Method and apparatus for magnetic particle imaging image reconstruction based on signal priors

Abstract

The application provides a magnetic particle imaging image reconstruction method and device based on signal prior, and belongs to the field of magnetic particle imaging. The method comprises the following steps: acquiring a magnetic particle imaging signal and an initial magnetic particle imaging image corresponding to the magnetic particle imaging signal; performing feature extraction on the magnetic particle imaging signal based on a squeezing excitation attention mechanism to obtain first signal features; performing feature fusion on the first signal features and the initial magnetic particle imaging image, and obtaining first noise features through feature encoding and decoding processing; and performing denoising on the initial magnetic particle imaging image through a reverse process of a Gaussian diffusion model by using the first noise features to obtain a reconstructed magnetic particle imaging image.

Description

Technical Field

[0001] This application relates to the field of magnetic particle imaging, and more specifically, to a method and apparatus for magnetic particle imaging image reconstruction based on signal prior. Background Technology

[0002] Magnetic particle imaging (MPI) is an emerging tracer-based medical imaging technique. Its principle involves utilizing the nonlinear magnetization response of superparamagnetic iron oxide nanoparticles (SPIONs) to an alternating magnetic field to generate signals, and then reconstructing a magnetic particle imaging image of the SPION concentration distribution based on these signals. MPI offers advantages such as real-time imaging, high spatiotemporal resolution, high sensitivity, and no ionizing radiation, and is currently widely used in various biomedical fields.

[0003] In MPI systems, effectively reducing interference information in imaging results to achieve accurate quantification of tracer concentration in samples has become a critical technical problem that urgently needs to be solved. Accurate quantification of tracer concentration in samples is mainly affected by factors such as concentration differences between adjacent signals, distance between signal sources, and background noise interference. When the distance between two MPI signal sources is small and the concentration difference between their SPIONs is large, the high-concentration signal can mask the low-concentration signal, making it difficult to effectively detect and accurately quantify the latter. Especially when the sample distance is small, signals are more prone to mutual interference, thus affecting effective signal differentiation and concentration quantification. Furthermore, background signals in MPI may drift nonlinearly over time, which traditional linear calibration methods struggle to effectively suppress, leading to image artifacts and quantitative bias. Simultaneously, the distribution of tracers in tissues may be uneven, and their dynamic metabolism and clearance processes in vivo may also cause changes in signal intensity, further exacerbating the bias in quantitative results.

[0004] Therefore, in the field of magnetic particle imaging, there is an urgent need to develop a method that can effectively recover the distribution of SPIONs and accurately quantify their concentration through image processing without significantly increasing hardware complexity and time costs. Summary of the Invention

[0005] In view of this, this application provides a method and apparatus for magnetic particle imaging image reconstruction based on signal prior.

[0006] One aspect of this application provides a magnetic particle imaging image reconstruction method based on signal prior, comprising: acquiring a magnetic particle imaging signal and its corresponding initial magnetic particle imaging image; extracting features from the magnetic particle imaging signal based on a squeeze-excited attention mechanism to obtain a first signal feature; fusing the first signal feature and the initial magnetic particle imaging image, and obtaining a first noise feature through feature encoding and decoding; and using the first noise feature to denoise the initial magnetic particle imaging image through the inverse process of a Gaussian diffusion model to obtain a reconstructed magnetic particle imaging image.

[0007] According to an embodiment of this application, the above-mentioned feature fusion of the first signal feature and the initial magnetic particle imaging image, and the obtaining of the first noise feature through feature encoding and decoding processing, includes: stitching the first signal feature and the initial magnetic particle imaging image in the channel dimension to obtain a first feature map; performing feature reconstruction processing on the first feature map to obtain a second feature map; and performing feature enhancement on the second feature map based on a dynamic range attention mechanism to obtain the first noise feature.

