Interpretable embedded high-resolution magnetic nanoparticle imaging system and method

Abstract

The present invention belongs to the field of magnetic nanoparticle imaging technology, and specifically relates to a magnetic nanoparticle imaging system and method that can interpret embedded high resolution, aiming to solve the problem that the existing technology cannot realize magnetic particle image reconstruction of different resolutions, and the robustness of reconstruction is poor. The system of the present invention includes: an excitation field, a focusing field, a selection field, and a receiving coil; the selection field includes two circular coils, and the two circular coils of the selection field are symmetrically arranged in parallel and have the same axis; the focusing field includes two circular coils, and the two circular coils of the focusing field are symmetrically arranged in parallel and have the same axis; the two circular coils of the focusing field are perpendicular to the axes of the two circular coils of the selection field; the excitation field includes a cylindrical coil; the excitation field is arranged in the enclosed space of the excitation field and the focusing field; the receiving coil includes two cylindrical coils, and the receiving coil is arranged inside the excitation field. The present invention realizes the reconstruction of magnetic particle images of different resolutions and improves the robustness of magnetic particle image reconstruction.

Description

Technical Field

[0001] The invention belongs to the technical field of magnetic nanoparticle imaging, and in particular relates to a magnetic nanoparticle imaging system and method capable of interpreting embedded high-resolution. Background Art

[0002] Magnetic nanoparticle imaging (MPI) technology is a new imaging method 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 in the body. It has high sensitivity and resolution and is not limited by scanning depth. The MPI device measures the magnetization response of the tracer to image its spatial distribution. Therefore, the spatial distribution and concentration of the tracer are crucial to clinical diagnosis, and the output of high-quality images is a basic requirement for MPI applications.

[0003] Traditional image processing methods may be affected by noise, artifacts or low resolution for various reasons. The presence of noise greatly reduces the image quality, obscures subtle but important morphological details, and is not conducive to clinical observation and diagnosis. Some deep learning-based methods, especially deep learning generative models, have shown great promise in image reconstruction and provide some advantages. However, these methods are all based on convolutional neural networks, which have a receptive field limited to the convolution kernel when extracting features and cannot provide a global view.

[0004] At the same time, the resolution of the magnetic nanoparticle imaging system is often limited by the system hardware itself and the reconstruction method. The deep learning method can realize the reconstruction of MPI with higher resolution, and can reconstruct images of different resolutions according to needs. By integrating the generation algorithms of different resolutions through software, combined with the MPI imaging device, the detection and application of tracers at different resolutions can be realized. Based on this, the present invention proposes a magnetic nanoparticle imaging device and method that can interpret embedded high-resolution. Summary of the invention

[0005] In order to solve the above problems in the prior art, that is, to solve the problem that the prior art cannot realize magnetic particle image reconstruction with different resolutions and the reconstruction robustness is poor, the present invention proposes an interpretable embedded high-resolution magnetic nanoparticle imaging system, the system comprising: an excitation field, a focusing field, a selection field, and a receiving coil;

[0006] The selection field comprises two circular coils; the two circular coils of the selection field are symmetrically arranged in parallel and have the same axis; the selection field is used to generate a stable gradient magnetic field;

[0007] The focusing field comprises two circular coils; the two circular coils of the focusing field are symmetrically arranged in parallel and have the same axis; the axes of the two circular coils of the focusing field are perpendicular to the axes of the two circular coils of the selection field; the focusing field is used to move the magnetic field-free point;

[0008] The excitation field includes a cylindrical coil; the excitation field is arranged in the enclosed space of the excitation field and the focusing field; the axis of the excitation field passes through the orthogonal point of the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field, and is perpendicular to the plane formed by the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field; the excitation field is used to generate a point without magnetic field;

[0009] The receiving coil includes two cylindrical coils, which are respectively used as a first cylindrical coil and a second cylindrical coil; the first cylindrical coil is coaxial with the second cylindrical coil, and the first cylindrical coil is arranged outside the second cylindrical coil; the receiving coil is coaxial with the cylindrical coil of the excitation field, and the cylindrical coil of the excitation field is arranged outside the first cylindrical coil;

[0010] The first cylindrical coil is used to offset the baseband signal; the second cylindrical coil is used to receive the magnetic particle signal.

