Superconducting magnetic nanoparticle imaging reconstruction method and system based on double gradient fusion

CN122510371BActive Publication Date: 2026-09-15BEIHANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610999143.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-07
Publication Date
2026-09-15
Estimated Expiration
2046-07-07

AI Technical Summary

Technical Problem

[0006]为了解决现有技术中的上述问题,即现有超导MPI图像重建方法在低信噪比条件下难以同时兼顾高灵敏度与高空间分辨率,重建图像存在局部失真的问题,本申请的第一方面提出了一种基于双梯度融合的超导磁纳米粒子成像重建方法,所述方法包括以下步骤:

Benefits of technology

通过同时输入低梯度模式和高梯度模式的MPI信号,利用低梯度图像提供稳定的轮廓先验,再通过级联双输入内容感知模块逐步注入高梯度图像的细节信息,从而在保持高空间分辨率的同时大幅提升重建图像的灵敏度,从而避免超导MPI因高梯度场导致信噪比下降所引起的局部失真问题。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122510371B_ABST
    Figure CN122510371B_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of magnetic nanoparticle imaging, and particularly relates to a superconducting magnetic nanoparticle imaging reconstruction method and system based on double-gradient fusion, aiming to solve the problem that superconducting MPI is difficult to balance high sensitivity and high resolution under low signal-to-noise ratio. The method comprises the following steps: acquiring MPI signals under different scanning modes, respectively solving the system matrix pseudo-inverse to obtain low-gradient MPI images and high-gradient MPI images; obtaining a contour feature map based on the low-gradient MPI images; inputting the contour feature map and the high-gradient MPI images into a first DCA module of a cascade DCA module, inputting the high-gradient MPI images and the fusion image output by the previous module into each subsequent DCA module; and taking the output of the last DCA module as a reconstructed image. The application utilizes the high signal-to-noise ratio prior of low gradient and the high resolution details of high gradient, improves the reconstruction quality of superconducting MPI images through cascade fusion, and balances high sensitivity and high spatial resolution.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of magnetic nanoparticle imaging technology, specifically involving a superconducting magnetic nanoparticle imaging reconstruction method and system based on dual gradient fusion. Background Technology

[0002] Magnetic Particle Imaging (MPI) is an emerging medical imaging technology that, with its advantages of high temporal and spatial resolution, shows broad application prospects in fields such as vascular imaging, tumor diagnosis, and drug tracing.

[0003] Superconducting MPI devices can provide higher magnetic field gradients than non-superconducting devices with lower coil power consumption, resulting in superior spatial resolution. However, while the introduction of superconducting coils increases the magnetic field gradient and spatial resolution, it also significantly reduces the area of ​​the free field point (FFP) and the number of magnetic nanoparticles within a single FFP, leading to a decrease in the sensitivity and signal-to-noise ratio (SNR) of MPI scans. Low sensitivity and low SNR often cause local distortion in superconducting MPI images, severely limiting the quality of image reconstruction.

[0004] To overcome the inherent limitations of traditional reconstruction methods, especially the quality issues caused by the low SNR of superconducting MPI, generative networks have been widely applied in medical image reconstruction tasks in recent years. These networks significantly enhance the performance of medical imaging technologies such as CT, PET, and MRI, enabling the recovery of high-quality images from low-dose, sparsely sampled, or high-noise data. In SM-based MPI reconstruction, generative networks can achieve end-to-end reconstruction, avoiding the numerical instability problems of solving ill-conditioned equations in traditional methods. They can also more fully utilize local spatial correlations and global semantic information, preserving fine structures such as edge textures while suppressing noise artifacts. However, most existing generative networks are designed for a single scanning mode and fail to effectively utilize the complementary characteristics between the low-gradient scanning mode (high SNR, low resolution) and high-gradient scanning mode (low SNR, high resolution) of superconducting MPI devices.