[0008] According to an embodiment of this application, the above-mentioned feature reconstruction processing of the first feature map to obtain the second feature map includes: downsampling and feature extraction of the first feature map using N cascaded feature coding units to obtain a third feature map, where N is a positive integer greater than 1; downsampling the third feature map to obtain a fourth feature map; upsampling the fourth feature map to obtain a fifth feature map; and upsampling the fifth feature map using N cascaded upsampling modules to obtain the second feature map, wherein the output of the i-th feature coding unit is input to the (N-i+1)-th upsampling module via a skip connection, where i is a positive integer not greater than N.

[0009] According to an embodiment of this application, the above method further includes: taking the input of the j-th feature coding unit as a first feature sub-map; obtaining a second feature sub-map based on the first feature sub-map through downsampling and feature extraction; and taking the second feature sub-map as the output of the j-th feature coding unit; wherein the output of the j-th feature coding unit is the input of the (j+1)-th feature coding unit, and 0 < j ≤ N.

[0010] According to an embodiment of this application, the above-mentioned method of obtaining a second feature sub-map based on the first feature sub-map through downsampling and feature extraction includes: downsampling the first feature sub-map to obtain a third feature sub-map; performing a fast Fourier transform on the third feature sub-map to obtain a frequency domain feature map; dividing the frequency domain feature map into an initial high-frequency feature map and an initial low-frequency feature map according to a preset frequency threshold; performing feature enhancement on the initial high-frequency feature map based on a Butterworth bandpass filter to obtain a high-frequency feature map; performing feature enhancement on the initial low-frequency feature map based on a Gaussian low-pass filter to obtain a low-frequency feature map; and concatenating the high-frequency feature map and the low-frequency feature map along the channel dimension to obtain the second feature sub-map.

[0011] According to an embodiment of this application, the above-mentioned feature enhancement of the second feature map based on the dynamic range attention mechanism to obtain the first noise feature includes: generating multiple interval masks based on multiple preset feature value intervals to identify whether the value of the second feature map is located within the feature value interval; and performing feature enhancement on the second feature map based on the multiple interval masks to obtain the first noise feature.

[0012] According to an embodiment of this application, the above-mentioned feature enhancement of the second feature map based on multiple interval masks to obtain a first noise feature includes: calculating the second norm of the second feature map and the multiple interval masks respectively, and performing normalization processing to obtain multiple attention weights; and obtaining the first noise feature by weighting the second feature map based on the multiple attention weights and the multiple interval masks.

[0013] According to an embodiment of this application, the above-mentioned feature extraction of the magnetic particle imaging signal based on the squeeze-excited attention mechanism to obtain a first signal feature includes: denoising the magnetic particle imaging signal based on Doshibe wavelets to obtain a first imaging signal; extracting features from the first imaging signal using multiple cascaded convolutional layers to obtain a second signal feature; enhancing the second signal feature based on the squeeze-excited attention mechanism module to obtain a third signal feature; performing global average pooling on the third signal feature to obtain a fourth signal feature; embedding and encoding the fourth signal feature to obtain a fifth signal feature; and decoding the fifth signal feature to obtain the second signal feature.

[0014] According to an embodiment of this application, the above method further includes: performing feature fusion on the first signal feature and the reconstructed magnetic particle imaging image, and obtaining an updated first noise feature through feature encoding and decoding; using the updated first noise feature, denoising the reconstructed magnetic particle imaging image through the inverse process of the Gaussian diffusion model, so as to update and obtain a reconstructed magnetic particle imaging image.

[0015] Another aspect of this application provides a magnetic particle imaging image reconstruction apparatus based on signal prior, comprising: a data acquisition module for acquiring a magnetic particle imaging signal and its corresponding initial magnetic particle imaging image; a signal feature extraction module for extracting features from the magnetic particle imaging signal based on a squeeze-excited attention mechanism to obtain a first signal feature; a noise feature extraction module for fusing the first signal feature and the initial magnetic particle imaging image and obtaining a first noise feature through feature encoding and decoding; and an image reconstruction module for denoising the initial magnetic particle imaging image using the first noise feature through the inverse process of a Gaussian diffusion model to obtain a reconstructed magnetic particle imaging image.