[0011] In some preferred embodiments, the interpretable embedded high-resolution magnetic nanoparticle imaging device takes the axis direction of the circular coil of the selection field as the y direction and takes the axis direction of the circular coil of the focusing field as the x direction.

[0012] In some preferred embodiments, the voltage across the cylindrical coil in the excitation field is a sinusoidal signal voltage.

[0013] In some preferred embodiments, the first cylindrical coil is connected in series with the second cylindrical coil.

[0014] In some preferred embodiments, the interpretable embedded high-resolution magnetic nanoparticle imaging system further comprises a bandpass filter, a signal generator, a high-pass filter, and a signal amplifier;

[0015] The signal generator is used to generate the sinusoidal signal required by the excitation field;

[0016] The bandpass filter is used to filter the sinusoidal signal and send it to the excitation field after filtering;

[0017] The high-pass filter is used to filter the baseband signal in the receiving coil and retain the high-frequency signal; the signal amplifier is used to amplify the signal processed by the high-pass filter.

[0018] In a second aspect of the present invention, a method for interpretable embedded high-resolution magnetic nanoparticle imaging is proposed. Based on the above-mentioned interpretable embedded high-resolution magnetic nanoparticle imaging system, the method comprises:

[0019] S100, collecting a one-dimensional MPI signal as an input signal through the constructed interpretable embedded high-resolution magnetic nanoparticle imaging system;

[0020] S200, based on the input signal, obtaining an MPI generated image through a generator in a pre-trained generative adversarial network;

[0021] S300, combining the one-dimensional signals z corresponding to the MPI images of different resolutions, obtaining the one-dimensional signals z corresponding to different pixel points of the MPI generated image, and then constructing a system matrix; based on the constructed system matrix, iteratively reconstructing through the Kaczmarz algorithm, and finally obtaining a reconstructed MPI image.

[0022] In some preferred embodiments, the one-dimensional signals corresponding to MPI images of different resolutions are obtained by:

[0023] A100, obtains MPI images of different resolutions and the corresponding one-dimensional MPI signals to construct a data set;

[0024] A200, inputting the one-dimensional MPI signal in the data set into the generator in the pre-trained generative adversarial network to obtain an MPI generated image;

[0025] A300, randomly generates a one-dimensional signal z, and generates an image through the generator as a forward result image;

[0026] A400, calculating the loss between the MPI image obtained in A200 and the forward result image, and based on the loss, updating the one-dimensional signal z by a back propagation algorithm;

[0027] A500, loop A100-A400, and obtain the one-dimensional signals corresponding to MPI images of different resolutions.

[0028] In some preferred embodiments, the generator includes a Swin-Transformer module and an upsampling module; the input of the upsampling module is used as the input of the Swin-Transformer module;

[0029] The Swin-Transformer module is constructed based on a position encoding module, a window attention module, a layer normalization module, a feedforward layer, and a layer normalization module connected in sequence.

[0030] In some preferred embodiments, a one-dimensional signal z is randomly generated by:

[0031] zint = Variable(torch.randn(nz,1,1))

[0032] Among them, zint represents the random generation of a one-dimensional signal z, torch.randn represents the function in pytorch that generates a tensor of a specified size, nz represents the length of the signal, and Variable() is the format defined by the function in pytorch.

[0033] In some preferred embodiments, the size of the randomly generated one-dimensional signal is the same as the size of the corresponding one-dimensional MPI signal in the data set.

[0034] Beneficial effects of the present invention:

[0035] The present invention realizes the reconstruction of magnetic particle images with different resolutions and improves the robustness of the magnetic particle image reconstruction.

[0036] 1) The present invention can explain that an embedded high-resolution magnetic nanoparticle imaging system uses two receiving coils, and the second receiving coil receives the same signal from the first receiving coil, but without the particle signal. Connecting the two signals in series but with opposite winding directions will cancel the transmitted signal and improve the accuracy of the collected one-dimensional MPI signal;

[0037] 2) The present invention is different from previous deep learning algorithms. The method of the present invention adds interpretable analysis of deep learning, turning it from a black box into a white box; it can realize the reconstruction process of images with different resolutions and improve the robustness of MPI image reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings.