[0005] Therefore, there is an urgent need for an image reconstruction method that can combine the advantages of two scanning modes and simultaneously improve the sensitivity and spatial resolution of superconducting MPI images. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, namely, the difficulty in simultaneously achieving high sensitivity and high spatial resolution under low signal-to-noise ratio conditions and the existence of local distortion in the reconstructed images, the first aspect of this application proposes a superconducting magnetic nanoparticle imaging reconstruction method based on dual gradient fusion, the method comprising the following steps:

[0007] Acquire the first one-dimensional MPI signal under low gradient scanning mode and the second one-dimensional MPI signal under high gradient scanning mode; The system matrix pseudo-inverse is solved by respectively solving the first one-dimensional MPI signal and the second one-dimensional MPI signal to obtain the corresponding low-gradient MPI image and high-gradient MPI image. The low-gradient MPI image is oversampled and contour segmented by a pre-trained encoder-decoder network to obtain a contour feature map with the same resolution as the high-gradient MPI image. The contour feature map and the high-gradient MPI image are input into multiple cascaded dual-input content-aware (DCA) modules for feature fusion. Each DCA module contains a multi-scale convolutional layer and a feature fusion layer. The input of the first DCA module is the contour feature map and the high-gradient MPI image, and the input of each subsequent DCA module is the high-gradient MPI image and the fused image output by the previous DCA module. The output of the final DCA module is used as the reconstructed target MPI image.

[0008] In some preferred embodiments, the encoder-decoder network is pre-trained, specifically as follows: Obtain the actual MPI concentration distribution image as the label image; The low-gradient scanning mode and high-gradient scanning mode of the superconducting MPI device are simulated respectively. According to the preset signal-to-noise ratio calculation formula, the signal amplitude corresponding to the tag image is adjusted and background noise is added to generate a first one-dimensional MPI signal in the high signal-to-noise ratio range and a second one-dimensional MPI signal in the low signal-to-noise ratio range respectively. The first one-dimensional MPI signal corresponds to the low-gradient scanning mode and the second one-dimensional MPI signal corresponds to the high-gradient scanning mode. The pseudo-inverse of the system matrix is ​​used to solve for the first and second one-dimensional MPI signals obtained from the simulation, respectively, to obtain the corresponding low-gradient MPI images and high-gradient MPI images as training samples; the encoder-decoder network is trained using the labeled images as supervision.

[0009] In some preferred embodiments, each DCA module processes the input, specifically including: The first and second inputs of the DCA module are convolved by the multi-scale convolutional layer to obtain the first and second multi-scale feature maps. The first multi-scale feature map and the second multi-scale feature map are fused by the feature fusion layer to obtain the fused image output by the DCA module.

[0010] In some preferred embodiments, feature fusion is performed on the first multi-scale feature map and the second multi-scale feature map, specifically as follows: Calculate the channel maximum value feature and channel average value feature of the second multi-scale feature map respectively; The maximum channel feature and the average channel feature are concatenated along the channel dimension, and then a mask is generated by 1×1 convolution and activation function. The activation mask is divided into a first gradient mask corresponding to the first input and a second gradient mask corresponding to the second input according to the input source, and the mask is averaged within each group. The first gradient mask is multiplied element-wise with the first multi-scale feature map, and the second gradient mask is multiplied element-wise with the second multi-scale feature map. The result of the multiplication is added to the corresponding input feature map, and the two added results are merged to obtain the fused image output by the DCA module.

[0011] In some preferred embodiments, when obtaining the target MPI image, a global skip connection is also introduced: the contour feature map and the high gradient MPI image are respectively added element-wise to the output of the last DCA module through skip connections, and the addition result is used as the target MPI image.