[0016] According to an embodiment of this application, by extracting features from the magnetic particle imaging signal, a first signal feature is obtained. Based on the first signal feature and the initial magnetic particle imaging image, a first noise feature is obtained. Thus, the magnetic particle imaging signal can be used as prior information. Using the first noise feature, the initial magnetic particle imaging image is denoised through the inverse process of the Gaussian diffusion model to obtain the reconstructed magnetic particle imaging image. Ultimately, the noise or redundant background of the magnetic particle imaging image is suppressed, and the restoration quality of the magnetic particle imaging image is enhanced. Attached Figure Description

[0017] The above and other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 A flowchart of a magnetic particle imaging image reconstruction method based on signal prior, according to an embodiment of this application, is shown;

[0019] Figure 2 A flowchart illustrating a magnetic particle imaging image reconstruction method based on signal prior, according to another embodiment of this application, is shown.

[0020] Figure 3 A schematic diagram of the architecture of a signal encoder for feature extraction of magnetic particle imaging signals according to an embodiment of this application is shown;

[0021] Figure 4 A schematic diagram of a model architecture for implementing feature fusion and feature encoding / decoding according to an embodiment of this application is shown;

[0022] Figure 5 A schematic diagram of the architecture of a frequency domain enhancement module according to an embodiment of this application is shown;

[0023] Figure 6 A block diagram of a magnetic particle imaging image reconstruction apparatus based on signal prior, implemented according to this application, is shown;

[0024] Figure 7 A schematic diagram illustrating the simulation data generation for training a model according to an implementation of this application is shown;

[0025] Figure 8 A flowchart illustrating a magnetic particle imaging image reconstruction method based on signal prior, according to another embodiment of this application, is shown.

[0026] Figure 9 A block diagram of an electronic device according to an embodiment of this application is shown. Detailed Implementation

[0027] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0028] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms "comprising," "including," etc., as used herein indicate the presence of the above-described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0029] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0030] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0031] In the embodiments of this application, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security, network security, and national security.

[0032] In the embodiments of this application, the user's authorization or consent was obtained before obtaining or collecting the user's personal information.

[0033] Figure 1 A flowchart of a magnetic particle imaging image reconstruction method based on signal prior, according to an embodiment of this application, is shown.

[0034] like Figure 1 As shown, the magnetic particle imaging image reconstruction method based on signal prior includes operations S110~S140.

[0035] In operation S110, the magnetic particle imaging signal and its corresponding initial magnetic particle imaging image are acquired.

[0036] Magnetic particle imaging signals are nonlinear magnetization responses generated by SPIONs under alternating magnetic fields, exhibiting as time-domain voltage signals containing higher harmonics. The initial magnetic particle imaging image is a spatial concentration distribution map converted from the acquired magnetic particle imaging signal using a reconstruction algorithm. The conversion from magnetic particle imaging signal to the initial magnetic particle imaging image can be achieved using mature algorithms and techniques, such as the system matrix method.

[0037] In operation S120, based on the squeeze-excited attention mechanism, feature extraction is performed on the magnetic particle imaging signal to obtain the first signal feature.

[0038] Squeeze-driven attention is a channel attention module widely used in computer vision. It allows neural networks to adaptively learn the importance weights of each feature channel, strengthen useful channels and suppress irrelevant channels, thereby improving the expressive power of the neural network. It has advantages such as lightweight structure and plug-and-play functionality.

[0039] In operation S130, the first signal feature and the initial magnetic particle imaging image are fused together, and the first noise feature is obtained through feature encoding and decoding.

[0040] In embodiments of this application, the first signal features and the initial magnetic particle imaging image can be processed based on a U-net (U-shaped network) architecture to obtain the first noise features. Feature fusion methods include, but are not limited to, stitching fusion, attention fusion, and gating fusion.

