A method, system, and apparatus for imaging and reconstruction using magnetic nanoparticles.

Abstract

This invention belongs to the field of image processing, specifically relating to a method, system, and apparatus for imaging and reconstructing magnetic nanoparticles. This invention solves the problem of blurriness and inaccuracy in X-Space image reconstruction caused by neglecting the relaxation effect of particles when using the Langevin model to describe the magnetization process of magnetic particles under the adiabatic assumption. The invention includes: acquiring a one-dimensional MPI voltage signal of the image to be reconstructed and inputting it into the generator of a trained generative adversarial network (GAN) model to obtain the reconstructed magnetic particle image; wherein the GAN model includes a generator and a discriminator. This invention uses a GAN model where the generator and discriminator compete against each other until an equilibrium is reached, at which point the samples generated by the generator can approximate the distribution of real samples. Simultaneously, conditional GAN ​​adds conditional information to the original GAN, realizing the reconstruction of a two-dimensional image from a one-dimensional signal, effectively ensuring the quality of the generated image.

Description

Technical Field

[0001] This invention belongs to the field of image processing, and specifically relates to a method, system and device for imaging reconstruction using magnetic nanoparticles. Background Technology

[0002] Magnetic particle imaging (MPI) is a novel imaging technique developed by Gleich et al. in 2005. It is a tracer-based imaging method that uses a gradient magnetic field to detect the distribution of superparamagnetic iron oxide nanoparticles (SPIONs) in vivo. It has high sensitivity and resolution and is not limited by scanning depth.

[0003] Currently, there are two main types of common MPI reconstruction algorithms: the system matrix method and the physical model method. The system matrix is ​​a discrete representation of the system function of an MPI device. It represents the harmonic components of the induced voltage signal generated by magnetic nanoparticles at different spatial locations, mapping the signal spectrum of the induced voltage to the spatial concentration distribution. Reconstruction using the system matrix method mainly involves two steps: obtaining the system matrix and solving the system's linear equations. The system matrix can be obtained through calibration scanning or physical modeling. Calibration scanning obtains the system matrix by measuring the signal at each pixel location of the sample, resulting in relatively accurate reconstruction, but the measurement process is time-consuming and labor-intensive, and recalibration is necessary when scanning conditions change. Physical modeling describes the mapping from the magnetic particle concentration distribution to the voltage signal using a physical model. While this avoids complex measurement processes, it cannot guarantee the accuracy of the obtained matrix. In solving the system's linear equations, due to factors such as measurement accuracy and external noise, the solved linear equations are often unsteady, thus requiring regularization to constrain the solution. Different regularization methods have been successfully used in MPI reconstruction, but these methods are based on assumptions about the data, and therefore the reconstruction results may be poor.

[0004] Another reconstruction algorithm is the X-Space method. It reconstructs the image using a point spread function and a voltage signal, and employs a two-step velocity compensation method to grid the received signal, converting it into magnetic particle concentration distribution information. The X-Space method avoids the complex process of constructing a system matrix, making imaging modular and enabling rapid image reconstruction and convolution, significantly improving the imaging speed of MPI. However, this method uses the Langevin model to describe the magnetization process of magnetic particles under the adiabatic assumption, neglecting the particle relaxation effect, leading to blurring and inaccuracies in the reconstructed image.

[0005] To address the shortcomings of traditional MPI reconstruction algorithms, some studies have applied deep neural networks to MPI reconstruction. However, these applications are mostly used to accelerate the calibration or optimization regularization process of the system matrix, and cannot further improve the reconstruction efficiency of MPI.

[0006] Generative Adversarial Networks (GANs), proposed in 2014, are image generation networks consisting of a generator and a discriminator. The generator learns complex data distributions from real images to generate near-realistic images, while the discriminator distinguishes between real and generated images. During training, the generator aims to generate samples that can fool the discriminator, while the discriminator aims to successfully identify real and fake samples. These two forces compete against each other until an equilibrium is reached, at which point the generator's samples approximate the distribution of real samples. Conditional GANs add conditional information to the original GAN, making the generated images controlled by the input signal, thus achieving the reconstruction of two-dimensional images from one-dimensional signals.

