Graph neural network-based magnetic resonance imaging permanent magnet design method and related apparatus

Abstract

The present application relates to the technical field of magnetic resonance imaging, and discloses a graph neural network-based magnetic resonance imaging permanent magnet design method and a related apparatus. The method comprises: acquiring a structural parameter of an initial magnetic resonance imaging permanent magnet, the structural parameter comprising an adjacency matrix and a parameter matrix; constructing a graph neural network model on the basis of the structural parameter of the initial magnetic resonance imaging permanent magnet; then, optimizing the parameter matrix on the basis of the graph neural network model and the adjacency matrix to obtain an optimized parameter matrix; and finally, adjusting the initial magnetic resonance imaging permanent magnet on the basis of the optimized parameter matrix to obtain a designed magnetic resonance imaging permanent magnet. In the present application, on the basis of graph data with a non-Euclidean structure consisting of an adjacency matrix and a parameter matrix, a magnetic resonance imaging permanent magnet is designed using a graph neural network model, with the objective of optimizing magnetic field uniformity, thereby effectively improving the magnetic field uniformity of the magnetic resonance imaging permanent magnet.

Description

Design Method and Related Devices for Magnetic Resonance Imaging Permanent Magnets Based on Graph Neural Networks

[0001] This application claims priority to Chinese Patent Application No. 2024117179934, filed on November 27, 2024, entitled “Design Method and Related Device for Magnetic Resonance Imaging Permanent Magnet Based on Graph Neural Network”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of magnetic resonance imaging technology, and in particular to a magnetic resonance imaging permanent magnet design method and related device based on graph neural networks. Background Technology

[0003] Magnetic resonance imaging (MRI), compared to medical imaging techniques such as X-ray and computed tomography (CT), offers advantages such as being radiation-free and providing excellent soft tissue imaging contrast, playing a significant role in medical diagnosis and the development of life sciences. The main magnet, as the most basic and crucial component of an MRI scanner, directly affects the quality of MRI images.

[0004] The development of magnetic resonance imaging (MRI) machines has primarily focused on superconducting technology. While this has achieved clearer imaging, the bulky solenoid-type main magnet system makes it difficult to conduct on-site diagnosis and monitoring near the patient. It also carries risks such as liquid helium leakage, burns from strong magnetic fields, and stimulation and damage to surrounding nerves. Furthermore, it is expensive and limited in number. These problems severely restrict the accessibility of MRI and fail to meet the rapidly growing demand for MRI testing in a deeply aging society.

[0005] In recent years, ultra-low field magnetic resonance imaging (MRI) technology has attracted increasing attention. Compared to solenoid magnets, the main magnets used in ultra-low field MRI scanners are generally made of rare-earth permanent magnet materials. Compared to solenoid magnets, permanent magnets have poorer uniformity of the main magnetic field, insufficient geometric accuracy, and severe image distortion. Existing permanent magnets are generally composed of multiple magnetic blocks arranged according to certain theories. To obtain a more uniform main magnetic field, researchers have optimized the arrangement of magnetic blocks. For example, Cooley et al. designed a Halbach-type magnet geometry model for head MRI, supporting three types of magnetic blocks: air, N52 neodymium iron boron (NdFeB) magnets, and N54 NdFeB magnets, and used a genetic algorithm to optimize the selection of the position of each magnetic block in the geometry model. Compared to the 137,870 ppm main magnetic field uniformity obtained when a single type of magnetic block was fully filled, Cooley et al. optimized the main magnetic field uniformity to 13,669 ppm. Reilly et al. designed a geometric model of the head magnet that supports optimization of the magnetic ring radius. They then used a genetic algorithm to iteratively adjust the magnetic ring radius, ultimately achieving a main magnetic field uniformity of 2400 ppm within a 20 cm spherical space using approximately 38 kg of N48 neodymium iron boron magnets. Building upon the main magnet design by Reilly et al., Tewari et al. further optimized the magnetization angle of the magnets within each magnetic ring using a genetic algorithm, improving the main magnet uniformity by approximately 18%.

