A low-field magnetic resonance rapid imaging method, a terminal device and a computer storage medium

Abstract

The application provides a low-field magnetic resonance fast imaging method, a terminal device and a computer storage medium. The low-field magnetic resonance fast imaging method comprises the following steps: reading undersampling imaging data about a target image; constructing a convolution feature subspace projection regularization model based on low-field data characteristics; and solving the convolution feature subspace projection regularization model in parallel based on a preset algorithm to obtain a high-quality magnetic resonance image. In the foregoing manner, the low-field magnetic resonance fast imaging method of the application can accelerate the obtaining of a magnetic resonance image comparable to a high-field magnetic resonance image by using the convolution feature subspace projection regularization model and the preset algorithm, thereby facilitating the improvement of clinical application efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of nuclear magnetic resonance imaging technology, in particular to a low-field magnetic resonance fast imaging method, a terminal device and a computer storage medium. BACKGROUND

[0002] Compared with medical imaging technologies such as CT, magnetic resonance imaging has no radioactive damage, and has rich contrast information and higher soft tissue resolution and functional imaging capability.

[0003] At present, the high-field magnetic resonance system with high image signal-to-noise ratio is complex, and for a long time, China's high-field magnetic resonance equipment has been heavily dependent on imports, and the price is expensive. The operation and maintenance of low-field magnetic resonance are simple, and its open magnet structure can eliminate the claustrophobia of the object, and at the same time facilitate the monitoring of life support system under intervention diagnosis and treatment. However, the signal-to-noise ratio of low-field magnetic resonance signal is low, and the image quality is relatively poor, and generally multiple repeated acquisition is needed to obtain high signal-to-noise ratio images. How to quickly obtain high-resolution magnetic resonance images has always been the core problem in the field of low-field magnetic resonance.

[0004] In recent years, the main strategy for fast magnetic resonance imaging technology is to reduce the data space (K-space) acquisition, and through mathematical modeling and optimization, the realization of accelerated imaging is transformed into a regularization solution of an ill-posed inverse problem, such as the compressed sensing method 1-3. However, the K-space observation noise under low-field conditions is large, and the low-field application based on the traditional compressed sensing sparse regularization method cannot meet the current clinical requirements of low-field ecological layout in terms of imaging quality and acceleration rate. SUMMARY

[0005] The present application provides a low-field magnetic resonance fast imaging method, a terminal device and a computer storage medium.

[0006] The present application provides a low-field magnetic resonance fast imaging method, which comprises:

[0007] Reading undersampled imaging data about a target image;

[0008] Constructing a convolution feature subspace projection regularization model based on low-field data characteristics;

[0009] Parallelly solving the convolution feature subspace projection regularization model based on a preset algorithm to obtain a high-quality magnetic resonance image.

[0010] In the embodiments of the present disclosure,

[0011] The reading of the undersampled imaging data about the target image comprises

[0012] Convolve the target image using a convolution feature code to obtain undersampled imaging data of the target image.

[0013] The undersampled imaging data includes structural information of the target image in several directions and / or several orders.

[0014] In the embodiments of the present disclosure,

[0015] The specific model of the convolution feature code is:

[0016]

[0017] Wherein, x is the target image, is a set of convolution filter kernels, λ is a preset weight parameter, z i is the i-th convolution feature.

[0018] In the embodiments of the present disclosure,

[0019] The convolution feature subspace projection regularization model is constructed using the undersampled imaging data, including:

[0020] Obtain the subspace of the undersampled imaging data and the basis function corresponding to the subspace;

[0021] Calculate the transpose of the basis function;

[0022] The convolution feature subspace projection regularization model is constructed using the undersampled imaging data, the basis function corresponding to the subspace and its transpose.

[0023] In the embodiments of the present disclosure,

[0024] The convolution feature subspace projection regularization model is:

[0025]

[0026] Wherein, is the target image, is a set of two-dimensional convolution filter kernels, u i is the i-th subspace projection coefficient. v i is the i-th orthogonal subspace, λ and γ are weight parameters, z i is the i-th convolution feature.

[0027] In the embodiments of the present disclosure,

[0028] The preset algorithm is a first boosting type plug and play iterative algorithm, and the algorithm framework is:

[0029] x t+1 = x t + D σ (xnoise -x t ), x noise =x k +l k-1

[0030] wherein x is the undersampled imaging data, D σ is a generalized denoising operator, x noise is noise information, and l is a multiplier term.

