Magnetic particle imaging reconstruction method and system based on time-frequency dual information

Abstract

This invention belongs to the field of magnetic particle imaging reconstruction, specifically involving a magnetic particle imaging reconstruction method and system based on time-frequency dual information. It aims to solve the problem that existing magnetic particle imaging methods cannot simultaneously improve resolution and signal-to-noise ratio, resulting in low MPI imaging quality. The invention includes: acquiring an MPI time-domain image under low field strength gradient acquisition parameters of an MPI device, and transforming the MPI time-domain image to the frequency domain using a two-dimensional Fourier transform; based on the MPI time-domain image and the MPI frequency-domain image, performing image reconstruction using a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results. This invention obtains a low-resolution MPI image by selecting low-gradient acquisition parameters. On this basis, fixed-power Gaussian white noise is added, and the time-frequency dual-branch neural network is used to improve image resolution and remove noise. Ultimately, it simultaneously improves the spatial resolution and signal-to-noise ratio of the MPI, enabling the detection and clear imaging of magnetic particles even at low magnetic particle concentrations.

Description

Technical Field

[0001] This invention belongs to the field of magnetic particle imaging reconstruction, and specifically relates to a magnetic particle imaging reconstruction method and system based on time-frequency dual information. Background Technology

[0002] MPI is a novel, cutting-edge tomographic imaging technique for imaging the concentration distribution of biocompatible superparamagnetic nanoparticles, offering high sensitivity, high resolution, and no radiation. This method utilizes the nonlinear magnetization behavior of magnetic nanoparticles to determine their local concentration. Superparamagnetic iron oxide (SPIO) is a suitable nanoparticle, used as a clinically approved contrast agent for liver examinations in magnetic resonance imaging (MRI), typically administered intravenously. MPI measures only the distribution of magnetic nanoparticles, achieving direct imaging by measuring their magnetism; therefore, it is a tracer-based method.

[0003] Spatial resolution is one of the most important parameters in MPI imaging. Essentially, spatial resolution describes the distance between two objects so that they can be distinguished. Distinguishing means there is a significant minimum grayscale value between the two objects; significant means the minimum value at the gap is less than half the maximum value at the object's location. MPI imaging is very flexible, allowing imaging performance to be adjusted by appropriately selecting acquisition parameters. For example, spatial resolution can be improved by enhancing the field intensity gradient of the selected field. Generally, MPI spatial resolution is proportional to the field intensity gradient; with high field intensity gradients, spatial resolution can reach 1 mm. However, a higher gradient is not always better; an excessively large gradient can affect the imaging signal-to-noise ratio, thus impacting sensitivity. Furthermore, noise is generated during the acquisition, encoding / decoding, and reconstruction of MPI images, which severely affects MPI imaging quality. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, namely the inability of current magnetic particle imaging to simultaneously improve resolution and signal-to-noise ratio, resulting in low MPI imaging quality, this invention provides a magnetic particle imaging reconstruction method based on time-frequency dual information. The magnetic particle imaging reconstruction method includes:

[0005] The MPI time-domain image under low field strength gradient acquisition parameters of the MPI device is acquired, and the MPI time-domain image is transformed to the frequency domain by two-dimensional Fourier transform to obtain the MPI frequency domain image.

[0006] Based on the MPI time-domain image and the MPI frequency-domain image, image reconstruction is performed using a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results.

[0007] In some preferred embodiments, the time-frequency dual-branch neural network includes a frequency domain branch, a time domain branch, a fusion module, and an output layer;

[0008] The frequency domain branch is used to extract features from the MPI frequency domain image.

[0009] The temporal branch performs feature extraction of the MPI temporal image;

[0010] The fusion module fuses the features extracted from the frequency domain branch and the features extracted from the time domain branch to obtain fused features;

[0011] The output layer performs feature extraction of the fused features to obtain the magnetic particle imaging reconstruction result.

[0012] In some preferred embodiments, the frequency domain branch includes a first complex convolutional layer, a second complex convolutional layer, a third complex convolutional layer, a first ReLU layer, and a second ReLU layer;

[0013] The first ReLU layer is disposed between the first complex convolutional layer and the second complex convolutional layer;

[0014] The second ReLU layer is disposed between the second complex convolutional layer and the third complex convolutional layer.

[0015] In some preferred embodiments, the temporal branch includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a first deconvolutional layer, a second deconvolutional layer, a first ReLU layer, a second ReLU layer, a third ReLU layer, and a fourth ReLU layer;

[0016] The first ReLU layer is disposed between the first convolutional layer and the second convolutional layer;

[0017] The second ReLU layer is disposed between the second convolutional layer and the first deconvolutional layer;

[0018] The third ReLU layer is disposed between the first deconvolution layer and the second deconvolution layer;

[0019] The fourth ReLU layer is positioned between the second deconvolution layer and the third convolution layer.