[0006] However, in existing research on permanent magnet design, the magnet geometric model supports only a single degree of freedom for optimization, and the uniformity optimization algorithm is generally a genetic algorithm. This leads to the following limitations: (1) The single degree of freedom of the magnet geometric model limits the upper limit of the performance of the main magnetic field uniformity. Each magnet block can only be adjusted in a single dimension, and the space for improvement is limited; (2) Although multi-degree-of-freedom adjustment can be achieved through cascade optimization, the existing magnet geometric model still lacks the ability to jointly optimize multiple degrees of freedom, and still has certain limitations on the uniformity of the main magnetic field; (3) The genetic algorithm lacks the utilization of the geometric relationship information of the magnet blocks in the magnet, which leads to the limitation of the optimization algorithm performance. Summary of the Invention

[0007] The purpose of this application is to provide a design method and related device for magnetic resonance imaging permanent magnets based on graph neural networks. By performing multi-degree-of-freedom joint optimization of magnetic resonance imaging permanent magnets based on the geometric relationship of magnetic blocks, the magnetic field uniformity of magnetic resonance imaging permanent magnets can be effectively improved.

[0008] To achieve the above objectives, this application provides the following solution:

[0009] A method for designing permanent magnets for magnetic resonance imaging based on graph neural networks, comprising:

[0010] Obtain the structural parameters of the initial magnetic resonance imaging permanent magnet; the initial magnetic resonance imaging permanent magnet is composed of several magnetic blocks; the structural parameters include a parameter matrix and an adjacency matrix; the parameter matrix is ​​a matrix including the three-dimensional center coordinates and three-dimensional rotation angles of each magnetic block; the adjacency matrix is ​​a matrix indicating whether there is an adjacency relationship between any two magnetic blocks.

[0011] A graph neural network model is constructed based on the structural parameters of the initial magnetic resonance imaging permanent magnet.

[0012] Based on the graph neural network model and the adjacency matrix, the parameter matrix is ​​optimized with the goal of optimizing magnetic field uniformity, resulting in an optimized parameter matrix.

[0013] Based on the optimized parameter matrix, the initial magnetic resonance imaging permanent magnet is adjusted to obtain the designed magnetic resonance imaging permanent magnet.

[0014] This application also discloses a magnetic resonance imaging permanent magnet design device based on graph neural networks, including:

[0015] The structural parameter acquisition module is used to acquire the structural parameters of the initial magnetic resonance imaging permanent magnet; the initial magnetic resonance imaging permanent magnet is composed of several magnetic blocks; the structural parameters include an adjacency matrix and a parameter matrix; the parameter matrix is ​​a matrix including the three-dimensional center coordinates and three-dimensional rotation angles of each magnetic block; the adjacency matrix is ​​a matrix indicating whether there is an adjacency relationship between any two magnetic blocks.

[0016] The graph neural network model construction module is used to construct a graph neural network model based on the structural parameters of the initial magnetic resonance imaging permanent magnet.

[0017] The parameter optimization module is used to optimize the parameter matrix based on the graph neural network model and the adjacency matrix, with the goal of optimizing the magnetic field uniformity, to obtain an optimized parameter matrix.

[0018] The adjustment module is used to adjust the initial magnetic resonance imaging permanent magnet according to the optimized parameter matrix to obtain the designed magnetic resonance imaging permanent magnet.

[0019] This application also discloses a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the graph neural network-based magnetic resonance imaging permanent magnet design method.

[0020] This application also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the graph neural network-based magnetic resonance imaging permanent magnet design method.

[0021] This application also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the graph neural network-based magnetic resonance imaging permanent magnet design method.

[0022] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0023] This application provides a method and related apparatus for designing magnetic resonance imaging (MRI) permanent magnets based on graph neural networks. The method involves acquiring the structural parameters of an initial MRI permanent magnet, including an adjacency matrix and a parameter matrix. The parameter matrix contains the three-dimensional center coordinates and three-dimensional rotation angles of each magnetic block, while the adjacency matrix indicates whether any two magnetic blocks are adjacent. A graph neural network model is constructed based on these parameters. The parameter matrix is ​​then optimized using the graph neural network model and the adjacency matrix to obtain an optimized parameter matrix. Finally, the initial MRI permanent magnet is adjusted based on the optimized parameter matrix to obtain the designed MRI permanent magnet. In other words, this application uses graph data—the non-Euclidean structure of the adjacency matrix and parameter matrix of the initial MRI permanent magnet—and a graph neural network model to design an MRI permanent magnet with the goal of optimizing magnetic field uniformity. This achieves multi-degree-of-freedom joint optimization of the MRI permanent magnet based on the geometric relationships of the magnetic blocks, effectively improving the magnetic field uniformity of MRI permanent magnets constructed from discrete magnetic blocks.