[0031] In the embodiments of the present disclosure,

[0032] The preset algorithm is a second lifting plug-and-play iterative algorithm, and the algorithm framework is:

[0033] x t+1 =D σ (x noise +x t )-x t , x noise =x k +l k-1

[0034] wherein x is the undersampled imaging data, D σ is a generalized denoising operator, x noise is noise information, and l is a multiplier term.

[0035] The present application also provides a terminal device, comprising:

[0036] A reading module is configured to read undersampled imaging data of a target image.

[0037] A construction module is configured to construct a convolution feature subspace projection regularization model based on low-field data characteristics.

[0038] A reconstruction module is configured to solve the convolution feature subspace projection regularization model in parallel based on a preset algorithm to obtain a high-quality magnetic resonance image.

[0039] The present application also provides another terminal device, comprising a memory and a processor, wherein the memory is coupled to the processor;

[0040] The memory is configured to store program data, and the processor is configured to execute the program data to implement the low-field magnetic resonance fast imaging method described above.

[0041] The present application also provides a computer storage medium for storing program data, which, when executed by a processor, is used to implement the low-field magnetic resonance fast imaging method described above.

[0042] The beneficial effects of the present application are: the terminal device reads the undersampling imaging data about the target image; a convolution feature subspace projection regularization model is constructed based on the low-field data characteristics; and the convolution feature subspace projection regularization model is solved in parallel based on a preset algorithm to obtain a high-quality magnetic resonance image. In the foregoing manner, the low-field magnetic resonance fast imaging method of the present application uses the convolution feature subspace projection regularization model and the preset algorithm to accelerate the obtaining of a magnetic resonance image comparable to a high-field magnetic resonance image, which is conducive to improving the clinical application efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0044] Figure 1 is a flowchart of an embodiment of the low-field magnetic resonance fast imaging method provided by the present application;

[0045] Figure 2 is a framework diagram of an embodiment of the low-field magnetic resonance fast imaging method provided by the present application;

[0046] Figure 3 is a comparison diagram of high-field data and data of a cardiac image provided by the present application;

[0047] Figure 4 is a structural diagram of an embodiment of the terminal device provided by the present application;

[0048] Figure 5 is a structural diagram of another embodiment of the terminal device provided by the present application;

[0049] Figure 6 is a structural diagram of an embodiment of the computer storage medium provided by the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0051] Currently, the main shortcomings of the traditional compressed sensing sparse regularization method in the face of low-field fast magnetic resonance imaging include: the strong degree of observation noise has an impact on the reliability of the measurement of the image sparse domain, that is, the sparse image features cannot be reliably expressed; the solving accuracy of the optimization problem and the algorithm implementation efficiency cannot meet the needs of actual clinical needs.

[0052] In view of the problems of the prior art, the present application provides a low-field magnetic resonance fast imaging method, specifically a convolution feature subspace projection regularization imaging method aiming to improve the quality of low-field magnetic resonance imaging.

[0053] Please refer to Figure 1 , Figure 1 is a flowchart of an embodiment of the low-field magnetic resonance fast imaging method provided by the present application.

[0054] Among them, the low-field magnetic resonance fast imaging method of the present application is applied to a terminal device, wherein the terminal device of the present application can be a server, or a system cooperated by a server and a terminal device. Accordingly, each part of the terminal device, such as each unit, sub-unit, module, and sub-module, can be all set in the server, or can be set in the server and the terminal device respectively.

[0055] Further, the above-mentioned server can be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software or software modules, such as software or software modules for providing a distributed server, or as a single software or software module, which is not specifically limited here. In some possible implementation manners, the magnetic resonance imaging method of the embodiment of the present application can be realized by a processor calling computer readable instructions stored in a memory.

[0056] Specifically, as shown in Figure 1 the low-field magnetic resonance fast imaging method of the embodiment of the present application specifically includes the following steps:

[0057] Step S11: reading the undersampled imaging data about the target image.

[0058] In the embodiment of the present disclosure, the terminal device reads the original data file collected by the nuclear magnetic resonance device, and the original data is used as full-sampling K-space data to perform an analog undersampling processing operation to obtain random variable-density undersampling space data. The more sparse the magnetic resonance image itself or its in a certain transform domain, the better the reconstruction quality.