[0007] The Transformer, proposed in 2017, is based on a multi-head self-attention mechanism. In image processing tasks, it expands the receptive field of the model, making it easier for neural networks to learn global connections within an image. Using the Transformer structure as both the generator and discriminator in a GAN effectively ensures the quality of generated images.

[0008] Based on this, the present invention provides a method, system and apparatus for imaging reconstruction using magnetic nanoparticles. Summary of the Invention

[0009] To address the aforementioned problems in the prior art, namely the use of the Langevin model to describe the magnetization process of magnetic particles under the adiabatic assumption, which neglects the relaxation effect of the particles and leads to blurring and inaccuracy in the X-Space method reconstructed images, this invention provides a method, system, and apparatus for magnetic nanoparticle imaging reconstruction.

[0010] In one aspect, the present invention provides a method for imaging reconstruction using magnetic nanoparticles, the method comprising:

[0011] The one-dimensional MPI voltage signal of the image to be reconstructed is acquired as the input signal;

[0012] The input signal is input into the generator of the trained generative adversarial network model to obtain the reconstructed magnetic particle image;

[0013] The generative adversarial network model includes a generator and a discriminator.

[0014] In some preferred embodiments, the generator is based on a multilayer perceptron, an encoding module, and a weighting module connected in sequence. The encoding module includes multiple upsampling units and multiple Transformer blocks, with each upsampling unit located between two Transformer blocks.

[0015] The Transformer block is constructed based on a first normalization layer, a multi-head self-attention layer, a first summing unit, a second normalization layer, a multilayer perceptron, and a second summing unit connected in sequence.

[0016] The input of the first normalization layer is connected to the output residual of the multi-head self-attention layer, and the output of the first summing unit is connected to the output residual of the multilayer perceptron.

[0017] In some preferred embodiments, the one-dimensional MPI voltage signal is obtained by:

[0018] Step S110: Inject a magnetic nanoparticle solution of a fixed concentration into a container;

[0019] Step S120: Place the container on a precisely movable displacement stage, move the displacement stage sequentially in the X-Y-Z directions, and simultaneously start the MPI device to scan the container to obtain an initial one-dimensional MPI voltage signal U(t).

[0020] Step S130: Repeat step S120 n times to obtain n initial one-dimensional MPI voltage signals U(t) and n system matrices A. Calculate the average value An of the n matrices A. c :

[0021] A c = (A1 + A2 ... + A n ) / n

[0022] Step S140, for the Perform interpolation to obtain Where N and M are dimensions;

[0023] Step S150, based on the image to be reconstructed Xr, Gaussian noise m, and the... The one-dimensional MPI voltage signal U(t) is obtained:

[0024]

[0025] In some preferred embodiments, the generative adversarial network model is trained using the following method:

[0026] Step S210: Construct a grayscale image of the concentration distribution of magnetic particles in the scanning space as a dataset. Use this dataset as the image to be reconstructed, Xr, and ground truth labels. Use a portion of the dataset as training data. Input the one-dimensional MPI voltage signal from the training data into the generator in the pre-constructed generative adversarial network model to obtain the reconstructed image Xr. f ;

[0027] Step S220, based on the reconstructed image X f The corresponding truth labels are used to obtain the loss value of the generator through the pre-constructed generator loss function, and the parameters of the generator are then corrected.

[0028] Step S230: Compare the truth labels with the reconstructed image X f The input is given to the discriminator to determine the reconstructed image X. f To determine whether something is true or false, a judgment result is obtained;

[0029] Step S240: Combining the discrimination result, the loss value corresponding to the discriminator is obtained through the pre-constructed discriminator loss function, and the parameters of the discriminator are corrected.

[0030] Step S250, repeat steps S210-S240 until a trained generative adversarial network model is obtained.

[0031] In some preferred embodiments, the discriminator is based on an image segmentation module, a linear weighting module, and multiple Transformer blocks connected in sequence;

[0032] The image segmentation module is used to segment the reconstructed image X f The image is divided into a set number of blocks of the same size; the linear weighting module is used to convert the image of the set number of blocks into a vector with an embedded set dimension.