[0024] Instruction manual illustrations

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 is a flowchart illustrating a magnetic resonance imaging permanent magnet design method based on graph neural networks according to an embodiment of this application.

[0027] Figure 2 is another flowchart illustrating a magnetic resonance imaging permanent magnet design method based on graph neural networks according to an embodiment of this application;

[0028] Figure 3(a) is a front view of an optimized Halbach-type permanent magnet provided in an embodiment of this application;

[0029] Figure 3(b) is a side view of an optimized Halbach-type permanent magnet provided in an embodiment of this application;

[0030] Figure 3(c) is a three-dimensional view of an optimized Halbach-type permanent magnet provided in an embodiment of this application;

[0031] Figure 4(a) is a three-dimensional field diagram of the magnetic field generated by the optimized Halbach-type permanent magnet provided in an embodiment of this application;

[0032] Figure 4(b) is a two-dimensional cross-sectional view of the magnetic field generated by the optimized Halbach-type permanent magnet provided in an embodiment of this application;

[0033] Figure 5(a) is a schematic diagram of the deflection angle distribution of the magnetic blocks in the magnetic ring with a frequency of 2 of the optimized Halbach-type permanent magnet provided in an embodiment of this application relative to the Halbach angle.

[0034] Figure 5(b) is a schematic diagram of the deflection angle distribution of the magnetic blocks in the magnetic ring with frequencies of 2 and 4 in an optimized Halbach-type permanent magnet according to an embodiment of this application, relative to the Halbach angle.

[0035] Figure 6(a) is a statistical histogram of the displacement of the magnetic block of the optimized Halbach-type permanent magnet in the x-axis direction according to an embodiment of this application;

[0036] Figure 6(b) is a statistical histogram of the displacement of the magnetic block of the optimized Halbach-type permanent magnet in the y-axis direction according to an embodiment of this application;

[0037] Figure 6(c) is a statistical histogram of the displacement of the magnetic block of the optimized Halbach-type permanent magnet in the z-axis direction according to an embodiment of this application;

[0038] Figure 7 is a schematic diagram of the functional modules of a magnetic resonance imaging permanent magnet design device based on graph neural network according to an embodiment of this application;

[0039] Figure 8 is a schematic diagram of the functional modules of a magnetic resonance imaging permanent magnet design device based on a graph neural network according to another embodiment of this application;

[0040] Figure 9 is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] The permanent magnet used in this application refers to the permanent magnet in a magnetic resonance imaging (MRI) machine. As the most basic and core component of an MRI machine, the permanent magnet needs to provide a sufficiently uniform main magnetic field to reduce image distortion and ensure spatial resolution and imaging quality. However, existing technologies typically use Euclidean data for magnet modeling, lacking an accurate description of the geometric relationships between magnetic blocks; the degrees of freedom supported by the magnet geometric model are relatively limited, restricting the optimization space; and the optimization algorithms are relatively traditional and have poor performance. To address these issues, this application proposes a method for designing permanent magnets for MRI based on graph neural networks. It uses non-Euclidean graph data for permanent magnet modeling, constructing a magnet geometric model that supports joint optimization of multiple degrees of freedom. Then, it utilizes a graph neural network in the high-dimensional solution space provided by the magnet geometric model, with uniformity as the primary objective, to achieve intelligent design of the permanent magnet. This application can significantly optimize the magnetic field uniformity of permanent magnets constructed from discrete magnetic blocks.

[0043] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] Example 1

[0045] As shown in Figure 1, this embodiment discloses a magnetic resonance imaging permanent magnet design method based on graph neural networks, which includes steps 101 to 104. Wherein:

[0046] Step 101: Obtain the structural parameters of the initial magnetic resonance imaging permanent magnet; the initial magnetic resonance imaging permanent magnet is composed of several magnetic blocks; the structural parameters include a parameter matrix and an adjacency matrix; the parameter matrix is ​​a matrix including the three-dimensional center coordinates and three-dimensional rotation angle (i.e., three-dimensional pose) of each magnetic block; the adjacency matrix is ​​a matrix indicating whether there is an adjacency relationship between any two magnetic blocks.