[0012] A second aspect of this application proposes a superconducting magnetic nanoparticle imaging and reconstruction system based on dual-gradient fusion, comprising: The signal acquisition module is used to acquire the first one-dimensional MPI signal in the low-gradient scanning mode and the second one-dimensional MPI signal in the high-gradient scanning mode. The preliminary reconstruction module is used to solve the pseudo-inverse of the system matrix for the first one-dimensional MPI signal and the second one-dimensional MPI signal respectively, and reconstruct them into a low-gradient MPI image and a high-gradient MPI image. The contour segmentation module is used to perform oversampling and contour segmentation on the low-gradient MPI image using an encoder-decoder network to obtain a contour feature map with the same resolution as the high-gradient MPI image. A cascaded fusion module is used to input the contour feature map and the high-gradient MPI image into multiple cascaded DCA modules for feature fusion. Each DCA module contains a multi-scale convolutional layer and a feature fusion layer. The input of the first DCA module is the contour feature map and the high-gradient MPI image, and the input of each subsequent DCA module is the high-gradient MPI image and the fused image output by the previous DCA module. The output module is used to obtain the target MPI image based on the output of the last DCA module.

[0013] The beneficial effects of this application are: By simultaneously inputting MPI signals in both low-gradient and high-gradient modes, the low-gradient image provides a stable contour prior, and then the detailed information of the high-gradient image is gradually injected through a cascaded dual-input content-aware module. This significantly improves the sensitivity of the reconstructed image while maintaining high spatial resolution, thereby avoiding the local distortion problem caused by the decrease in signal-to-noise ratio due to the high gradient field in superconducting MPI.

[0014] Simultaneously, it integrates multi-scale convolution and content-aware self-attention mechanisms, enabling the full extraction and utilization of local and global features. While suppressing noise and artifacts, it sharpens edges and texture structures, resulting in higher numerical stability. An encoder-decoder network performs oversampling and contour segmentation on low-gradient images, and a feature fusion module achieves step-by-step alignment and fusion of the two types of information. This fully leverages the complementary advantages of the two scanning modes of the superconducting MPI device, ensuring that the final image possesses both low-gradient contour integrity and high-gradient detail clarity.

[0015] The DCA cascaded fusion structure does not rely on specific system matrices or scanning parameters and can be adapted to different models of superconducting MPI equipment. Furthermore, the model training process simultaneously employs Dice loss (segmentation task) and L1+MS-SSIM loss (reconstruction task), ensuring the accuracy of contour prior extraction and the visual quality of the reconstructed image. This provides reliable technical support for the clinical application of superconducting MPI in high-sensitivity scenarios such as vascular imaging. Attached Figure Description

[0016] 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: Figure 1 This is a schematic flowchart of a superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion provided in one embodiment of this application; Figure 2 This is a schematic diagram of the structure of an encoder-decoder network provided in one embodiment of this application; Figure 3 This is a schematic diagram of the structure of a cascaded DCA module provided in one embodiment of this application; Figure 4 This is a schematic diagram of the structure of a single dual-input content-aware (DCA) module provided in one embodiment of this application; Figure 5 This is a schematic diagram of the structure of a computer system used to implement the methods, systems, and electronic devices of this application. Detailed Implementation

[0017] 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.

[0018] 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.

[0019] Based on a two-stream generative model, this application provides a superconducting magnetic nanoparticle imaging reconstruction method based on dual-gradient fusion. One-dimensional MPI signals under different scanning modes are acquired, and the pseudo-inverse of the system matrix is ​​performed to obtain low-gradient MPI images and high-gradient MPI images. A contour feature map is obtained based on the low-gradient MPI image, and the contour feature map and high-gradient MPI image are input into multiple cascaded DCA modules. The output of the final-stage DCA module is used as the reconstructed image. The technical solution of this invention leverages the high sensitivity of the low-gradient field to supplement the prior knowledge of the imaging object's contour in the superconducting MPI reconstruction algorithm, while using the high resolution of the high-gradient field to supplement image details, thereby improving the model's reconstruction performance for superconducting MPI images and enabling the detection of even lower magnetic nanoparticle concentrations.

[0020] To more clearly explain the superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion in this application, the following will combine... Figure 1 The steps in the embodiments of this application are described in detail.