[0039] Figure 1 It is a schematic structural diagram of an interpretable embedded high-resolution magnetic nanoparticle imaging system according to an embodiment of the present invention;

[0040] Figure 2 It is a schematic diagram of other hardware connections of an explainable embedded high-resolution magnetic nanoparticle imaging system according to an embodiment of the present invention;

[0041] Figure 3 It is a schematic flow chart of an interpretable embedded high-resolution magnetic nanoparticle imaging method according to an embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of a generative adversarial network training process according to an embodiment of the present invention;

[0043] Figure 5 It is a schematic diagram of the process of inverse solution from image to signal according to an embodiment of the present invention;

[0044] Figure 6 Schematic diagram of a Swin-Transformer module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should be noted that, in the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0047] The first embodiment of the present invention is an embedded high-resolution magnetic nanoparticle imaging device, such as Figure 1 As shown, the device includes: an excitation field, a focusing field, a selection field, and a receiving coil;

[0048] The selection field comprises two circular coils; the two circular coils of the selection field are symmetrically arranged in parallel and have the same axis; the selection field is used to generate a stable gradient magnetic field;

[0049] The focusing field comprises two circular coils; the two circular coils of the focusing field are symmetrically arranged in parallel and have the same axis; the axes of the two circular coils of the focusing field are perpendicular to the axes of the two circular coils of the selection field; the focusing field is used to move the magnetic field-free point;

[0050] The excitation field includes a cylindrical coil; the excitation field is arranged in the enclosed space of the excitation field and the focusing field; the axis of the excitation field passes through the orthogonal point of the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field, and is perpendicular to the plane formed by the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field; the excitation field is used to generate a point without magnetic field;

[0051] The receiving coil includes two cylindrical coils, which are respectively used as a first cylindrical coil and a second cylindrical coil; the first cylindrical coil is coaxial with the second cylindrical coil, and the first cylindrical coil is arranged outside the second cylindrical coil; the receiving coil is coaxial with the cylindrical coil of the excitation field, and the cylindrical coil of the excitation field is arranged outside the first cylindrical coil;

[0052] The first cylindrical coil is used to offset the baseband signal; the second cylindrical coil is used to receive the magnetic particle signal.

[0053] In order to more clearly illustrate an interpretable embedded high-resolution magnetic nanoparticle imaging device of the present invention, each module in an embodiment of the device of the present invention is described in detail below with reference to the accompanying drawings.

[0054] The present invention provides an embedded magnetic nanoparticle imaging system and method that provides reconstruction of different spatial resolutions based on interpretability. First, the corresponding magnetic nanoparticle imaging system is designed, and the equipment parameters are adjusted to collect data; then the network is trained through a deep learning algorithm to train the matching relationship between images and signals of different resolutions; secondly, an interpretable analysis algorithm is introduced to invert the algorithm network that generates the image, infer the signal value from the image, and use the system matrix composed of the obtained signal to achieve image reconstruction; finally, the software is used to integrate reconstruction algorithms at different resolutions to achieve system automated acquisition, reconstruction and analysis. The details are as follows:

[0055] An interpretable embedded high-resolution magnetic nanoparticle imaging system according to a first embodiment of the present invention comprises an excitation field, a focusing field, a selection field, and a receiving coil;

[0056] The selection field comprises two circular coils; the two circular coils of the selection field are symmetrically arranged in parallel and have the same axis; the selection field is used to generate a stable gradient magnetic field;

[0057] The focusing field includes two circular coils; the two circular coils of the focusing field are symmetrically arranged in parallel and have the same axis; the axes of the two circular coils of the focusing field are perpendicular to the axes of the two circular coils of the selection field; the focusing field is used to move the magnetic field-free point, improve the field of view, and make the FFP move in a wider range;