[0047] This application proposes a magnet geometry model that supports multi-degree-of-freedom optimization (i.e., the parameter matrix and adjacency matrix of a magnetic resonance imaging permanent magnet) to achieve joint and continuous optimization of multiple degrees of freedom during the subsequent training of the graph neural network.

[0048] The multiple degrees of freedom are the three-dimensional center coordinates and three-dimensional rotation angles of the magnetic block (i.e., six degrees of freedom). Each small magnetic block can rotate around its center in three orthogonal directions, and there is space between the initial positions of each magnetic block to support the displacement of the small magnetic block in the three orthogonal directions.

[0049] For N magnetic blocks with d0 attributes (corresponding to d0 degrees of freedom), the parameter matrix can be obtained. Use adjacency matrix This is used to describe the spatial relationships between the magnetic blocks. This process is called permanent magnet modeling with non-Euclidean data. Considering that magnetic blocks that are closer together have stronger spatial correlation, if magnetic block i and magnetic block j are close to each other, the corresponding value A(i,j) in the adjacency matrix is ​​assigned a larger value, and vice versa.

[0050] Step 102: Construct a graph neural network model (GNN) based on the structural parameters of the initial magnetic resonance imaging permanent magnet.

[0051] Step 103: Based on the graph neural network model and the adjacency matrix, optimize the parameter matrix with the goal of optimizing the magnetic field uniformity to obtain the optimized parameter matrix.

[0052] Step 104: Adjust the initial magnetic resonance imaging permanent magnet according to the optimized parameter matrix to obtain the designed magnetic resonance imaging permanent magnet.

[0053] Step 103 specifically includes:

[0054] Step 103.1, set constraints; the constraints are the movement threshold of the magnetic blocks; the movement threshold is a value not greater than the gap between the magnetic blocks at the initial moment.

[0055] To ensure that the magnetic blocks do not interfere with each other during the optimization process, the movement threshold of the magnetic blocks in the three orthogonal directions is limited to no more than the gap between the initial positions of the magnetic blocks in the magnet geometry model.

[0056] Step 103.2: Optimize the parameter matrix based on the graph neural network model, the constraints, and the adjacency matrix to obtain the optimized parameter matrix.

[0057] Step 103.2 specifically involves: inputting the adjacency matrix and parameter matrix into the graph neural network model for forward propagation, and truncating the parameter matrix output by the graph neural network model according to the constraints to obtain the optimized parameter matrix.

[0058] The above process will be described in more detail below:

[0059] Step ①: For the parameter matrix X and the adjacency matrix A, input the weight parameter W. (1) The first graph convolutional layer produces a d1-dimensional vector. d1 is the weight parameter W (1) The dimensions and length.

[0060] Step ②, for the parameter matrix H (1) Given an adjacency matrix A, its input weight parameter is W. (2) The second graph convolutional layer produces a vector of dimension d2. d2 is the weight parameter W (2) The dimensions and length.

[0061] Step ③, for the parameter matrix H (2) Given an adjacency matrix A, input weight parameters W. (3) The third graph convolutional layer produces a vector of dimension d3. d3 is the weight parameter W (3) The dimensions and length.

[0062] Step ④, for the parameter matrix H (3) Given an adjacency matrix A, input weight parameters W. (4) The fourth graph convolutional layer produces a vector of dimension d4. d4 is the weight parameter W (4) The dimensions and length.

[0063] Among them, X to H (4) The dimensions maintain a trend of first increasing and then decreasing: d0-d1↑-d2↑-d3↑-d4↓.

[0064] Step 5, H (4) Take a fully connected layer (FCL) as input and output a parameter matrix with the same dimension as d0.

[0065] Step 6: Truncate X* according to the constraints to obtain the optimized parameter matrix.

[0066] Step 103.3: Determine the magnetic field uniformity of the preset region based on the magnetic resonance imaging permanent magnet corresponding to the optimized parameter matrix.

[0067] Step 103.4: Based on the magnetic field uniformity of the preset region, the parameters of the graph neural network model are updated using the gradient backpropagation algorithm to obtain the updated graph neural network model.

[0068] Step 103.5: Update the graph neural network model to the updated graph neural network model and update the parameter matrix to the optimized parameter matrix, then return to step 103.2; until the magnetic field uniformity of the preset region converges, the optimized parameter matrix is ​​obtained.