[0021] The superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion in the first embodiment of this application, steps S100-S400, are described in detail below: S100: Acquire the first one-dimensional MPI signal under low gradient scanning mode and the second one-dimensional MPI signal under high gradient scanning mode.

[0022] In one specific embodiment, two scanning modes, low gradient and high gradient, are achieved by energizing the gradient coil of the superconducting MPI scanner to different degrees. The one-dimensional MPI signal obtained by the superconducting MPI device in the two modes is obtained respectively. The excitation frequency of 1MHz is used for scanning in both scanning modes. In practical applications, it can be adjusted appropriately according to the characteristics of the instrument.

[0023] S200. Solve the pseudo-inverse of the system matrix for the first one-dimensional MPI signal and the second one-dimensional MPI signal respectively to obtain the corresponding low-gradient MPI image and high-gradient MPI image.

[0024] Preferably, the pseudo-inverse of the system matrix is ​​solved for the first one-dimensional MPI signal and the second one-dimensional MPI signal respectively, specifically as follows: ; ; in, The system matrix (SM) is obtained by point-by-point scanning. To receive the first one-dimensional MPI signal or the second one-dimensional MPI signal generated by the receiving coil due to electromagnetic induction, This is a concentration distribution image of magnetic nanoparticles in MPI, where M is the frequency of the system matrix and N is the number of pixels in the MPI image. for The false rebellion.

[0025] S300. The low-gradient MPI image is oversampled and contour segmented using a pre-trained encoder-decoder network to obtain a contour feature map with the same resolution as the high-gradient MPI image.

[0026] Preferably, the encoder-decoder network is a U-net with a depth of three. Its encoder consists of three downsampling layers, each containing two convolutional layers and one max-pooling layer, with the number of output channels increasing sequentially among the three downsampling layers. The decoder consists of three corresponding upsampling layers, which restore spatial resolution through transposed convolutions and concatenate the feature maps of the corresponding layers of the encoder to the decoder through skip connections to preserve detailed information.

[0027] In one specific embodiment, the downsampling layer of the encoder includes two 3×3 convolutional layers with a stride of 1 and a 2×2 max pooling layer with a stride of 2, with the number of output channels being 64, 128, and 256 respectively, and the ReLU activation function is used after each convolutional layer.

[0028] More preferably, the encoder is: ; in, This serves as the input to the function. By introducing the ReLU function, the gradient explosion problem can be mitigated without increasing the computational cost, facilitating network parameter convergence and helping the network enhance effective features such as edges, textures, and contours.

[0029] The U-Net performs two tasks simultaneously: first, it upsamples the spatial resolution of the first image to be the same as that of the second image; second, it performs contour segmentation on the imaging object and outputs a binary contour map (i.e., contour prior knowledge). This contour prior provides stable shape information for subsequent image reconstruction, which helps to correct local distortions that may occur in high-gradient modes.

[0030] S400. Input the contour feature map and the high gradient MPI image into multiple cascaded DCA modules to perform feature fusion, and use the output of the final DCA module as the reconstructed target MPI image. Each DCA module contains a multi-scale convolutional layer and a feature fusion layer. The input to the first DCA module is the contour feature map and the high-gradient MPI image. The input to each subsequent DCA module is the high-gradient MPI image and the fused image output by the previous DCA module.

[0031] Preferably, the number of the plurality of cascaded DCA modules is at least a first preset number, and the first preset number is four.

[0032] Preferably, the multi-scale convolutional layers within each DCA module contain convolutional kernels of sizes including 1×1, 3×3, and 7×7.

[0033] Preferably, each DCA module processes the input, specifically including: The first and second inputs of the DCA module are convolved by the multi-scale convolutional layer to obtain the first and second multi-scale feature maps. The first multi-scale feature map and the second multi-scale feature map are fused by the feature fusion layer to obtain the fused image output by the DCA module.