[0041] In operation S140, the initial magnetic particle imaging image is denoised using the first noise feature through the inverse process of the Gaussian diffusion model to obtain the reconstructed magnetic particle imaging image.

[0042] The Gaussian diffusion model is a generative probabilistic model. Its core idea is to transform data into pure random noise by simulating a forward process of gradually adding Gaussian noise. The inverse process of the Gaussian diffusion model is to recover the original data step by step from the noise. If a general Gaussian diffusion model is used to directly perform unconditional or image-conditional inverse denoising on the initial magnetic particle imaging image, the model struggles to distinguish between real magnetic particle signals and noise components in the image. Therefore, it is necessary to use the first noise feature as a guided noise prior, allowing the model to more accurately estimate the noise components that should be removed based on the known noise intensity prior, thus obtaining the reconstructed magnetic particle imaging image.

[0043] Through the embodiments of this application, feature extraction is performed on the magnetic particle imaging signal to obtain a first signal feature. Based on the first signal feature and the initial magnetic particle imaging image, a first noise feature is obtained. Thus, the magnetic particle imaging signal can be used as prior information. Using the first noise feature, the initial magnetic particle imaging image is denoised through the inverse process of the Gaussian diffusion model to obtain the reconstructed magnetic particle imaging image. Ultimately, the noise or redundant background of the magnetic particle imaging image is suppressed, and the restoration quality of the magnetic particle imaging image is enhanced.

[0044] According to an embodiment of this application, the magnetic particle imaging image reconstruction method based on signal prior further includes: performing feature fusion on the first signal feature and the reconstructed magnetic particle imaging image, and obtaining an updated first noise feature through feature encoding and decoding; using the updated first noise feature, denoising the reconstructed magnetic particle imaging image through the inverse process of the Gaussian diffusion model, so as to update the reconstructed magnetic particle imaging image.

[0045] Figure 2 A schematic flowchart of a magnetic particle imaging image reconstruction method based on signal prior, according to another embodiment of this application, is shown.

[0046] like Figure 2 As shown, after the magnetic particle imaging signal passes through the signal encoder, the first signal feature is obtained. The first signal feature is used as prior information and input together with the initial magnetic particle imaging image corresponding to the magnetic particle imaging signal into the Gaussian diffusion model. Denoising is performed based on label condition supervision. This process is iterated multiple times to obtain the final reconstructed magnetic particle imaging image.

[0047] In the Gaussian diffusion model, the original magnetic particle imaging image without any noise can be taken as x0. By simulating the Gaussian diffusion process, the magnetic particle imaging image x at time step t is obtained. t Add a Gaussian noise layer to get x t+1 That is, starting from x0, Gaussian noise is gradually added, and after T steps, x is obtained. T x TThis is the initial magnetic particle imaging image. Reconstructing the initial magnetic particle imaging image into a noise-free original magnetic particle imaging image is achieved through the inverse process of the Gaussian diffusion model, based on x... T The process iterates continuously to obtain x0. To achieve this reverse process, we need the magnetic particle imaging signal as prior information and the magnetic particle imaging image x corresponding to time step t. t Predicting noise layer ε t The parameters, then through ε t To achieve x t To x t-1 The reverse process. For example, using the initial magnetic particle imaging image x T Predicting the first noise feature ε using the first signal feature yields the first noise feature ε. T Subsequently, through the first noise feature ε T For the initial magnetic particle imaging image x T Perform the inverse Gaussian diffusion process to denoise, and obtain the reconstructed magnetic particle imaging image x. T-1 Subsequently, using the first signal features and the reconstructed magnetic particle imaging image x T-1 Continuing the prediction process, we obtain the updated first noise feature, ε. T-1 ; Utilizing the updated first noise feature ε T-1 The reconstructed magnetic particle imaging image x T-1 Continue performing the inverse Gaussian diffusion process to update the reconstructed magnetic particle imaging image x. T-2 By iterating the above process multiple times, a raw magnetic particle imaging image x0 without any noise can be obtained. In the embodiments of this application, the number of time steps T is a preset constant value, which can be continuously adjusted according to the reconstruction effect of the initial magnetic particle imaging image.