[0033] In some preferred embodiments, the loss value corresponding to the generator is calculated as follows:

[0034] Step S221, calculate the reconstructed image X f =G(v) and the mean absolute error L of each pixel in the image to be reconstructed Xr MAE :

[0035] L MAE (G,Xr)=E(||Xr-G(v)||);

[0036] Step S222, based on the mean absolute error L MAE Calculate the loss value L generated by it. G :

[0037] L G =-E(D(G(v)))+λL MAE (G,Xr);

[0038] Wherein, λ is a parameter that adjusts the proportion of the two losses in the generator, D represents the discriminator's judgment process on the reconstructed image, and E represents the average value of the discriminator's output.

[0039] In some preferred embodiments, the loss value corresponding to the discriminator is calculated as follows:

[0040] L D =-E(max(0,1-D(x)))+E(max(0,D(1+G(v)))).

[0041] Among them, L D This represents the loss value corresponding to the discriminator.

[0042] In another aspect, the present invention provides a system for imaging reconstruction using magnetic nanoparticles, the system comprising:

[0043] The signal acquisition module is configured to acquire a one-dimensional MPI voltage signal of the image to be reconstructed as an input signal.

[0044] The training module is configured to input the input signal into the generator of the trained generative adversarial network model to obtain the reconstructed magnetic particle image.

[0045] The generative adversarial network model includes a generator and a discriminator.

[0046] A third aspect of the invention provides a storage device storing a plurality of programs adapted for loading and execution by a processor of a method for magnetic nanoparticle imaging reconstruction.

[0047] A fourth aspect of the invention provides a processing apparatus comprising a processor and a storage device; the processor being adapted to execute various programs; the storage device being adapted to store multiple programs; the programs being adapted to be loaded and executed by the processor to realize a method for magnetic nanoparticle imaging reconstruction.

[0048] The beneficial effects of this invention are:

[0049] 1. Introducing the transformer structure into the generator and discriminator makes it easier for the model to learn the global relationships of the image. This makes it easier for the generator to pay attention to the overall effect of the image when generating it, and the discriminator can more quickly distinguish between real and fake images from a global perspective, thereby improving the model performance.

[0050] 2. During the acquisition of the MPI voltage signal, the system matrix was measured multiple times, averaged, and Gaussian noise was added to make the simulated system matrix closer to the real system matrix; interpolation was used on the system matrix to avoid ill-conditioned problems in solving the inverse problem.

[0051] 3. This invention utilizes a generative adversarial network (GAN) model, where the generator and discriminator compete against each other until an equilibrium is reached. At this point, the samples generated by the generator can closely approximate the distribution of real samples. Simultaneously, conditional GAN ​​adds conditional information to the original GAN, making the images generated by the GAN controlled by the input signal, thereby achieving the reconstruction of a two-dimensional image from a one-dimensional signal and effectively ensuring the quality of the generated image. Attached Figure Description

[0052] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0053] Figure 1 This is a flowchart of a magnetic nanoparticle imaging reconstruction method according to the present invention;

[0054] Figure 2 This is a flowchart of the training and testing process of a magnetic nanoparticle imaging reconstruction method according to the present invention;

[0055] Figure 3 This is a schematic diagram of a dataset for a magnetic nanoparticle imaging reconstruction method according to the present invention;

[0056] Figure 4 This is a generator structure diagram of a magnetic nanoparticle imaging reconstruction method according to the present invention;

[0057] Figure 5 This is a diagram of the discriminator structure of a magnetic nanoparticle imaging reconstruction method according to the present invention;

[0058] Figure 6 This is a Transformer block structure diagram of a magnetic nanoparticle imaging reconstruction method according to the present invention;

[0059] Figure 7 This is a schematic diagram of the bicubic interpolation upsampling method and pixel reconstruction method of the magnetic nanoparticle imaging reconstruction method of the present invention;

[0060] Figure 8 This is a schematic diagram of the multi-head self-attention calculation range of a magnetic nanoparticle imaging reconstruction method according to the present invention;

[0061] Figure 9 This is a schematic diagram of the structure of a computer system used to implement the methods, systems, and apparatus embodiments of this application. Detailed Implementation

[0062] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0063] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0064] like Figure 1-9 As shown, see Figure 1 , Figure 2 , Figure 3 The first embodiment of the present invention provides a method for imaging reconstruction using magnetic nanoparticles, the method comprising:

[0065] The one-dimensional MPI voltage signal of the image to be reconstructed is acquired as the input signal;

[0066] The input signal is input into the generator of the trained generative adversarial network model to obtain the reconstructed magnetic particle image;

[0067] The generative adversarial network model includes a generator and a discriminator.