[0034] More preferably, the first input of the first DCA module is the contour feature map, and the second input is the high-gradient MPI image; the first input of each subsequent DCA module is the fused image output by the previous DCA module, and the second input is the high-gradient MPI image. By introducing high-gradient MPI images multiple times, the details of the contour feature map are gradually filled in, thereby compensating for the lack of spatial resolution in the low-gradient scanning mode.

[0035] More preferably, feature fusion is performed on the first multi-scale feature map and the second multi-scale feature map, specifically as follows: Calculate the channel maximum value feature and channel average value feature of the second multi-scale feature map respectively; The maximum channel feature and the average channel feature are concatenated along the channel dimension, and then a mask is generated by 1×1 convolution and activation function. The activation mask is divided into a first gradient mask corresponding to the first input and a second gradient mask corresponding to the second input according to the input source, and the mask is averaged within each group. The first gradient mask is multiplied element-wise with the first multi-scale feature map, and the second gradient mask is multiplied element-wise with the second multi-scale feature map. The result of the multiplication is added to the corresponding input feature map, and the two added results are merged to obtain the fused image output by the DCA module.

[0036] In the DCA module, multi-scale convolution is used to extract low-gradient contour scale information and high-gradient detail scale information separately. Then, the content-aware module modulates the MPI features from the two different sources to achieve source-distinguished adaptive fusion. By establishing a step-by-step detail injection mechanism through cascaded DCA modules, high-resolution details in the high-gradient image are gradually supplemented under the constraints of the low-gradient contour, rather than being stitched together all at once.

[0037] Specifically, a content-aware (CA) module is used to connect the maximum and average values ​​of the channels of the features using a 1×1 convolutional kernel and pass them through a sigmoid activation function to obtain the mask information of the feature map.

[0038] Preferably, when obtaining the target MPI image, a global skip connection is also introduced: the contour feature map and the high-gradient MPI image are respectively skip-connected and added element-wise to the output of the last DCA module, and the sum is used as the target MPI image. This better preserves the MPI signal features under the two gradients.

[0039] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that in order to achieve the effects of the embodiments, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. 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 changes are all within the protection scope of this application.

[0040] The second embodiment of this application, based on the first embodiment, provides a specific example of the training process and illustrates the pre-training process of the encoder-decoder network and the DCA module. The pre-training method specifically includes the following steps: 1) Training data needs to be generated, specifically: Obtain the actual MPI concentration distribution image as the label image; The low-gradient scanning mode and high-gradient scanning mode of the superconducting MPI device are simulated respectively. Based on a preset signal-to-noise ratio (SNR) calculation formula, the signal amplitude corresponding to the tag image is adjusted and background noise is added to generate a first one-dimensional MPI signal in the high SNR range and a second one-dimensional MPI signal in the low SNR range, respectively. Specifically: ; in, Indicates the signal peak value of MNPs, The maximum noise peak value is indicated; the first one-dimensional MPI signal corresponds to the low gradient scan mode, and the second one-dimensional MPI signal corresponds to the high gradient scan mode; the signal-to-noise ratio of the low gradient scan mode is set to 30dB to 40dB, and the signal-to-noise ratio of the high gradient scan mode is set to 15dB to 25dB. The pseudo-inverse of the system matrix is ​​used to solve for the first and second one-dimensional MPI signals obtained from the simulation, respectively: ; ; in, The system matrix (SM) is obtained by point-by-point scanning. To receive the first one-dimensional MPI signal or the second one-dimensional MPI signal generated by the receiving coil due to electromagnetic induction, This is a concentration distribution image of magnetic nanoparticles in MPI, where M is the frequency of the system matrix and N is the number of pixels in the MPI image. for The false rebellion.

[0041] The corresponding low-gradient MPI image and high-gradient MPI image are obtained as training samples; the label image is used as supervision to form training sample pairs (first one-dimensional MPI signal, second one-dimensional MPI signal, label image and its contour label).