[0069] In other words, the general process of optimizing the main magnetic field uniformity based on GNN is as follows: ① Input the construction parameters (parameter matrix) of the non-Euclidean permanent magnet structure and the corresponding adjacency matrix into the GNN. The GNN constructs the spatial relationship between the magnetic blocks and mines their deep features and dependencies, outputting the optimized permanent magnet construction parameters (parameter matrix); ② Calculate the three-dimensional magnetic field and its corresponding uniformity based on the output permanent magnet construction parameters (parameter matrix); ③ Use the magnetic field uniformity as a loss function, perform gradient backpropagation, and update the parameters of the GNN; ④ Replace the original input with the permanent magnet construction parameters output by the GNN in step ①; ⑤ Repeat steps ① to ④ until the uniformity converges, finally obtaining the optimal permanent magnet construction parameters and completing the magnetic field uniformity optimization. Finally, complete the permanent magnet design based on the permanent magnet construction parameters output by the GNN.

[0070] The loss value is the difference between the magnetic field uniformity and 0, so the magnetic field uniformity value itself is the loss value.

[0071] Step 103.3 specifically includes:

[0072] Step 103.3.1: Based on the Biot-Savart law and the magnetic dipole model, calculate the magnetic field generated in the preset region by the magnetic block in the magnetic resonance imaging permanent magnet corresponding to the optimized parameter matrix.

[0073] Specifically, according to the Biot-Savart Law, the magnetic field generated by a current-carrying conductor at a point in space is:

[0074] Where B is the three-dimensional vector magnetic field, and μ0 is the free permeability (4π×10⁻⁶). -7 T·m / A), I is the current in the conductor, dl is the vector representing a small segment of the conductor, and r is the vector distance from the conductor to the observation point, with unit vector distance. By integrating along the path C of the conductor, the magnetic field generated by the conductor at the observation point can be obtained.

[0075] To facilitate the calculation of the three-dimensional magnetic field generated by the magnetic block, the magnetic block can be regarded as an equivalent tiny toroidal current magnetic dipole, and its corresponding magnetic field is:

[0076] Where B(r) is the magnetic field generated by the magnetic dipole at position r, and m = M·V, where M and V represent the magnetization and volume of the magnetic block, respectively. This formula is used to calculate the field diagram of a single magnetic block.

[0077] The calculation process can be assisted by simulation. First, a permanent magnet model is established in the simulation software. In establishing the permanent magnet model, the pose and polarization direction of the magnetic block are determined as follows: a single magnetic block is rotated around the axis of the field of view (FOV) and then around the axis of its own center of gravity. Based on the pose of the single magnetic block, the corresponding poses of the other magnetic blocks on the same magnetic ring are determined. Then, through the same affine transformation as the magnetic block pose transformation, the polarization direction of the magnetic block is matched with its pose.

[0078] A magnetic ring is composed of several discrete magnetic blocks. Here, we first determine a single magnetic block, and then obtain the remaining magnetic blocks through operations similar to "copy" and "paste". The ring arrangement of several magnetic blocks forms a magnetic ring, as shown in Figures 3(a)-3(c).

[0079] The field diagram of all magnetic rings is calculated using formula (2), and the main magnetic field in the field of view is obtained by linear superposition.

[0080] For example, the selected magnetic block has a side length of 2.54 cm and is of neodymium iron boron (N57) grade. Simultaneously, to reduce the quantization error of the affine transformation, the dimension of the three-dimensional field map within the field of view is expanded to 100. 3 In this embodiment, the FOV is a spherical region with a diameter of 45 cm.

[0081] Step 103.3.2: Linearly superimpose the magnetic fields generated by each magnetic block in the preset area to obtain the magnetic field generated by the magnetic resonance imaging permanent magnet in the preset area corresponding to the optimized parameter matrix.

[0082] Step 103.3.3: Determine the magnetic field uniformity of the preset region based on the magnetic field generated by the magnetic resonance imaging permanent magnet in the preset region corresponding to the optimized parameter matrix.

[0083] The types of permanent magnets used in magnetic resonance imaging include Halbach type, C type and H type.

[0084] When the magnetic resonance imaging permanent magnet is of the Halbach type, the central magnetic ring of the magnetic resonance imaging permanent magnet has a large aperture, while the two side magnetic rings have small apertures; the magnetic ring is a ring formed by arranging several magnetic blocks.