[0068] To more clearly illustrate the method for magnetic nanoparticle imaging reconstruction according to the present invention, the following is in conjunction with... Figure 1 , 2 3. The embodiments of the present invention will be described in detail below.

[0069] The one-dimensional MPI voltage signal of the image to be reconstructed is acquired as the input signal;

[0070] This invention uses MATLAB to generate 260,000 binary images of size 128×128 pixels, which can be evenly divided into three types of patterns: parallel vertically arranged bar charts, parallel horizontally arranged bar charts, and bar charts with cross-shaped arrangements in two directions. The vertical bars in each image vary in number, width, spacing, and length. The MPI simulation software MPIRF is used to simulate the one-dimensional MPI voltage signal corresponding to each binary image generated in the receiving coil, and then normalized as the input signal for the model, with a dimension of 1×1632. Of the 260,000 paired data, 240,000 pairs are used as the training set, and the remaining 20,000 pairs are used as the test set.

[0071] Two-dimensional image generation:

[0072] We plan to construct a dataset containing 600,000 128×128 dimensional grayscale images. The patterns in the dataset are divided into 6 types, with 100,000 images of each type, representing the concentration distribution of magnetic particles in the scanning space. Figure 3 (a) represents a bar distribution, with the bar spacing and width increasing from 1 mm to 10 mm in increments of 1 mm; the number of bars increases from 1 until it does not exceed the image boundary. Figure 3 (b) Represents a circular distribution, with the ring radius increasing from 10 mm to 64 mm in increments of 1 mm, and the difference between the outer and inner ring diameters increasing from 1 mm to 10 mm in increments of 1 mm. Figure 3 (c) and (d) represent triangular and square distributions, respectively, with the inner and outer diameters taking the same values ​​as the annular distribution; Figure 3 (e) is a random combination of the four shapes mentioned above; Figure 3 (f) represents a random letter shape.

[0073] The one-dimensional MPI voltage signal is obtained by the following method:

[0074] Step S110: Inject a magnetic nanoparticle solution of a fixed concentration into a container;

[0075] Step S120: Place the container on a precisely movable displacement stage, move the displacement stage sequentially in the X-Y-Z directions, and simultaneously start the MPI device to scan the container to obtain an initial one-dimensional MPI voltage signal U(t).

[0076] Step S130: Repeat step S120 n times to obtain n initial one-dimensional MPI voltage signals U(t) and n system matrices A. Calculate the average value An of the n matrices A. c :

[0077] A c = (A1 + A2 ... + A n ) / n;

[0078] Step S140, for the Perform interpolation to obtain Where N and M are dimensions;

[0079] Step S150, based on the image to be reconstructed Xr, Gaussian noise m, and the... The one-dimensional MPI voltage signal U(t) is obtained:

[0080]

[0081] In this embodiment, N is 120 and M is 240.

[0082] In particular, this invention innovatively employs interpolation processing, which can avoid ill-conditioned problems in solving inverse problems.

[0083] This invention incorporates Gaussian noise in order to make the simulated system matrix closer to the real system matrix.

[0084] The input signal is fed into the generator of a pre-trained generative adversarial network (GAN) model to obtain a reconstructed magnetic particle image. The GAN model includes a generator and a discriminator.

[0085] Specifically, any horizontal direction on the displacement stage is taken as the X direction, the direction perpendicular to the X direction is the Y direction, and the direction perpendicular to the plane formed by the X direction and the Y direction is the Z direction, which is located above the displacement stage.

[0086] See Figure 4 The generator is based on a multilayer perceptron, an encoding module, and a weighting module connected in sequence. The encoding module includes multiple upsampling units and multiple Transformer blocks, with each upsampling unit located between two Transformer blocks.

[0087] See Figure 6 The Transformer block is constructed based on a first normalization layer, a multi-head self-attention layer, a first summing unit, a second normalization layer, a multilayer perceptron, and a second summing unit connected in sequence.

[0088] The input of the first normalization layer is connected to the output residual of the multi-head self-attention layer, and the output of the first summing unit is connected to the output residual of the multilayer perceptron.

[0089] The input to the first normalization layer is a vector sequence stored in the embedding layer.