[0042] Based on a real MPI concentration distribution image (as a label image), the signal amplitude is adjusted according to the signal-to-noise ratio formula, and background noise acquired during a blank scan of the superconducting MPI scanner is added to simulate the actual acquisition environment in both modes. More preferably, the added background noise is the actual background noise acquired during a blank scan of the superconducting MPI scanner.

[0043] 2) Based on the training sample pairs obtained above, the U-net encoder-decoder network (responsible for super-resolution and contour segmentation) and the feature fusion module (composed of multiple cascaded DCA modules, responsible for fusion and reconstruction) were trained respectively. The training adopted a phased approach to ensure that the objectives of each sub-network were clear and the training was stable.

[0044] The training data consists of simulated MPI signals and their reconstructed images generated based on signal functions, including 50,000 data points each for high and low gradients. Specifically, the training data for Unet consists of 50,000 low-gradient scan-reconstructed MPI images; the training data for the dual-input fusion network consists of 50,000 low-scan gradient MPI feature maps extracted by Unet and their corresponding high-scan gradient reconstructed MPI images; and the training data for joint training consists of 50,000 pairs of high- and low-gradient reconstructed MPI images.

[0045] Phase 1: Training the U-net network.

[0046] Low-gradient MPI images are input into U-net to obtain contour feature maps upsampled to the target resolution. (with label image) (Same size) and the corresponding binary profile map ; Binary contour plotting based on loss function supervision The degree of matching between the contour label and the image's outline, and the resulting loss function, Dice loss, such as: ; in, For the predicted segmentation mask, This stage involves training independently and does not involve the feature fusion module. Training parameters for the U-net network are set according to requirements. For example, when using the Adam optimizer, the learning rate is set to 1e-5, the batch size is 32, the total training data is approximately 50,000 images, and the iterations are performed for 50 epochs.

[0047] Phase 2: Training the feature fusion module (with fixed U-net weights).

[0048] The trained U-net weights are fixed and used as the feature extraction front end. For each training sample, the following steps are performed: The low-gradient MPI image is fed into U-net to obtain the contour feature map. and the corresponding binary profile map Note that although U-net has segmentation output, only its output contour feature map is used in the feature fusion stage. (Contour priors are already included).

[0049] The contour feature map and high-gradient MPI image are taken as input and fed into a feature fusion network consisting of at least four cascaded DCA modules to output a reconstructed image. .

[0050] Computational reconstruction of images With label image The losses between them include L1 and MS-SSIM functions: ; ; ; in, For brightness contrast, For contrast comparison; For structural comparison; , These are the local pixel mean values ​​of image x and y calculated using Gaussian filtering. 、 Image x (referring to the reconstructed image) ) and y (referring to the label image) Local standard deviation; It is the local covariance of the image x and y; , , This small constant is set to avoid division by zero, and it is usually taken as... ; 、 、 These are parameters used to adjust the relative weights of different components. L1 loss ensures pixel-level accuracy, while MS-SSIM loss constrains image quality from a multi-scale structural similarity perspective. Only the parameters of the feature fusion module are updated during training.

[0051] After the two training phases are completed, U-net and the feature fusion module are concatenated into a complete network for end-to-end fine-tuning, with the total loss function being...

[0052] ; in, The coefficients of the MS-SSIM loss function are preferably in the range of 0.1 to 0.5. In this embodiment, the coefficients are... =0.2. In this embodiment, another 50 epochs can be trained together, reducing the learning rate to 1e-5, further optimizing the overall performance.