[0020] In some preferred embodiments, the fusion module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a first ReLU layer, and a second ReLU layer;

[0021] The first ReLU layer is positioned between the first convolutional layer and the second convolutional layer;

[0022] The second ReLU layer is positioned between the second convolutional layer and the third convolutional layer.

[0023] In some preferred embodiments, the training method for the time-frequency dual-branch neural network is as follows:

[0024] Acquire a set number of low-gradient MPI time-domain images, add a set amount of Gaussian noise to the images, obtain the corresponding clear and noise-free images, perform nonlinear normalization of the images, and obtain normalized MPI time-domain images.

[0025] The low-gradient MPI time-domain image is transformed to the frequency domain by two-dimensional Fourier transform, and the image is normalized to obtain a normalized MPI frequency-domain image.

[0026] Using the normalized MPI time-domain image and the normalized MPI frequency-domain image as training sample pairs, and the corresponding normalized clear and noise-free image as sample labels, the time-frequency dual-branch neural network is iteratively trained until the set training termination condition is met, and the trained time-frequency dual-branch neural network is obtained.

[0027] In some preferred embodiments, the loss function during the training of the time-frequency dual-branch neural network is a function that fuses the loss functions of the two branches of the time-frequency dual-branch neural network according to a set ratio.

[0028] In some preferred embodiments, the performance of the time-frequency dual-branch neural network is evaluated using a set evaluation metric.

[0029] In some preferred embodiments, the set evaluation indicators include:

[0030] The measures of image half-width, Dice coefficient, root mean square error, peak signal-to-noise ratio, and structural similarity index.

[0031] In another aspect, the present invention proposes a magnetic particle imaging reconstruction system based on time-frequency dual information, the magnetic particle imaging reconstruction system comprising:

[0032] The image acquisition module acquires MPI time-domain images under low field strength gradient acquisition parameters of the MPI device;

[0033] The image conversion module transforms the MPI time-domain image to the frequency domain using a two-dimensional Fourier transform to obtain an MPI frequency-domain image.

[0034] The reconstruction module, based on the MPI time-domain image and the MPI frequency-domain image, performs image reconstruction using a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results.

[0035] The beneficial effects of this invention are:

[0036] (1) The magnetic particle imaging reconstruction method based on time and frequency dual information of the present invention can simultaneously improve the signal-to-noise ratio and resolution of MPI equipment by means of a dual-branch neural network. It makes full use of the characteristics of time domain data and frequency domain data, so that high resolution and high signal-to-noise ratio MPI images can still be obtained under the setting of acquisition parameters of low field strength gradient of MPI equipment.

[0037] (2) The present invention is a magnetic particle imaging reconstruction method based on time and frequency dual information. In the field of medical imaging, high resolution and high signal-to-noise ratio are important indicators. It is necessary to meet the clinical needs for safe and rapid vascular imaging and tumor imaging, help researchers to better understand the disease process from the organ, cell and molecular level, improve the imaging potential of magnetic particle imaging system, and overcome the problem of mutual restriction between signal-to-noise ratio and resolution in MPI equipment. Attached Figure Description

[0038] 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:

[0039] Figure 1 This is a schematic flowchart of the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention;

[0040] Figure 2 This is a framework diagram of the time-frequency dual-branch neural network of the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention;

[0041] Figure 3 This is a schematic diagram of the training process of the time-frequency dual-branch neural network in the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention;

[0042] Figure 4 This is a comparison of images of various states in an embodiment of the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention. Detailed Implementation

[0043] 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.

[0044] 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.

[0045] The present invention provides a magnetic particle imaging reconstruction method based on time-frequency dual information, the magnetic particle imaging reconstruction method comprising:

[0046] The MPI time-domain image under low field strength gradient acquisition parameters of the MPI device is acquired, and the MPI time-domain image is transformed to the frequency domain by two-dimensional Fourier transform to obtain the MPI frequency domain image.

[0047] Based on the MPI time-domain image and the MPI frequency-domain image, image reconstruction is performed using a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results.

[0048] To more clearly explain the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention, the following will be combined with... Figure 1 The steps in the embodiments of the present invention will be described in detail below.