[0085] The adjacency matrix A has a size of N×N, where N is the number of magnetic blocks in the magnetic resonance imaging permanent magnet. When magnetic block i and magnetic block j in the magnetic resonance imaging permanent magnet are adjacent, the value of the corresponding element A(i,j) in the adjacency matrix is ​​1; when magnetic block i and magnetic block j in the magnetic resonance imaging permanent magnet are not adjacent, the value of the corresponding element A(i,j) in the adjacency matrix is ​​0.

[0086] For example, if the magnet's geometric model has 2594 magnet block positions, when modeling a permanent magnet using non-Euclidean data, a parameter matrix can be obtained for the 2594 magnet blocks with 6 attributes (degrees of freedom). Use adjacency matrix This describes the spatial relationship between the magnetic blocks. If magnetic block i is adjacent to any magnetic block j, then the corresponding value A(i,j) in the adjacency matrix is ​​assigned a value of 1, and the remaining elements in row i are assigned a value of 0.

[0087] In this embodiment, the appearance of the permanent magnet is shown in Figures 3(a)-3(c). The permanent magnet can generate magnetic fields up to order 10, and its three-dimensional field diagram and corresponding two-dimensional cross-sectional diagram are shown in Figures 4(a)-4(b). The deflection angle of the magnetic block in the optimized permanent magnet ring relative to the Halbach angle has multiple frequency components. Figure 5(a) shows the deflection angle distribution of the magnetic ring with frequency 2, and Figure 5(b) shows the deflection angle distribution of the magnetic ring containing both frequencies 2 and 4. Here, the frequency refers to the number of times the rotation angle of the magnetic block in the corresponding magnetic ring periodically repeats within the azimuth angle range of 0 to 360 degrees. The overall displacement of the magnetic block in the optimized permanent magnet is basically symmetrical with the deflection angle distribution. Figures 6(a)-6(c) show the statistical histograms of the magnetic block displacement.

[0088] This application also provides an application scenario in which the above-described graph neural network-based permanent magnet design method for magnetic resonance imaging (MRI) is applied. Specifically, the graph neural network-based permanent magnet design method for MRI provided in this embodiment can be applied in the design scenario of MRI permanent magnets. The MRI permanent magnet design scenario includes an initial MRI permanent magnet determination stage and an initial MRI permanent magnet optimization stage. After the initial MRI permanent magnet is determined in the initial MRI permanent magnet determination stage, the initial MRI permanent magnet optimization stage is entered. In the initial MRI permanent magnet optimization stage, the structural parameters of the initial MRI permanent magnet are optimized, and the initial MRI permanent magnet is adjusted according to the optimized structural parameters. The graph neural network-based MRI permanent magnet design method provided in this embodiment belongs to the initial MRI permanent magnet optimization stage. Specifically, based on the structural parameters of the initial magnetic resonance imaging permanent magnet, a graph neural network model is constructed, and the parameter matrix is ​​optimized based on the graph neural network model and the adjacency matrix to obtain the optimized parameter matrix. Based on the optimized parameter matrix, the initial magnetic resonance imaging permanent magnet is adjusted to obtain the designed magnetic resonance imaging permanent magnet.

[0089] Example 2

[0090] Based on the same inventive concept, this application also discloses a graph neural network-based magnetic resonance imaging permanent magnet design device for implementing the above-mentioned graph neural network-based magnetic resonance imaging permanent magnet design method. The solution provided by this device is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more graph neural network-based magnetic resonance imaging permanent magnet design device embodiments provided below can be found in the limitations of the graph neural network-based magnetic resonance imaging permanent magnet design method described above, and will not be repeated here.

[0091] As shown in Figure 7, a magnetic resonance imaging permanent magnet design device based on graph neural networks includes:

[0092] The structural parameter acquisition module M01 is used to acquire the structural parameters of the initial magnetic resonance imaging permanent magnet; the initial magnetic resonance imaging permanent magnet is composed of several magnetic blocks; the structural parameters include an adjacency matrix and a parameter matrix; the parameter matrix is ​​a matrix including the three-dimensional center coordinates and three-dimensional rotation angles of each magnetic block; the adjacency matrix is ​​a matrix indicating whether there is an adjacency relationship between any two magnetic blocks.