[0058] The excitation field includes a cylindrical coil; the excitation field is set in the enclosed space of the excitation field and the focusing field; the axis of the excitation field passes through the orthogonal point of the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field, and is perpendicular to the plane formed by the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field; the voltage applied to the excitation field is a sinusoidal signal, and when the excitation field is superimposed on the selection field, a point without magnetic field (FFP) will be generated due to the superposition of magnetic fields. By changing the magnitude of the voltage at both ends of the excitation field, this point without magnetic field will move, and the range of movement is the range of the final imaging (that is, the movement directions of the magnetic field free points in three directions are generated to achieve three-dimensional imaging);

[0059] The explainable embedded high-resolution magnetic nanoparticle imaging device takes the axial direction of the circular coil of the selection field as the y direction (that is, the central axis of the selection field is parallel to the y-axis direction), the axial direction of the circular coil of the focusing field as the x direction (that is, the central axis of the focusing field is parallel to the x-axis direction), and the central axis of the excitation field is parallel to the z-axis direction.

[0060] The receiving coil includes two cylindrical coils, which are respectively used as the first cylindrical coil and the second cylindrical coil; the first cylindrical coil is coaxial with the second cylindrical coil, and the first cylindrical coil is arranged on the outside of the second cylindrical coil; the receiving coil is coaxial with the cylindrical coil of the excitation field, and the cylindrical coil of the excitation field is arranged on the outside of the first cylindrical coil; the first cylindrical coil (i.e., the outer ring, i.e., Figure 1 The receiving coil 2 in the embodiment is used to cancel the baseband signal; the second cylindrical coil (i.e., the inner ring, i.e. Figure 1 The receiving coil 1 in the embodiment is used to receive magnetic particle signals. The first cylindrical coil is connected in series with the second cylindrical coil.

[0061] The magnetic nanoparticle imaging system that can interpret embedded high resolution also includes a bandpass filter, a signal generator, a high-pass filter, a signal amplifier, such as Figure 2 As shown;

[0062] A signal generator is used to generate the sinusoidal signal required for the excitation field; a bandpass filter is used to filter the sinusoidal signal and send it to the excitation field after processing; a high-pass filter is used to filter the baseband signal in the receiving coil and retain the high-frequency signal; a signal amplifier is used to amplify the signal processed by the high-pass filter to obtain a one-dimensional MPI signal and send it to a computer for magnetic particle imaging.

[0063] A computer, used for generating scanning parameters of an interpretable embedded high-resolution magnetic nanoparticle imaging system and setting parameters of the interpretable embedded high-resolution magnetic nanoparticle imaging system through cables or wireless communications, and sending imaging instructions to scan the target to be measured when the target to be measured is set in a scanning area of ​​the interpretable embedded high-resolution magnetic nanoparticle imaging system;

[0064] It is also used to obtain an MPI generated image based on the collected one-dimensional MPI signal through a generator in a pre-trained generative adversarial network; combine the one-dimensional signals z corresponding to the MPI images of different resolutions to obtain the one-dimensional signals z corresponding to different pixel points of the MPI generated image, and then construct a system matrix; based on the constructed system matrix, iteratively reconstruct through the Kaczmarz algorithm to finally obtain a reconstructed MPI image. The specific process can be referred to the second embodiment, which will not be described in detail here.

[0065] It should be noted that the interpretable embedded high-resolution magnetic nanoparticle imaging system provided in the above embodiment is only illustrated by 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 decomposed or combined. For example, the modules in the above embodiment can be combined 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 modules or steps, and are not regarded as improper limitations of the present invention.

[0066] A second embodiment of the present invention is a method for interpreting embedded high-resolution magnetic nanoparticle imaging, such as Figure 3 As shown, based on the above-mentioned interpretable embedded high-resolution magnetic nanoparticle imaging system, the method comprises:

[0067] S100, collecting a one-dimensional MPI signal as an input signal through the constructed interpretable embedded high-resolution magnetic nanoparticle imaging system;

[0068] S200, based on the input signal, obtaining an MPI generated image through a generator in a pre-trained generative adversarial network;

[0069] S300, combining the one-dimensional signals z corresponding to the MPI images of different resolutions, obtaining the one-dimensional signals z corresponding to different pixel points of the MPI generated image, and then constructing a system matrix; based on the constructed system matrix, iteratively reconstructing through the Kaczmarz algorithm, and finally obtaining a reconstructed MPI image.