[0049] The magnetic particle imaging reconstruction method based on time-frequency dual information according to the first embodiment of the present invention is described in detail below:

[0050] The MPI time-domain image under low field strength gradient acquisition parameters of the MPI device is acquired, and the MPI time-domain image is transformed to the frequency domain by two-dimensional Fourier transform to obtain the MPI frequency domain image.

[0051] Based on the MPI time-domain image and the MPI frequency-domain image, image reconstruction is performed using a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results.

[0052] like Figure 2 The diagram shown is a framework diagram of the time-frequency dual-branch neural network for the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention. The time-frequency dual-branch neural network includes a frequency domain branch, a time domain branch, a fusion module, and an output layer.

[0053] The frequency domain branch includes a first complex convolutional layer, a second complex convolutional layer, a third complex convolutional layer, a first ReLU layer, and a second ReLU layer, which are used to extract features from the MPI frequency domain image.

[0054] The network layer connections in the frequency domain branch are as follows: from input to output, they are: first complex convolutional layer, first ReLU layer, second complex convolutional layer, second ReLU layer, and third complex convolutional layer. That is, the first ReLU layer is located between the first and second complex convolutional layers, and the second ReLU layer is located between the second and third complex convolutional layers.

[0055] The temporal branch includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a first deconvolutional layer, a second deconvolutional layer, a first ReLU layer, a second ReLU layer, a third ReLU layer, and a fourth ReLU layer, for feature extraction of the MPI temporal image.

[0056] The connection relationships of the network layers in the temporal branch are as follows, from input to output: first convolutional layer, first ReLU layer, second convolutional layer, second ReLU layer, first deconvolutional layer, third ReLU layer, second deconvolutional layer, fourth ReLU layer, and third convolutional layer. Specifically, the first ReLU layer is positioned between the first and second convolutional layers, the second ReLU layer is positioned between the second convolutional layer and the first deconvolutional layer, the third ReLU layer is positioned between the first and second deconvolutional layers, and the fourth ReLU layer is positioned between the second deconvolutional layer and the third convolutional layer.

[0057] The fusion module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a first ReLU layer, and a second ReLU layer, which fuse the features extracted from the frequency domain branch and the features extracted from the time domain branch to obtain fused features.

[0058] The connection relationship of each network layer in the fusion module is as follows: from input to output, it is the first convolutional layer, the first ReLU layer, the second convolutional layer, the second ReLU layer, and the third convolutional layer. That is, the first ReLU layer is set between the first and second convolutional layers, and the second ReLU layer is set between the second and third convolutional layers.

[0059] The output layer performs feature extraction of fused features to obtain the magnetic particle imaging reconstruction results.

[0060] like Figure 3 The diagram shown illustrates the training process of the time-frequency dual-branch neural network in the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention, including:

[0061] The first step is dataset construction. A dataset containing 5000 MNIST images is constructed, with 4000 used as the training set and 1000 as the test set. The image size is set to 128×128, and the entire dataset is normalized to a constant value defined by the maximum intensity of the dataset.

[0062] The second step is data acquisition and preprocessing. A set number of low-gradient MPI time-domain images are acquired, and a set amount of Gaussian noise is added to the images to obtain corresponding clear, noise-free images. The images are then nonlinearly normalized to obtain normalized MPI time-domain images.

[0063] The third step is to acquire the frequency domain MPI image. The low-gradient MPI time-domain image is transformed to the frequency domain using a two-dimensional Fourier transform, and then the image is normalized to obtain a normalized MPI frequency domain image.

[0064] The fourth step involves the time-frequency dual-branch neural network structure and its input and output information. The input to the neural network includes the aforementioned low-gradient time-domain image data and the frequency-domain image data after two-dimensional Fourier transform.

[0065] The frequency domain data uses a complex convolution module to extract frequency domain image features. In the complex convolution module, complex convolution kernels are used to extract real and imaginary features. The module consists of three complex convolution layers, with a ReLU layer added after each complex convolution layer.

[0066] Temporal data is processed using a standard convolutional module to extract useful features from blurred MPI images while filtering out useless blur artifacts. The standard convolutional module employs an encoder-decoder symmetric structure, consisting of two convolutional layers and two deconvolutional layers. Finally, a separate convolutional layer is used to integrate the extracted features.

[0067] To further optimize the MPI image, a fusion module is used to fuse the results of the frequency domain branch and the temporal branch. After the frequency domain convolution result and the temporal convolution result are fused, features are further extracted using convolutional layers, and finally a clear MPI image is output.

[0068] Step 5, Neural Network Training. Using the normalized MPI time-domain image and the normalized MPI frequency-domain image as training sample pairs, and the corresponding normalized clear and noise-free image as sample labels, the time-frequency dual-branch neural network is iteratively trained until the set training termination condition is met, and the trained time-frequency dual-branch neural network is obtained.