[0093] The graph neural network model building module M02 is used to build a graph neural network model based on the structural parameters of the initial magnetic resonance imaging permanent magnet.

[0094] The parameter optimization module M03 is used to optimize the parameter matrix based on the graph neural network model and the adjacency matrix, with the goal of optimizing the magnetic field uniformity, to obtain an optimized parameter matrix.

[0095] The adjustment module M04 is used to adjust the initial magnetic resonance imaging permanent magnet according to the optimized parameter matrix to obtain the designed magnetic resonance imaging permanent magnet.

[0096] Example 3

[0097] As shown in Figure 8, this application also discloses another magnetic resonance imaging permanent magnet design device based on graph neural networks, including:

[0098] Acquire module M11, which is used to construct data for a non-Euclidean structure to describe the permanent magnet.

[0099] The output module M12 is used to input the non-Euclidean structure data into the graph neural network model and output permanent magnet parameters after optimization; the permanent magnet parameters include the three-dimensional pose of each magnetic block in the magnet geometry model.

[0100] The determination module M13 is used to determine the three-dimensional spatial structure of the permanent magnet based on the permanent magnet parameters.

[0101] Compared with the prior art, this application has the following advantages:

[0102] 1. This application proposes a permanent magnet modeling method based on non-Euclidean structure data, which can effectively model and deeply explore the spatial relationship of magnetic blocks through adjacency matrix, thereby improving the performance of the final designed magnet.

[0103] 2. This application proposes a magnet geometry model that supports multi-degree-of-freedom optimization. This model can jointly optimize the continuous three-dimensional pose parameters of the magnet. The optimization space that this model can achieve covers the entire optimization space that existing single-degree-of-freedom optimization methods can achieve. By greatly expanding the optimization space, the performance ceiling of magnet optimization is significantly improved.

[0104] 3. This application proposes the application of Graph Neural Networks (GNNs) to permanent magnet design, pioneering the introduction of GNN technology into the design and optimization process of ultra-low field magnetic resonance imaging (ULF MRI) permanent magnets, filling a research gap in this field. It utilizes adjacency matrices to model the geometric relationships of non-Euclidean magnetic blocks and achieves multi-dimensional joint optimization in a multi-degree-of-freedom optimization space, overcoming the difficulties of traditional genetic algorithms in effectively exploring multi-degree-of-freedom optimization spaces and limiting the performance of the designed magnets.

[0105] 4. The permanent magnet optimized in this application can generate a uniform magnetic field of up to 10th order, which is significantly better than the permanent magnet designed by existing algorithms.

[0106] Example 4

[0107] A computer device, which can be a server or a terminal, has an internal structure as shown in Figure 9. The computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database stores the structural parameters of permanent magnets. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a magnetic resonance imaging permanent magnet design method based on a graph neural network.

[0108] Those skilled in the art will understand that the structure shown in FIG9 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0109] Example 5

[0110] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0111] Example 6

[0112] A computer program product includes a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0114] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0115] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0116] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0117] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for designing permanent magnets for magnetic resonance imaging based on graph neural networks, characterized in that, The magnetic resonance imaging permanent magnet design method based on graph neural networks includes: Obtain the structural parameters of the initial magnetic resonance imaging permanent magnet; the initial magnetic resonance imaging permanent magnet is composed of several magnetic blocks; the structural parameters include a parameter matrix and an adjacency matrix; the parameter matrix is ​​a matrix including the three-dimensional center coordinates and three-dimensional rotation angles of each magnetic block; the adjacency matrix is ​​a matrix indicating whether there is an adjacency relationship between any two magnetic blocks; Based on the structural parameters of the initial magnetic resonance imaging permanent magnet, a graph neural network model is constructed. Based on the graph neural network model and the adjacency matrix, with the goal of optimizing magnetic field uniformity, the parameter matrix is ​​optimized to obtain an optimized parameter matrix; Based on the optimized parameter matrix, the initial magnetic resonance imaging permanent magnet is adjusted to obtain the designed magnetic resonance imaging permanent magnet.