[0048] According to an embodiment of this application, feature extraction is performed on a magnetic particle imaging signal based on a squeeze-excited attention mechanism to obtain a first signal feature, including: denoising the magnetic particle imaging signal based on Doshiby wavelets to obtain a first imaging signal; extracting features from the first imaging signal using multiple cascaded convolutional layers to obtain a second signal feature; enhancing the second signal feature based on the squeeze-excited attention mechanism module to obtain a third signal feature; performing global average pooling on the third signal feature to obtain a fourth signal feature; embedding and encoding the fourth signal feature to obtain a fifth signal feature; and decoding the fifth signal feature to obtain the second signal feature.

[0049] Figure 3 A schematic diagram of the architecture of a signal encoder for feature extraction of magnetic particle imaging signals according to an embodiment of this application is shown.

[0050] like Figure 3As shown, the magnetic particle imaging signal is first denoised and normalized using Dobese wavelets to obtain the first imaging signal. Then, the first imaging signal undergoes feature extraction using three convolutional layers with 3×3 kernels and a squeeze-excited attention mechanism module to obtain the second signal features. The number of kernels in the three convolutional layers are 16, 8, and 4, respectively. Next, the second signal features are enhanced using the squeeze-excited attention mechanism module to obtain the third signal features. Then, global average pooling is performed on the third signal features to obtain the fourth signal features. Next, the fourth signal features are embedded and encoded using two first fully connected layers to obtain the fifth signal features. Finally, the fifth signal features are decoded using two second fully connected layers to decode accurate concentration information, obtaining the second signal features, which serve as prior information for subsequent inputs.

[0051] In the embodiments of this application, the feature extraction process of the signal encoder can be constrained by a concentration loss function, which can be expressed as:

[0052] (1);

[0053] in, This represents the actual concentration. To predict concentration, For embedding regularization, λ can be 0.001.

[0054] According to an embodiment of this application, a first signal feature and an initial magnetic particle imaging image are fused, and a first noise feature is obtained through feature encoding and decoding. This includes: stitching the first signal feature and the initial magnetic particle imaging image together in the channel dimension to obtain a first feature map; performing feature reconstruction processing on the first feature map to obtain a second feature map; and performing feature enhancement on the second feature map based on a dynamic range attention mechanism to obtain the first noise feature.

[0055] The first signal feature and the initial magnetic particle imaging image are stitched together in the channel dimension. This operation does not change the spatial size, but only increases the number of feature channels, so that the first signal feature and the initial magnetic particle imaging image can be used simultaneously in subsequent processing.

[0056] Feature reconstruction refers to the nonlinear transformation of the concatenated first feature map through convolutional layers, residual blocks, or encoder-decoder structures, aiming to capture key information in the first feature map, rather than noise or irrelevant details, so as to make the features more stable and have stronger generalization.

[0057] Dynamic range attention is an adaptive weighting mechanism that dynamically adjusts the weights of different amplitude ranges of the signal to enhance key information in the second feature map (such as strong responses of high-concentration samples or weak signals of low-concentration samples), while suppressing noise or redundant background and enhancing the image restoration quality. Its core is to dynamically calculate the importance weights of samples of different concentrations instead of using a fixed threshold.

[0058] According to an embodiment of this application, feature reconstruction processing is performed on the first feature map to obtain the second feature map, including: downsampling and feature extraction of the first feature map using N cascaded feature coding units to obtain the third feature map, where N is a positive integer greater than 1; downsampling the third feature map to obtain the fourth feature map; upsampling the fourth feature map to obtain the fifth feature map; and upsampling the fifth feature map using N cascaded upsampling modules to obtain the second feature map, wherein the output of the i-th feature coding unit is input to the (N-i+1)-th upsampling module through a skip connection, where i is a positive integer not greater than N.