[0069] The time-frequency bibranch neural network has a loss function during training that is a fusion of the loss functions of the two branches of the time-frequency bibranch neural network according to a set ratio.

[0070] In one embodiment of the present invention, the network input is a low-gradient noise image, the label is the corresponding high-quality clean image, and the number of iterations is set to 1000. During training, the loss function is a fusion of the two-branch network loss function at a certain ratio, with the ratio coefficient of the time-domain module network being 0.9 and the ratio coefficient of the frequency-domain module network being 0.1.

[0071] The sixth step is to evaluate the trained time-frequency dual-branch neural network using the established evaluation metrics. The results are quantitatively analyzed from four aspects: resolution, localization accuracy, image contrast, and shape restoration. This invention selects half-width at half-maximum (FWHM), Dice coefficient, root mean square error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM) as evaluation metrics to quantitatively assess image quality and modify the network training parameters based on the evaluation results.

[0072] To verify the effectiveness of the magnetic particle imaging reconstruction based on time-frequency dual information proposed in this invention, low-gradient MPI images were first simulated in a simulation system. Then, deep neural networks were used for image reconstruction, and simulation experiments were conducted. The main process is as follows:

[0073] (1) Parameter settings:

[0074] The simulation parameters are designed as follows: the two-dimensional field of view is 20×20mm; the voltage signal simulation is based on the paramagnetic Langevin model; the field strength gradient is set to 3T / m; and the particle size is set to 20nm. A Cartesian trajectory is simulated, with a sampling time of 2.5MS / s. Furthermore, to illustrate the robustness of the network in this invention, 20dB of Gaussian white noise is added to the blurred MPI image.

[0075] The network architecture is as follows: In the complex convolution module, complex convolution kernels are used to extract real and imaginary features. This module consists of three complex convolutional layers, with a ReLU layer added after each convolutional layer. In the ordinary convolution module, an encoder-decoder symmetric structure is used. This module consists of two convolutional layers and two deconvolutional layers. At the end of the module, a separate convolutional layer is used to integrate the extracted features again. To further optimize the MPI image, a fusion module is used to fuse the results from both domains. After fusing the frequency domain convolution results and the temporal domain convolution results, further features are extracted using convolutional layers, finally outputting a clear and unified MPI image.

[0076] like Figure 4 The image shown is a comparison of images in various states of an embodiment of the magnetic particle imaging reconstruction method based on time-frequency dual information of the present invention. The first row is the time-domain image and frequency-domain image of the blurred MPI image, the second row is the time-domain image and frequency-domain image of the reconstructed MPI image, and the third row is the time-domain image and frequency-domain image of the clear MPI image.

[0077] As can be seen, this invention can simultaneously improve the signal-to-noise ratio and resolution of MPI devices by using a dual-branch neural network, making full use of the characteristics of time-domain and frequency-domain data, so that high-resolution and high signal-to-noise ratio MPI images can still be obtained even with the low field strength gradient acquisition parameters of the MPI device.

[0078] 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.

[0079] The second embodiment of the magnetic particle imaging reconstruction system based on time-frequency dual information of the present invention includes:

[0080] The image acquisition module acquires MPI time-domain images under low field strength gradient acquisition parameters of the MPI device;

[0081] The image conversion module transforms the MPI time-domain image to the frequency domain using a two-dimensional Fourier transform to obtain an MPI frequency-domain image.

[0082] The reconstruction module, based on the MPI time-domain image and the MPI frequency-domain image, performs image reconstruction using a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results.

[0083] 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.

[0084] It should be noted that the magnetic particle imaging reconstruction system based on time-frequency dual information 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.

[0085] An electronic device according to a third embodiment of the present invention includes:

[0086] At least one processor;

[0087] and a memory communicatively connected to at least one of the processors;

[0088] The memory stores instructions that can be executed by the processor to implement the above-described magnetic particle imaging reconstruction method based on time-frequency dual information.

[0089] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, which are executed by the computer to implement the above-described magnetic particle imaging reconstruction method based on time-frequency dual information.

[0090] 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 processes in the foregoing method embodiments, and will not be repeated here.

[0091] 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.

[0092] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0093] 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.