[0059] Specifically, the magnetic resonance imaging data acquisition process based on the K-space undersampling mechanism can be discretely represented as:

[0060] b = Ax + ξ

[0061] wherein, is a target image to be reconstructed, is under-sampled data of K-space, is an under-sampled Fourier transform matrix, is a strong observation noise under the condition of assuming Gaussian distribution.

[0062] Generally speaking, it is difficult to solve the above underdetermined problem, and a regularization method can construct an unconstrained optimization model as follows:

[0063]

[0064] wherein, λ ∈ R + is a positive constant balancing data fitting and regularization term.

[0065] In the embodiment of the present disclosure, the terminal device convolves the target image by using the convolution feature coding to obtain under-sampled imaging data of the target image.

[0066] Specifically, the convolution sparse coding is equivalent to a group of linear filter kernels and corresponding convolution with sparse feature mapping. The specific model of the convolution sparse coding provided by the embodiment of the present disclosure is as follows:

[0067]

[0068] wherein, is an input image, is a group of convolution filter kernels, z i is the i-th convolution feature, and λ is a weight parameter.

[0069] The advantage of using the convolution feature coding in the embodiment of the present disclosure is that different levels of structural information of the target image can be obtained, including different directions and different orders.

[0070] Step S12: constructing a convolution feature subspace projection regularization model based on the low-field data characteristics.

[0071] Based on the existing cognition, the image signal-to-noise ratio is low under the low-field condition, and the convolution feature z i is sensitive to noise and other interference factors. Direct modeling based on the feature z i (such as sparsity measurement in the horizontal and vertical direction gradient domain) will limit the reliability of the overall regularization.

[0072] The embodiment of the present disclosure proposes a convolution feature subspace projection regularization model based on anti-noise interference, which is specifically as follows:

[0073]

[0074] where, is the subspace regularization of the convolution feature.

[0075] where, is the image to be reconstructed, is a set of two-dimensional convolution filter kernels, u i is the ith subspace projection coefficient. v i is the ith orthogonal subspace. λ and γ are weight parameters. z i is the ith convolution feature.

[0076] Step S13: Parallelly solving the convolution feature subspace projection regularization model based on a preset algorithm to obtain a high-quality magnetic resonance image.

[0077] In the embodiments of the present disclosure, the preset algorithm is specifically a plug-and-play iterative algorithm.

[0078] Specifically, the plug-and-play prior (P3) iteration is rooted in the proximal gradient algorithm, and the ADMM algorithm of the primal-dual alternating optimization is taken as an example.

[0079] Further, the specific convolution feature subspace projection regularization solving model contained in the embodiments of the present disclosure is:

[0080]

[0081] If the Lagrange multiplier term is introduced, an iterative optimization problem containing multiple subtasks is solved:

[0082]

[0083] where, for each u i , the subproblem

[0084] u i :

[0085] Further simplification gives

[0086] u i :

[0087] The proximal gradient projection used for solving u i in the traditional plug-and-play iteration directly uses a certain generalized denoising operator D σ to replace

[0088] u i :

[0089] In particular, the embodiments of the present application propose two kinds of promotion plug-and-play iterative algorithms, which differ in that the de-noising step uses a promotion de-noising mechanism, further improving the imaging quality and avoiding excessive smoothing in the reconstruction process:

[0090]

[0091]

[0092] For the optimization problem of low-field magnetic resonance fast imaging, the variational model is:

[0093]

[0094] An auxiliary variable w is introduced to solve a saddle-point problem as follows:

[0095]

[0096] Among them, the standard ADMM algorithm iteration framework includes a quadratic programming subproblem about x, a proximal gradient projection of the auxiliary variable w, and an update of the multiplier term l.

[0097] The embodiments of the present disclosure directly prox Φ (w; η) (subproblem about w) uses a generalized de-noising operator D σ Instead, such as non-local matching and filtering (BM3D) 10. The algorithm steps of P3-ADMM are:

[0098] x k :

[0099] w k =D σ (x k +l k-1 )

[0100] l k =l k-1 +(x k -x k )

[0101] Therefore, the two promotion plug-and-play iterative algorithm frameworks provided by the embodiments of the present application can be derived as follows:

[0102] x t+1 =x t +D σ (x noise -x t ), x noise =x k +l k-1

[0103] x t+1 = D σ (x noise + x t ) - x t , x noise = x k + l k-1

[0104] wherein t represents the iteration number of the de-noising link.