2. The magnetic resonance imaging permanent magnet design method based on graph neural networks according to claim 1, characterized in that, Based on the graph neural network model and the adjacency matrix, with the goal of optimizing magnetic field uniformity, the parameter matrix is ​​optimized to obtain an optimized parameter matrix, specifically including: Set constraints; the constraints are the movement thresholds of the magnetic blocks; the movement thresholds are values ​​not greater than the gap between the magnetic blocks at the initial moment; Based on the graph neural network model, the constraints, and the adjacency matrix, the parameter matrix is ​​optimized to obtain the optimized parameter matrix. Based on the magnetic resonance imaging permanent magnet corresponding to the optimized parameter matrix, the magnetic field uniformity of the preset area is determined. Based on the magnetic field uniformity of the preset region, the parameters of the graph neural network model are updated using the gradient backpropagation algorithm to obtain the updated graph neural network model. The graph neural network model is updated to the updated graph neural network model, and the parameter matrix is ​​updated to the optimized parameter matrix. The process returns to the previous step, where the parameter matrix is ​​optimized based on the graph neural network model, the constraints, and the adjacency matrix to obtain the optimized parameter matrix. This process continues until the magnetic field uniformity of the preset region converges, resulting in the optimized parameter matrix.

3. The magnetic resonance imaging permanent magnet design method based on graph neural networks according to claim 2, characterized in that, Based on the magnetic resonance imaging permanent magnet corresponding to the optimized parameter matrix, the magnetic field uniformity of the preset region is determined, specifically including: Based on the Biot-Savart law and the magnetic dipole model, the magnetic field generated by the magnetic block in the magnetic resonance imaging permanent magnet in the preset region is calculated according to the optimized parameter matrix; The magnetic fields generated by each magnetic block in the preset area are linearly superimposed to obtain the magnetic field generated by the magnetic resonance imaging permanent magnet in the preset area corresponding to the optimized parameter matrix. The magnetic field uniformity of the preset region is determined based on the magnetic field generated by the magnetic resonance imaging permanent magnet in the preset region corresponding to the optimized parameter matrix.

4. The magnetic resonance imaging permanent magnet design method based on graph neural networks according to claim 1, characterized in that, The types of permanent magnets used in magnetic resonance imaging include Halbach type, C type and H type.

5. The magnetic resonance imaging permanent magnet design method based on graph neural networks according to claim 1, characterized in that, When the magnetic resonance imaging permanent magnet is of the Halbach type, the central magnetic ring of the magnetic resonance imaging permanent magnet has a large aperture, while the two side magnetic rings have small apertures; the magnetic ring is a ring formed by arranging several magnetic blocks.

6. The magnetic resonance imaging permanent magnet design method based on graph neural networks according to claim 1, characterized in that, The adjacency matrix A has a size of N×N, where N is the number of magnetic blocks in the magnetic resonance imaging permanent magnet. When magnetic block i and magnetic block j in the magnetic resonance imaging permanent magnet are adjacent, the value of the corresponding element A(i,j) in the adjacency matrix is ​​1; when magnetic block i and magnetic block j in the magnetic resonance imaging permanent magnet are not adjacent, the value of the corresponding element A(i,j) in the adjacency matrix is ​​0.

7. A magnetic resonance imaging permanent magnet design device based on graph neural networks, characterized in that, The magnetic resonance imaging permanent magnet design device based on graph neural network includes: A structural parameter acquisition module is used to acquire the structural parameters of an initial magnetic resonance imaging permanent magnet; the initial magnetic resonance imaging permanent magnet is composed of several magnetic blocks; the structural parameters include an adjacency matrix and a parameter matrix; the parameter matrix is ​​a matrix including the three-dimensional center coordinates and three-dimensional rotation angles of each magnetic block; the adjacency matrix is ​​a matrix indicating whether there is an adjacency relationship between any two magnetic blocks; The graph neural network model construction module is used to construct a graph neural network model based on the structural parameters of the initial magnetic resonance imaging permanent magnet. The parameter optimization module is used to optimize the parameter matrix based on the graph neural network model and the adjacency matrix, with the goal of optimizing the magnetic field uniformity, to obtain an optimized parameter matrix. The adjustment module is used to adjust the initial magnetic resonance imaging permanent magnet according to the optimized parameter matrix to obtain the designed magnetic resonance imaging permanent magnet.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the magnetic resonance imaging permanent magnet design method based on graph neural networks as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the magnetic resonance imaging permanent magnet design method based on graph neural networks as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the magnetic resonance imaging permanent magnet design method based on graph neural networks as described in any one of claims 1-6.

Images