[0070] In the following, the process of one-dimensional signals corresponding to MPI images of different resolutions is first described, and then the process of image reconstruction by interpretable embedded high-resolution magnetic nanoparticle imaging method is described.

[0071] The process of acquiring the one-dimensional signal corresponding to MPI images of different resolutions is as follows:

[0072] A100, obtains MPI images of different resolutions and the corresponding one-dimensional MPI signals to construct a data set;

[0073] A200, inputting the one-dimensional MPI signal in the data set into the generator in the pre-trained generative adversarial network to obtain an MPI generated image;

[0074] A300, randomly generates a one-dimensional signal z, and generates an image through the generator as a forward result image;

[0075] A400, calculating the loss between the MPI image obtained in A200 and the forward result image, and based on the loss, updating the one-dimensional signal z by a back propagation algorithm;

[0076] A500, loop A100-A400, get the one-dimensional signal corresponding to the MPI image with different resolutions

[0077] In this embodiment, a signal-image matching data set for training is first constructed, the data set includes images of different resolutions and their corresponding one-dimensional MPI signals, and a Swin-Transformer is used to replace the convolutional neural network to form a generative adversarial network (the generator includes a Swin-Transformer module, an upsampling module, such as Figure 6 As shown in the figure, the input of the upsampling module is used as the input of the Swin-Transformer module), the MPI signal is used as the input of the generator, and the image is used as the output of the generator. The reconstructed image and ground truth are input into the discriminator for real or fake discrimination, and the iterative network parameters are updated, as shown in the figure. Figure 4 As shown in the figure, the parameters of the generator and discriminator are changed, and finally the results are quantified through evaluation indicators, and the model with the best results is selected.

[0078] Then use interpretability algorithms based on generative adversarial networks, such as Figure 5 As shown in the figure, first a random one-dimensional signal z is generated, and its size is defined to be the same as the one-dimensional signal in the data set. Then, the image x' generated by z after passing through the generator is calculated. This image is different from x in the data set, so the loss function between x and x' is calculated, and the random one-dimensional signal z is reversely updated through the loss function. Finally, the inverse solution from image x to signal z can be achieved.

[0079] This algorithm can be used to reversely obtain the result of the generator, that is, to obtain the mapping from image to signal. Since the system matrix is ​​the superposition of signals of unit points at different positions, this method can be used to construct the system matrix corresponding to the generative adversarial network, and use the system matrix to reconstruct the signal. The spatial resolution in the traditional MPI system is determined by the actual acquired system matrix, but using this method, system matrices of different resolutions can be obtained according to the resolution of the training data set, so images with different spatial resolutions can be reconstructed. That is:

[0080] S100, collecting a one-dimensional MPI signal as an input signal through the constructed interpretable embedded high-resolution magnetic nanoparticle imaging system;

[0081] S200, based on the input signal, obtaining an MPI generated image through a generator in a pre-trained generative adversarial network;

[0082] S300, combining the one-dimensional signals z corresponding to the MPI images of different resolutions, obtaining the one-dimensional signals z corresponding to different pixel points of the MPI generated image, and then constructing a system matrix; based on the constructed system matrix, iteratively reconstructing through the Kaczmarz algorithm, and finally obtaining a reconstructed MPI image.

[0083] In this embodiment, the specific process of constructing the system matrix is ​​as follows:

[0084] The image size generated by MPI is M×N, where M represents the number of pixels on the horizontal axis and N represents the number of pixels on the vertical axis. Each pixel can obtain a one-dimensional signal z with a dimension of 1×L. In this way, a three-dimensional matrix of size M×N×L can be obtained, namely the system matrix.

[0085] In order to more clearly understand the specific application process of the above system and method, an example is given below:

[0086] The selection field uses permanent magnets, placed in a symmetrical orientation so that the central field between the two magnets is canceled; the excitation field coil is installed near the scanner to reduce eddy current losses. Typically, the excitation field is made of Litz coils to mitigate skin effects; the receiving coils are made of twisted wires and are implemented in the form of solenoids or Helmholtz coil pairs. In the case of multiple transmit and receive coils, a dedicated signal chain is implemented for each excitation direction. The signal to move the FFP is generated by a power amplifier. Since MPI requires a pure sine wave for signal excitation, the use of a bandpass filter can improve the quality of the power amplifier signal. The signal is then applied to the transmission (transmission generally uses a transmission coil, which is usually implemented as an oscillating circuit resonating at the transmission frequency).