[0053] A third embodiment of this application proposes a superconducting magnetic nanoparticle imaging and reconstruction system based on dual gradient fusion, used to implement the superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion described in the above embodiments. The system includes: The signal acquisition module is used to acquire the first one-dimensional MPI signal in the low-gradient scanning mode and the second one-dimensional MPI signal in the high-gradient scanning mode. The preliminary reconstruction module is used to solve the pseudo-inverse of the system matrix for the first one-dimensional MPI signal and the second one-dimensional MPI signal respectively, and reconstruct them into a low-gradient MPI image and a high-gradient MPI image. The contour segmentation module is used to perform oversampling and contour segmentation on the low-gradient MPI image using an encoder-decoder network to obtain a contour feature map with the same resolution as the high-gradient MPI image. A cascaded fusion module is used to input the contour feature map and the high-gradient MPI image into multiple cascaded DCA modules for feature fusion. Each DCA module contains a multi-scale convolutional layer and a feature fusion layer. The input of the first DCA module is the contour feature map and the high-gradient MPI image, and the input of each subsequent DCA module is the high-gradient MPI image and the fused image output by the previous DCA module. The output module is used to obtain the target MPI image based on the output of the last DCA module.

[0054] Preferably, in a plurality of cascaded DCA modules, the first input of the first DCA module is the contour feature map, and the second input is the high-gradient MPI image; the first input of each subsequent DCA module is the fused image output by the previous DCA module, and the second input is the high-gradient MPI image. By introducing high-gradient MPI images multiple times, the details of the contour feature map are gradually filled in, thereby compensating for the lack of spatial resolution in the low-gradient scanning mode.

[0055] A device according to the fourth embodiment of this application includes: At least one processor; and a memory communicatively connected to at least one of the processors; The memory stores instructions that can be executed by the processor. These instructions are used to implement the superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion described above. When the processor executes the instructions, it is specifically configured to perform the steps of acquiring, predicting, and constructing the total loss function and training the model as described in the above embodiments.

[0056] A computer-readable storage medium according to a fifth embodiment of this application stores computer instructions that are executed by the computer to implement the above-described superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion.

[0057] 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.

[0058] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system for implementing embodiments of the systems, methods, and electronic devices of this application. Figure 5 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.

[0059] like Figure 5 As shown, the computer system includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in Read Only Memory (ROM) 502 or programs loaded from storage section 508 into Random Access Memory (RAM) 503. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0060] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0061] 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 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, 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. A 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.

[0062] 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).

[0063] 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.

[0064] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0065] 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.

[0066] The technical solutions of this application have 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 this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, 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 this application.

Claims

1. A method for superconducting magnetic nanoparticle imaging reconstruction based on dual-gradient fusion, characterized in that, Includes the following steps: Acquire the first one-dimensional MPI signal under low gradient scanning mode and the second one-dimensional MPI signal under high gradient scanning mode; The system matrix pseudo-inverse is solved by respectively solving the first one-dimensional MPI signal and the second one-dimensional MPI signal to obtain the corresponding low-gradient MPI image and high-gradient MPI image. The low-gradient MPI image is oversampled and contour segmented by a pre-trained encoder-decoder network to obtain a contour feature map with the same resolution as the high-gradient MPI image. The contour feature map and the high gradient MPI image are input into multiple cascaded DCA modules for feature fusion. Each DCA module contains a multi-scale convolutional layer and a feature fusion layer. The input of the first DCA module is the contour feature map and the high gradient MPI image. The input of each subsequent DCA module is the high gradient MPI image and the fused image output by the previous DCA module. The output of the final DCA module is used as the reconstructed target MPI image.

2. The dual-gradient fusion based superconducting magnetic nanoparticle imaging reconstruction method of claim 1, wherein, The encoder-decoder network is a U-net with a depth of three. Its encoder consists of three downsampling layers, each containing two convolutional layers and one max pooling layer. The number of output channels of the three downsampling layers increases sequentially. 3.The dual-gradient fusion based superconducting magnetic nanoparticle imaging reconstruction method of claim 2, wherein, The encoder-decoder network is pre-trained, specifically as follows: Obtain the actual MPI concentration distribution image as the label image; The low-gradient scanning mode and high-gradient scanning mode of the superconducting MPI device are simulated respectively. According to the preset signal-to-noise ratio calculation formula, the signal amplitude corresponding to the tag image is adjusted and background noise is added to generate a first one-dimensional MPI signal in the high signal-to-noise ratio range and a second one-dimensional MPI signal in the low signal-to-noise ratio range respectively. The first one-dimensional MPI signal corresponds to the low-gradient scanning mode and the second one-dimensional MPI signal corresponds to the high-gradient scanning mode. The pseudo-inverse of the system matrix is ​​used to solve for the first and second one-dimensional MPI signals obtained from the simulation, respectively, to obtain the corresponding low-gradient MPI images and high-gradient MPI images as training samples; the encoder-decoder network is trained using the labeled images as supervision.