[0090] Figure 4The generator structure involves a one-dimensional signal first passing through a multilayer perceptron (MLP) to obtain a 64×128 long sequence. This sequence is added to a learnable positional code, passed through three Transformer blocks, and then reconstructed into an 8×8×128 image feature. Each vector in the sequence represents a pixel in the image, and the embedding dimension of the sequence is the same as the feature dimension of the image (128). Four upsampling steps are then performed. In the first upsampling step, the image is upsampled by a factor of 2 to obtain a 16×16×128 feature map, which is then reconstructed into a 256×128 sequence and passed through two Transformer blocks. This upsampling step is repeated once more to obtain a 1024×128 sequence. In the next two upsampling steps, to reduce computational complexity, a pixel-recombination upsampling method is used, increasing resolution while reducing the embedding dimension of the sequence, ultimately resulting in a (128×128)×8 sequence. The sequence is deflattened and linearly weighted to obtain a 128×128 grayscale image.

[0091] In the four upsampling steps, upsampling steps 1 and 2 employ bicubic interpolation, maintaining the feature dimension unchanged while increasing image resolution. Upsampling steps 3 and 4 utilize pixel rearrangement, which achieves upsampling by arranging and rearranging image pixels, reducing image distortion and noise, and resulting in more stable upsampling quality. Through pixel rearrangement, image resolution is doubled while the feature dimension is reduced to one-quarter of its original value. A comparison of the two upsampling methods is provided. Figure 8 As shown.

[0092] Figure 6 This describes the structure of a Transformer block. Within each Transformer block, when the feature map resolution is lower than 64×64, the multi-head self-attention module performs global attention calculations on the image. When the image resolution exceeds 64×64, a grid self-attention method is used to divide the feature map into multiple blocks, performing attention calculations only within each block, thereby reducing the computational cost of the attention mechanism. Figure 4 In Transformer blocks 1, 2, and 3, the input feature maps are relatively small, so global attention is calculated. However, Transformer block 4 processes feature maps of 64×64, so the feature map is first divided into four 32×32 windows, and then self-attention is calculated only within the window. The same applies to Transformer block 5.

[0093] The training method for the generative adversarial network model is as follows:

[0094] Step S210: Construct a grayscale image of the concentration distribution of magnetic particles in the scanning space as a dataset. Use this dataset as the image to be reconstructed, Xr, and ground truth labels. Use a portion of the dataset as training data. Input the one-dimensional MPI voltage signal from the training data into the generator in the pre-constructed generative adversarial network model to obtain the reconstructed image Xr. f ;

[0095] Step S220, based on the reconstructed image X f The corresponding truth labels are used to obtain the loss value of the generator through the pre-constructed generator loss function, and the parameters of the generator are then corrected.

[0096] Step S230: Compare the truth labels with the reconstructed image X f The input is given to the discriminator to determine the reconstructed image X. f To determine whether something is true or false, a judgment result is obtained;

[0097] Step S240: Combining the discrimination result, the loss value corresponding to the discriminator is obtained through the pre-constructed discriminator loss function, and the parameters of the discriminator are corrected.

[0098] Step S250, repeat steps S210-S240 until a trained generative adversarial network model is obtained.

[0099] The discriminator is composed of an image segmentation module, a linear weighting module, and multiple Transformer blocks connected in sequence.

[0100] The image segmentation module is used to segment the reconstructed image X f The image is divided into a set number of blocks of the same size; the linear weighting module is used to convert the set number of p blocks of image into vectors with an embedded set dimension.

[0101] Figure 5 The discriminator structure determines whether an entire image is true or false, therefore it does not need to encode each pixel individually, but rather encodes image blocks. The discriminator divides the image into p blocks of equal size; for example, if each block is 8×8, then p = 256.

[0102] After linear weighting, each image block is transformed into a 128-dimensional vector, resulting in a p×128 sequence. A classification vector is added to the beginning of the sequence to learn the image's true / false category. This (p+1)×128 sequence is added to the learnable positional encoding and used as input to the Transformer layer. The Transformer blocks in the discriminator compute global attention for the image. After a series of Transformer blocks, only the output corresponding to the classification vector is used for true / false prediction. This vector undergoes a linear transformation, ultimately outputting a probability value for judging whether the image is true or false.