[0087] The signal is measured by a receiving coil. This signal is rich in high harmonics, so a broadband receiving chain is required. The transmitted signal is inevitably coupled directly to the receiving coil. Since the direct feedthrough of the transmitting coil is several orders of magnitude larger than the particle signal, the excitation signal needs to be eliminated in the analog receiving chain. For this purpose, the gradient measurement method can be used, which uses two receiving coils, the second receiving coil receives the same signal from the first receiving coil, but without the particle signal. Connecting the two signals in series but with opposite windings will cancel the transmitted signal. After signal compensation and / or filtering, the voltage is usually in the pV range, so it must be amplified to fit the input range of the digital-to-analog converter. The amplifier needs very low noise characteristics to detect the small pV signal. After amplification, the signal is digitized and stored in a file on the measurement computer for further processing.

[0088] The parameters of the constructed MPI hardware system are as follows: the field strength in three directions is 2T / m, the excitation field strength in three directions is 15mT, and the frequencies in three directions are 24.51KHz, 26.042KHz and 25KHz respectively. The sampling frequency of the receiving coil is 2.5MHz, and the focusing field strength is 1T / m. In the experiment, the scanning method used is 3D Lissajous, and the size of the region of interest (field of view) is 15mm×15mm×15mm. After scanning the phantom, the dimension of the signal in each direction is the same, and the total length is 812×3.

[0089] like Figure 3 As shown in the figure, the Swin-Transformer module and the upsampling module are used in the generator. The Swin-Transformer contains an input position encoding, which divides the input signal into several patches, each of which is 8×8 in size. The patches are then positionally encoded, and the weights between each patch are calculated through window attention to obtain the calculated parameters, which are then normalized through a layer, a feedforward layer, and a layer normalization to obtain the output result.

[0090] Randomly generate a z signal, zint = Variable(torch.randn(nz,1,1)), where nz represents the length of the signal. Then use the optim iterator to update zint, calculate the forward result of the generator x'=gen.forward(zint), and then calculate the mse loss of x' and x, recloss = F.mse_loss(x, x'), back-propagate to update zint, and finally get the reverse z corresponding to each image.

[0091] If the resolution of the generated image is 128×128, then 128×128 z signals corresponding to each point-by-point activation image can be obtained. These signals can be assembled into a system matrix and iteratively reconstructed using the Kaczmarz algorithm to finally obtain the reconstructed image.

[0092] In addition, python and Qt can be used to develop software, which includes a signal acquisition interface for automated signal acquisition. A data construction module is used to store the acquired signals and phantom images in a one-to-one pairing and construct a data set. Generative adversarial network training is then used to obtain the system matrix through an interpretable algorithm, and finally the obtained system matrix is ​​used for reconstruction.

[0093] An electronic device according to the third embodiment of the present invention comprises at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned interpretable embedded high-resolution magnetic nanoparticle imaging method.

[0094] A fourth embodiment of the present invention is a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned interpretable embedded high-resolution magnetic nanoparticle imaging method.

[0095] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the electronic device and computer-readable storage medium described above can refer to the corresponding process in the aforementioned method example and will not be repeated here.

[0096] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, the square boxes of two consecutive representations can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0097] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.

[0098] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.

[0099] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle 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 fall within the protection scope of the present invention.