4. The superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion according to claim 3, characterized in that, The added background noise is the actual background noise obtained during the empty scan of the superconducting MPI scanner.

5. The superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion according to claim 1, characterized in that, The number of the multiple cascaded DCA modules is at least a first preset number, and the first preset number is four.

6. The superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion according to claim 1, characterized in that, The multi-scale convolutional layers within each DCA module contain convolutional kernels of sizes including 1×1, 3×3, and 7×7.

7. The superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion according to claim 1, characterized in that, Each DCA module processes the input, specifically including: The first and second inputs of the DCA module are convolved by the multi-scale convolutional layer to obtain the first and second multi-scale feature maps. The first multi-scale feature map and the second multi-scale feature map are fused by the feature fusion layer to obtain the fused image output by the DCA module.

8. The superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion according to claim 7, characterized in that, Feature fusion is performed on the first multi-scale feature map and the second multi-scale feature map, specifically as follows: Calculate the channel maximum value feature and channel average value feature of the second multi-scale feature map respectively; The maximum channel feature and the average channel feature are concatenated along the channel dimension, and then a mask is generated by 1×1 convolution and activation function. The activation mask is divided into a first gradient mask corresponding to the first input and a second gradient mask corresponding to the second input according to the input source, and the mask is averaged within each group. The first gradient mask is multiplied element-wise with the first multi-scale feature map, and the second gradient mask is multiplied element-wise with the second multi-scale feature map. The result of the multiplication is added to the corresponding input feature map, and the two added results are merged to obtain the fused image output by the DCA module.

9. The superconducting magnetic nanoparticle imaging and reconstruction method based on dual gradient fusion according to claim 1, characterized in that, When obtaining the target MPI image, a global skip connection is also introduced: the contour feature map and the high gradient MPI image are respectively added element-wise to the output of the last DCA module through skip connections, and the sum is used as the target MPI image.

10. A superconducting magnetic nanoparticle imaging and reconstruction system based on dual gradient fusion, characterized in that, include: The signal acquisition module is used to acquire the first one-dimensional MPI signal in the low-gradient scanning mode and the second one-dimensional MPI signal in the high-gradient scanning mode. The preliminary reconstruction module is used to solve the pseudo-inverse of the system matrix for the first one-dimensional MPI signal and the second one-dimensional MPI signal respectively, and reconstruct them into a low-gradient MPI image and a high-gradient MPI image. The contour segmentation module is used to perform oversampling and contour segmentation on the low-gradient MPI image using an encoder-decoder network to obtain a contour feature map with the same resolution as the high-gradient MPI image. A cascaded fusion module is used to input the contour feature map and the high-gradient MPI image into multiple cascaded DCA modules for feature fusion. Each DCA module contains a multi-scale convolutional layer and a feature fusion layer. The input of the first DCA module is the contour feature map and the high-gradient MPI image, and the input of each subsequent DCA module is the high-gradient MPI image and the fused image output by the previous DCA module. The output module is used to obtain the target MPI image based on the output of the last DCA module.

Citation Information

Patent Citations

  • Magnetic particle imaging spatial resolution improving method based on FDNN model

    CN116934597A

  • Magnetic particle high-resolution reconstruction method based on rapid feature fusion

    CN120235980A