[0103] The loss value corresponding to the generator is calculated as follows:

[0104] Step S221, calculate the reconstructed image X f =G(v) and the mean absolute error L of each pixel in the image to be reconstructed Xr MAE :

[0105] L MAE (G,Xr)=E(||Xr-G(v)||);

[0106] Step S222, based on the mean absolute error L MAE Calculate the loss value L generated by it. G :

[0107] L G =-E(D(G(v)))+λL MAE (G,Xr);

[0108] Wherein, λ is a parameter that adjusts the proportion of the two losses in the generator so that the generated samples not only closely approximate the distribution of the real image, but also conform to the correspondence between the one-dimensional signal and the reconstructed image. D represents the discriminator's judgment process on the reconstructed image, and E represents the average value of the discriminator's output.

[0109] Mean Absolute Error (MAE) between Generated Image and Actual Image: To ensure that the image generated by the generator is a unique image corresponding to the one-dimensional signal, the reconstructed image X is calculated. f =G(v) is the average absolute error between each pixel of the actual image and serves as a conditional constraint on the generator.

[0110] The loss value corresponding to the discriminator is calculated as follows:

[0111] L D =-E(max(0,1-D(x)))+E(max(0,D(1+G(v))));

[0112] Among them, L D This represents the loss value corresponding to the discriminator.

[0113] Adversarial Loss in GANs: To increase training stability, both the generator and discriminator employ the Hinge loss function. The discriminator predicts 1 for real samples and -1 for fake samples. When the correct prediction exceeds this range, the loss stops decreasing, allowing the discriminator to focus more on the overall error. For the generator, this loss encourages it to generate samples that the discriminator cannot distinguish, meaning the discriminator's prediction for that sample is close to the actual value.

[0114] Although the steps in the above embodiments are described in the above order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not need to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple variations are all within the protection scope of this invention.

[0115] A magnetic nanoparticle imaging reconstruction system according to a second embodiment of the present invention, the system comprising:

[0116] The signal acquisition module is configured to acquire a one-dimensional MPI voltage signal of the image to be reconstructed as an input signal.

[0117] The training module is configured to input the input signal into the generator of the trained generative adversarial network model to obtain the reconstructed magnetic particle image.

[0118] The generative adversarial network model includes a generator and a discriminator.

[0119] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0120] It should be noted that the magnetic nanoparticle imaging and reconstruction system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0121] like Figure 9 As shown, a storage device according to a third embodiment of the present invention stores a plurality of programs adapted to be loaded and executed by a processor to implement a method for magnetic nanoparticle imaging reconstruction.

[0122] like Figure 9 As shown, the fourth embodiment of the present invention includes a processor and a storage device; the processor is adapted to execute various programs; the storage device is adapted to store multiple programs; the programs are adapted to be loaded and executed by the processor to realize a method for magnetic nanoparticle imaging reconstruction.

[0123] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the storage device and processing device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0124] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0125] The following is for reference. Figure 9 It shows a schematic diagram of the structure of a computer system for implementing the methods, systems, and apparatus embodiments of this application. Figure 9 The server shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0126] like Figure 9As shown, the computer system includes a Central Processing Unit (CPU) 901, which can perform various appropriate actions and processes based on programs stored in Read Only Memory (ROM) 902 or programs loaded from storage section 908 into Random Access Memory (RAM) 903. RAM 903 also stores various programs and data required for system operation. The CPU 901, ROM 902, and RAM 903 are interconnected via bus 904. Input / output (I / O) interface 905 is also connected to bus 904.

[0127] The following components are connected to I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to I / O interface 905 as needed. Removable media 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 910 as needed so that computer programs read from them can be installed into storage section 908 as needed.