Claims

1. A method for interpreting embedded high-resolution magnetic nanoparticle imaging, characterized in that: The method includes: S100, collecting a one-dimensional MPI signal as an input signal through the constructed interpretable embedded high-resolution magnetic nanoparticle imaging system; S200, based on the input signal, obtaining an MPI generated image through a generator in a pre-trained generative adversarial network; S300, combining the one-dimensional signals z corresponding to the MPI images with different resolutions, obtaining the one-dimensional signals z corresponding to different pixel points of the MPI generated image, and then constructing a system matrix; based on the constructed system matrix, iteratively reconstructing by using the Kaczmarz algorithm, and finally obtaining a reconstructed MPI image; The method for obtaining the one-dimensional signal corresponding to the MPI images of different resolutions is as follows: A100, obtains MPI images of different resolutions and the corresponding one-dimensional MPI signals to construct a data set; A200, inputting the one-dimensional MPI signal in the data set into the generator in the pre-trained generative adversarial network to obtain an MPI generated image; A300, randomly generates a one-dimensional signal z, and generates an image through the generator as a forward result image; A400, calculating the loss between the MPI image obtained in A200 and the forward result image, and based on the loss, updating the one-dimensional signal z by a back propagation algorithm; A500, loop A100-A400, and obtain the one-dimensional signals corresponding to MPI images of different resolutions.

2. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to claim 1, characterized in that: The generator includes a Swin-Transformer module and an upsampling module; the input of the upsampling module is used as the input of the Swin-Transformer module; The Swin-Transformer module is constructed based on a position encoding module, a window attention module, a layer normalization module, a feedforward layer, and a layer normalization module connected in sequence.

3. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to claim 2, characterized in that: Randomly generate a one-dimensional signal z, the method is: zint = Variable(torch.randn(nz,1,1)); Among them, zint represents the random generation of a one-dimensional signal z, torch.randn represents the function in pytorch that generates a tensor of a specified size, nz represents the length of the signal, and Variable() is the format defined by the function in pytorch.

4. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to claim 3, characterized in that: The size of the randomly generated one-dimensional signal is the same as the size of the corresponding one-dimensional MPI signal in the data set.

5. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to any one of claims 1 to 4, characterized in that: Based on an interpretable embedded high-resolution magnetic nanoparticle imaging system, the system includes: an excitation field, a focusing field, a selection field, and a receiving coil; The selection field comprises two circular coils; the two circular coils of the selection field are symmetrically arranged in parallel and have the same axis; the selection field is used to generate a stable gradient magnetic field; The focusing field comprises two circular coils; the two circular coils of the focusing field are symmetrically arranged in parallel and have the same axis; the axes of the two circular coils of the focusing field are perpendicular to the axes of the two circular coils of the selection field; the focusing field is used to move the magnetic field-free point; The excitation field includes a cylindrical coil; the excitation field is arranged in the enclosed space of the excitation field and the focusing field; the axis of the excitation field passes through the orthogonal point of the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field, and is perpendicular to the plane formed by the axis of the circular coil of the focusing field and the axis of the circular coil of the excitation field; the excitation field is used to generate a point without magnetic field; The receiving coil includes two cylindrical coils, which are respectively used as a first cylindrical coil and a second cylindrical coil; the first cylindrical coil is coaxial with the second cylindrical coil, and the first cylindrical coil is arranged outside the second cylindrical coil; the receiving coil is coaxial with the cylindrical coil of the excitation field, and the cylindrical coil of the excitation field is arranged outside the first cylindrical coil; The first cylindrical coil is used to offset the baseband signal; the second cylindrical coil is used to receive the magnetic particle signal.

6. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to claim 5, characterized in that: The interpretable embedded high-resolution magnetic nanoparticle imaging device takes the axis direction of the circular coil of the selection field as the y direction and takes the axis direction of the circular coil of the focusing field as the x direction.

7. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to claim 6, characterized in that: The voltage across the cylindrical coil in the excitation field is a sinusoidal signal voltage.

8. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to claim 7, characterized in that: The first cylindrical coil is connected in series with the second cylindrical coil.

9. The method for interpretable embedded high-resolution magnetic nanoparticle imaging according to claim 8, characterized in that: The interpretable embedded high-resolution magnetic nanoparticle imaging system also includes a bandpass filter, a signal generator, a high-pass filter, and a signal amplifier; The signal generator is used to generate the sinusoidal signal required by the excitation field; The bandpass filter is used to filter the sinusoidal signal and send it to the excitation field after filtering; The high-pass filter is used to filter the baseband signal in the receiving coil and retain the high-frequency signal; The signal amplifier is used to amplify the signal processed by the high-pass filter.

Images