[0094] 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 magnetic particle imaging reconstruction method based on time-frequency dual information, characterized in that, The magnetic particle imaging reconstruction method includes: The MPI time-domain image under low field strength gradient acquisition parameters of the MPI device is acquired, and the MPI time-domain image is transformed to the frequency domain by two-dimensional Fourier transform to obtain the MPI frequency domain image. Based on the MPI time-domain image and the MPI frequency-domain image, image reconstruction is performed using a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results. The time-frequency dual-branch neural network includes a frequency domain branch, a time domain branch, a fusion module, and an output layer; The frequency domain branch is used to extract features from the MPI frequency domain image. The temporal branch performs feature extraction of the MPI temporal image; The fusion module fuses the features extracted from the frequency domain branch and the features extracted from the time domain branch to obtain fused features; The output layer performs feature extraction of the fused features to obtain the magnetic particle imaging reconstruction result.

2. The magnetic particle imaging reconstruction method based on time-frequency dual information according to claim 1, characterized in that, The frequency domain branch includes a first complex convolutional layer, a second complex convolutional layer, a third complex convolutional layer, a first ReLU layer, and a second ReLU layer; The first ReLU layer is disposed between the first complex convolutional layer and the second complex convolutional layer; The second ReLU layer is disposed between the second complex convolutional layer and the third complex convolutional layer.

3. The magnetic particle imaging reconstruction method based on time-frequency dual information according to claim 1, characterized in that, The temporal branch includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a first deconvolutional layer, a second deconvolutional layer, a first ReLU layer, a second ReLU layer, a third ReLU layer, and a fourth ReLU layer; The first ReLU layer is disposed between the first convolutional layer and the second convolutional layer; The second ReLU layer is disposed between the second convolutional layer and the first deconvolutional layer; The third ReLU layer is disposed between the first deconvolution layer and the second deconvolution layer; The fourth ReLU layer is positioned between the second deconvolution layer and the third convolution layer.

4. The magnetic particle imaging reconstruction method based on time-frequency dual information according to claim 1, characterized in that, The fusion module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a first ReLU layer, and a second ReLU layer; The first ReLU layer is positioned between the first convolutional layer and the second convolutional layer; The second ReLU layer is positioned between the second convolutional layer and the third convolutional layer.

5. The magnetic particle imaging reconstruction method based on time-frequency dual information according to any one of claims 1-4, characterized in that, The training method for the time-frequency dual-branch neural network is as follows: Acquire a set number of low-gradient MPI time-domain images, add a set amount of Gaussian noise to the images, obtain the corresponding clear and noise-free images, perform nonlinear normalization of the images, and obtain normalized MPI time-domain images. The low-gradient MPI time-domain image is transformed to the frequency domain by two-dimensional Fourier transform, and the image is normalized to obtain a normalized MPI frequency-domain image. Using the normalized MPI time-domain image and the normalized MPI frequency-domain image as training sample pairs, and the corresponding normalized clear and noise-free image as sample labels, the time-frequency dual-branch neural network is iteratively trained until the set training termination condition is met, and the trained time-frequency dual-branch neural network is obtained.

6. The magnetic particle imaging reconstruction method based on time-frequency dual information according to claim 5, characterized in that, The time-frequency dual-branch neural network has a loss function during training that is a fusion of the loss functions of the two branches of the time-frequency dual-branch neural network according to a set ratio.

7. The magnetic particle imaging reconstruction method based on time-frequency dual information according to claim 5, characterized in that, The performance of the time-frequency dual-branch neural network is evaluated using a set evaluation index.

8. The magnetic particle imaging reconstruction method based on time-frequency dual information according to claim 7, characterized in that, The established evaluation indicators include: The measures of image half-width, Dice coefficient, root mean square error, peak signal-to-noise ratio, and structural similarity index.

9. A magnetic particle imaging reconstruction system based on time-frequency dual information, characterized in that, The magnetic particle imaging reconstruction system includes: The image acquisition module acquires MPI time-domain images under low field strength gradient acquisition parameters of the MPI device; The image conversion module transforms the MPI time-domain image to the frequency domain using a two-dimensional Fourier transform to obtain an MPI frequency-domain image. The reconstruction module, based on the MPI time-domain image and the MPI frequency-domain image, performs image reconstruction through a trained time-frequency dual-branch neural network to obtain high-quality magnetic particle imaging reconstruction results. The time-frequency dual-branch neural network includes a frequency domain branch, a time domain branch, a fusion module, and an output layer; The frequency domain branch is used to extract features from the MPI frequency domain image. The temporal branch performs feature extraction of the MPI temporal image; The fusion module fuses the features extracted from the frequency domain branch and the features extracted from the time domain branch to obtain fused features; The output layer performs feature extraction of the fused features to obtain the magnetic particle imaging reconstruction result.

Images