[0105] The embodiment of the application can further depict image details by using the above-mentioned boosting plug-and-play prior (Boosting P3) iterative algorithm, and avoid over-smoothing in the reconstruction process.

[0106] Further, the embodiment of the application can also use the double means of data end accelerated acquisition and algorithm parallel computing to further shorten the overall imaging time of low-field magnetic resonance. As shown in Figure 2 , Figure 2 is a framework schematic diagram of an embodiment of the low-field magnetic resonance fast imaging method provided by the application.

[0107] Specifically, the feature subspace operators under different convolution kernels are separable, so different calculation targets can be allocated to multiple calculation units, thereby realizing algorithm parallel iteration and being beneficial to avoiding lengthy waiting of the algorithm process.

[0108] In the embodiment of the application, the terminal device reads the undersampled imaging data of a target image; a convolution feature subspace projection regularization model is constructed based on the low-field data characteristics; and the convolution feature subspace projection regularization model is solved in parallel based on a preset algorithm to obtain a high-quality magnetic resonance image. Through the above-mentioned manner, the low-field magnetic resonance fast imaging method of the application uses the convolution feature subspace projection regularization model and the preset algorithm to accelerate to obtain a magnetic resonance image comparable to a high-field one, which is beneficial to improving the clinical application efficiency.

[0109] For details, please continue to refer to Figure 3 , Figure 3 is a comparison schematic diagram of high-field data and low-field data of a cardiac image provided by the application.

[0110] As shown in Figure 3 , the embodiment of the application verifies the effectiveness of the proposed magnetic resonance imaging method by simulating a low-field scene by adding noise and taking cardiac data as an example. Figure 3 The left side is a cardiac image of a certain frame (top: high-field data; bottom: low-field data), Figure 3 the middle column of the image is a horizontal gradient direction convolution feature amplitude graph, and it can be obviously observed that the sparsity of the original image is significantly reduced due to the influence of noise.Figure 3 The right side of each is a feature amplitude map based on subspace learning (low-rank space), and it is not difficult to find that the sparsity of the subspace of the noisy feature can still be ensured, and even similar sparsity to the original image is shown.

[0111] Therefore, compared with the traditional method, the low-field magnetic resonance fast imaging method provided by the application is more reliable and robust in imaging modeling on the one hand, and is more efficient in problem solving on the other hand. Specifically, the magnetic resonance imaging method provided by the application first constructs a convolution feature subspace regularization model, focusing on the effective expression of the convolution feature, to avoid the sensitivity of the original space regularization to noise; secondly, a boosting plug-and-play prior iterative algorithm is designed, and the parallel computing advantage under the condition that part of the operator is separable is considered, and a more accurate problem solution is obtained by using operator separation and parallel computing.

[0112] Those skilled in the art can understand that in the above method of the specific embodiment, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0113] To realize the low-field magnetic resonance fast imaging method of the above-mentioned embodiments, the application further provides a terminal device, please refer to Figure 4 , Figure 4 is a structural schematic diagram of an embodiment of the terminal device provided by the application.

[0114] As Figure 4 shown, the terminal device 300 provided by the application includes a reading module 31, a construction module 32, and a reconstruction module 33.

[0115] The reading module 31 is configured to read undersampled imaging data about a target image; the construction module 32 is configured to construct a convolution feature subspace projection regularization model based on low-field data characteristics; and the reconstruction module 33 is configured to solve the convolution feature subspace projection regularization model in parallel based on a preset algorithm to obtain a high-quality magnetic resonance image.

[0116] To realize the low-field magnetic resonance fast imaging method of the above-mentioned embodiments, the application further provides another terminal device, please refer to Figure 5 , Figure 5 is a structural schematic diagram of another embodiment of the terminal device provided by the application.

[0117] The terminal device 400 of the embodiment of the application includes a memory 41 and a processor 42, wherein the memory 41 and the processor 42 are coupled.

[0118] The memory 41 is configured to store program data, and the processor 42 is configured to execute the program data to implement the low-field magnetic resonance fast imaging method in the above embodiments.

