 中国科学技术大学 一种基于深度图像先验的电阻抗图像重建方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110052023A1 (en) * 2009-08-28 2011-03-03 International Business Machines Corporation Reconstruction of Images Using Sparse Representation
CN109658469A (zh) * 2018-12-13 2019-04-19 深圳先进技术研究院 一种基于深度先验学习的头颈联合成像方法和装置
CN109671129A (zh) * 2018-12-14 2019-04-23 深圳先进技术研究院 一种自适应参数学习的动态磁共振图像重建方法和装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108335339B (zh) * 2018-04-08 2021-10-22 朱高杰 一种基于深度学习和凸集投影的磁共振重建方法
CN109325985B (zh) * 2018-09-18 2020-07-21 上海联影智能医疗科技有限公司 磁共振图像重建方法、装置和计算机可读存储介质
CN109375125B (zh) * 2018-10-25 2020-12-25 哈尔滨理工大学 一种修正正则化参数的压缩感知磁共振成像重建方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110052023A1 (en) * 2009-08-28 2011-03-03 International Business Machines Corporation Reconstruction of Images Using Sparse Representation
CN109658469A (zh) * 2018-12-13 2019-04-19 深圳先进技术研究院 一种基于深度先验学习的头颈联合成像方法和装置
CN109671129A (zh) * 2018-12-14 2019-04-23 深圳先进技术研究院 一种自适应参数学习的动态磁共振图像重建方法和装置

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113592973A (zh) * 2021-07-30 2021-11-02 哈尔滨工业大学（深圳） 一种基于多频复数卷积的磁共振图像重建方法及装置
CN113592972A (zh) * 2021-07-30 2021-11-02 哈尔滨工业大学（深圳） 一种基于多模态聚合的磁共振图像重建方法及装置
CN113592972B (zh) * 2021-07-30 2023-11-14 哈尔滨工业大学（深圳） 一种基于多模态聚合的磁共振图像重建方法及装置
CN113763501A (zh) * 2021-09-08 2021-12-07 上海壁仞智能科技有限公司 图像重建模型的迭代方法和图像重建方法
CN113763501B (zh) * 2021-09-08 2024-02-27 上海壁仞智能科技有限公司 图像重建模型的迭代方法和图像重建方法

Also Published As

Publication number Publication date
CN110766769A (zh) 2020-02-07
CN110766769B (zh) 2023-08-11

Similar Documents

Publication Publication Date Title
WO2021077557A1 (zh) 一种磁共振图像重建方法、装置、设备和介质
CN110766768B (zh) 一种磁共振图像重建方法、装置、设备和介质
EP3822865A1 (en) Synthesizing data for training one or more neural networks
CN109712234B (zh) 三维人体模型的生成方法、装置、设备和存储介质
CN114365156A (zh) 用于神经网络的迁移学习
DE112021002631T5 (de) Bilderzeugung unter verwendung eines oder mehrerer neuronaler netzwerke
WO2021098534A1 (zh) 相似度确定、网络训练、查找方法及装置、电子装置和存储介质
DE112021002657T5 (de) Bilderzeugung unter verwendung eines oder mehrerer neuronaler netzwerke
DE112020002425T5 (de) Bewegungsvorhersage unter verwendung eines oder mehrerer neuronaler netzwerke
CN114581587A (zh) 采用隐式形状的实时渲染
DE112021004543T5 (de) Bilderzeugung unter verwendung eines oder mehrerer neuronaler netzwerke
CN115081616A (zh) 一种数据的去噪方法以及相关设备
CN110717958A (zh) 一种图像重建方法、装置、设备及介质
CN115439341A (zh) 用于图像处理和光传输模拟系统和应用的使用向量的时空噪声掩模和采样
DE102021122615A1 (de) Bildverbesserung unter verwendung eines oder mehrerer neuronaler netzwerke
CN110728732A (zh) 一种图像重建方法、装置、设备及介质
DE102023125188A1 (de) Überblenden von videorahmen
DE102023125014A1 (de) Anwendungsprogrammierschnittstelle zum ausführen der rahmeninterpolation
DE102023102679A1 (de) Berechnung des optischen flusses mittels semiglobalem abgleich
WO2022193379A1 (zh) 图像重建模型生成及装置、图像重建方法及装置、设备、介质
US20220392023A1 (en) Spatio-temporal noise masks for image processing
US20220415481A1 (en) Mesh topology adaptation
WO2022193378A1 (zh) 图像重建模型生成方法及装置、图像重建方法及装置、设备、介质
US11922550B1 (en) Systems and methods for hierarchical text-conditional image generation
US20240095880A1 (en) Using a neural network to generate an upsampled image

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19950075

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19950075

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19950075

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13.10.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19950075

Country of ref document: EP

Kind code of ref document: A1