[0128] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by central processing unit (CPU) 901, it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having 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 devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0129] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0130] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0131] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0132] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0133] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method of magnetic nanoparticle imaging reconstruction, the method comprising: The method includes: The one-dimensional MPI voltage signal of the image to be reconstructed is acquired as the input signal; The input signal is input into the generator of the trained generative adversarial network model to obtain the reconstructed magnetic particle image; The generative adversarial network model includes a generator and a discriminator; The generator is based on a multilayer perceptron, an encoding module, and a weighting module connected in sequence. The encoding module includes multiple upsampling units and multiple Transformer blocks, with each upsampling unit located between two Transformer blocks. The Transformer block is constructed based on a first normalization layer, a multi-head self-attention layer, a first summing unit, a second normalization layer, a multilayer perceptron, and a second summing unit connected in sequence. The input of the first normalization layer is connected to the output residual of the multi-head self-attention layer, and the output of the first summing unit is connected to the output residual of the multilayer perceptron. The discriminator is composed of an image segmentation module, a linear weighting module, and multiple Transformer blocks connected in sequence. The image segmentation module is used to segment the reconstructed image into a set number of blocks of the same size X f The linear weighting module is used to convert the set number of block images into vectors embedded in a set number of dimensions.

2. The method for imaging and reconstruction using magnetic nanoparticles according to claim 1, characterized in that, The one-dimensional MPI A voltage signal, the acquisition method is: Step S110: Inject a magnetic nanoparticle solution of a fixed concentration into a container; Step S120: Place the container on a precisely movable displacement stage. X direction- Y direction- Z The displacement stages are moved sequentially in the direction of movement, and the operation is started simultaneously. MPI The device scans the container to obtain an initial one-dimensional image. MPI voltage signal ; Step S130, repeat n In step S120, we obtain n An initial one-dimensional MPI voltage signal and n System Matrix A ,calculate n The one mentioned A average : ； Step S140, for the Perform interpolation to obtain Where N and M are dimensions; Step S150: Reconstruct the image based on the image to be imaged. Xr Gaussian noise m The above , to obtain one dimension MPI voltage signal : 。 3. The method for imaging and reconstruction using magnetic nanoparticles according to claim 1, characterized in that, The training method for the generative adversarial network model is as follows: Step S210: Construct a grayscale image of the concentration distribution of magnetic particles in the scanning space as a dataset, and use the dataset as the image to be reconstructed. Xr And ground value labels, and use a portion of the dataset as training data; One dimension of the training data MPI A voltage signal is input into the generator in the pre-constructed generative adversarial network model to obtain a reconstructed image. ; Step S220, based on the reconstructed image The corresponding truth labels are used to obtain the loss value of the generator through the pre-constructed generator loss function, and the parameters of the generator are then corrected. Step S230: Match the truth labels with the reconstructed image. The image is input into the discriminator to determine the reconstructed image. To determine whether something is true or false, a judgment result is obtained; Step S240: Combining the discrimination result, the loss value corresponding to the discriminator is obtained through the pre-constructed discriminator loss function, and the parameters of the discriminator are corrected. Step S250, repeat steps S210-S240 until a trained generative adversarial network model is obtained.

4. The method for imaging and reconstruction using magnetic nanoparticles according to claim 3, characterized in that, The loss value corresponding to the generator is calculated as follows: Step S221, calculate the reconstructed image = With the image to be reconstructed Mean absolute error of each pixel L MAE : ； Step S222, based on the mean absolute error L MAE Calculate the loss value generated from it. : ； Among them, the To adjust the parameters of the proportion of the two losses in the generator, D This represents the discriminator's judgment process on the reconstructed image. E This indicates that the average value of the results output by the discriminator is calculated.

5. The method for imaging and reconstruction using magnetic nanoparticles according to claim 3, characterized in that, The loss value corresponding to the discriminator is calculated as follows: ； in, This represents the loss value corresponding to the discriminator.

6. A system for imaging and reconstruction using magnetic nanoparticles, based on the method for imaging and reconstruction using magnetic nanoparticles according to any one of claims 1-5, characterized in that, The system includes: The signal acquisition module is configured to acquire a one-dimensional MPI voltage signal of the image to be reconstructed as an input signal. The training module is configured to input the input signal into the generator of the trained generative adversarial network model to obtain the reconstructed magnetic particle image. The generative adversarial network model includes a generator and a discriminator.

7. A storage device storing a plurality of programs, characterized in that, The program is applicable to being loaded and executed by a processor to implement a method for magnetic nanoparticle imaging reconstruction as described in any one of claims 1-5.

8. A processing apparatus, comprising a processor and a storage device; the processor being adapted to execute various programs; the storage device being adapted to store multiple programs; characterized in that, The program is applicable to being loaded and executed by a processor to implement a method for magnetic nanoparticle imaging reconstruction as described in any one of claims 1-5.

Images