[0119] In this embodiment, the processor 42 can also be referred to as a CPU (Central Processing Unit). The processor 42 can be an integrated circuit chip having a processing capability of signals. The processor 42 can also be a general processor, a DSP (Digital Signal Process), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The general processor can be a microprocessor or the processor 42 can also be any conventional processor.

[0120] The present application also provides a computer storage medium, such as a computer-readable storage medium 300 as shown in the figure. Figure 6 The computer storage medium 300 is configured to store program data 31, and the program data 31 is used to implement the low-field magnetic resonance fast imaging method in the above embodiments when executed by a processor.

[0121] The present application also provides a computer program product, wherein the computer program product includes a computer program, and the computer program is operable to cause a computer to execute the magnetic resonance imaging method in the embodiments of the present application. The computer program product can be a software installation package.

[0122] The low-field magnetic resonance fast imaging method in the above embodiments of the present application, when implemented in the form of a software function unit and sold or used as an independent product, can be stored in an apparatus, such as a computer-readable storage medium. Based on this understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk or an optical disk, and various other media that can store program codes.

[0123] The above merely describes the embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, which is made according to the content of the present application specification and drawings, is also included in the patent protection scope of the present application.

Claims

1. A method of low-field magnetic resonance fast imaging, characterized by, The low-field magnetic resonance fast imaging method comprises: reading undersampled imaging data about a target image; constructing a convolution feature subspace projection regularization model based on low-field data characteristics; solving the convolution feature subspace projection regularization model in parallel based on a preset algorithm to obtain a high-quality magnetic resonance image; wherein the convolution feature subspace projection regularization model is constructed based on low-field data characteristics, comprising: obtaining a subspace of the undersampled imaging data and a basis function corresponding to the subspace; calculating the transpose of the basis function; constructing the convolution feature subspace projection regularization model using the undersampled imaging data, the basis function corresponding to the subspace, and the transpose thereof; wherein the convolution feature subspace projection regularization model is: wherein, is the target image, is a set of two-dimensional convolution filter kernels, u i is the i-th subspace projection coefficient, v i is the i-th orthogonal subspace, λ and γ are weight parameters, z i is the i-th convolution feature.

2. The low-field magnetic resonance fast imaging method of claim 1, wherein the reading of the undersampled imaging data about the target image comprises convolving the target image using a convolution feature code to obtain the undersampled imaging data of the target image; wherein the undersampled imaging data comprises structural information of the target image in several directions and / or several orders.

3. The low-field magnetic resonance fast imaging method of claim 2, wherein the specific model of the convolution feature code is: Wherein x is the target image, is a set of convolution filter kernels, λ is a preset weight parameter, z i is the i-th convolution feature.

4. The low-field magnetic resonance fast imaging method of claim 1, wherein the preset algorithm is a first boosting plug-and-play iterative algorithm, and the algorithm framework is: x t+1 = x t + D σ (x noise - x t ), x noise = x k + l k-1 where x is the undersampled imaging data, D σ is a generalized denoising operator, x noise is noise information, and l is a multiplier term.

5. The low-field magnetic resonance fast imaging method of claim 1, wherein the preset algorithm is a second boosting plug-and-play iterative algorithm, and the algorithm framework is: x t+1 = D σ (x noise + x t ) - x t , x noise = x k + l k-1 where x is the undersampled imaging data, D σ is a generalized denoising operator, x noise is noise information, and l is a multiplier term.

6. A terminal device, characterized by comprising: The terminal device executes the low-field magnetic resonance fast imaging method of any one of claims 1-5, wherein the terminal device comprises: a reading module for reading undersampled imaging data about the target image; a construction module for constructing a convolution feature subspace projection regularization model based on low-field data characteristics; a reconstruction module for solving the convolution feature subspace projection regularization model in parallel based on a preset algorithm to obtain a high-quality magnetic resonance image.

7. A terminal device, characterized by comprising: The terminal device comprises a memory and a processor, wherein the memory and the processor are coupled; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the low-field magnetic resonance fast imaging method of any one of claims 1-5.

8. A computer storage medium, characterized in that The computer storage medium is used to store program data, which when executed by a processor, is used to implement the low-field magnetic resonance fast imaging method of any one of claims 1-5.

Images