[0059] According to an embodiment of this application, the above method further includes: using the input of the j-th feature coding unit as a first feature sub-map; obtaining a second feature sub-map based on the first feature sub-map through downsampling and feature extraction; using the second feature sub-map as the output of the j-th feature coding unit; wherein the output of the j-th feature coding unit is the input of the (j+1)-th feature coding unit, and 0 < j ≤ N.

[0060] Figure 4 A schematic diagram of a model architecture for implementing feature fusion and feature encoding / decoding according to an embodiment of this application is shown.

[0061] like Figure 4 As shown, the first feature map is obtained by feature fusion of the first signal and the initial magnetic particle imaging image. The first feature map is then passed through two cascaded feature coding units to obtain a third feature map. The input of the second feature coding unit is the output of the first feature coding unit. Each feature coding unit has the same function and includes a downsampling module and a frequency domain enhancement module. The third feature map then passes through a downsampling module and an upsampling module to obtain a fifth feature map. The fifth feature map is then passed through two cascaded upsampling modules to obtain a second feature map. The output of the first feature coding unit is used as the input of the cascaded second upsampling module via a skip connection, and the output of the second feature coding unit is used as the input of the cascaded first upsampling module via a skip connection.

[0062] According to an embodiment of this application, a second feature sub-map is obtained based on a first feature sub-map through downsampling and feature extraction, including: downsampling the first feature sub-map to obtain a third feature sub-map; performing a fast Fourier transform on the third feature sub-map to obtain a frequency domain feature map; dividing the frequency domain feature map into an initial high-frequency feature map and an initial low-frequency feature map according to a preset frequency threshold; performing feature enhancement on the initial high-frequency feature map based on a Butterworth bandpass filter to obtain a high-frequency feature map; performing feature enhancement on the initial low-frequency feature map based on a Gaussian low-pass filter to obtain a low-frequency feature map; and concatenating the high-frequency feature map and the low-frequency feature map along the channel dimension to obtain the second feature sub-map.

[0063] Figure 5 A schematic diagram of the architecture of a frequency domain enhancement module according to an embodiment of this application is shown.

[0064] like Figure 5 As shown, the frequency domain enhancement module includes a high-frequency enhancement submodule, a low-frequency enhancement submodule, and a feature fusion submodule. The high-frequency enhancement submodule includes a Butterworth bandpass filter and an 8x8 convolutional kernel to suppress noise and enhance the details of the high-frequency magnetization response. The low-frequency enhancement submodule includes a Gaussian low-pass filter and a 16x16 convolutional kernel to enhance the stability of low-frequency background signals and slow magnetization changes. For the first feature map, downsampling and Fast Fourier Transform are first performed to obtain the frequency domain feature map. Then, the high-frequency enhancement submodule enhances the high-frequency portion of the frequency domain feature map to obtain the high-frequency feature map, and the low-frequency enhancement submodule enhances the low-frequency portion of the frequency domain feature map to obtain the low-frequency feature map. Finally, the feature fusion submodule concatenates the high-frequency and low-frequency feature maps along the channel dimension, adaptively fusing the high-frequency and low-frequency features while retaining key information to obtain the second feature map.

[0065] According to an embodiment of this application, based on a dynamic range attention mechanism, feature enhancement is performed on a second feature map to obtain a first noise feature, including: generating multiple interval masks based on multiple preset feature value intervals to identify whether the value of the second feature map is located within the feature value interval; calculating the second norm of the second feature map and the multiple interval masks respectively, and performing normalization processing to obtain multiple attention weights; and obtaining the first noise feature by weighting the second feature map based on the multiple attention weights and the multiple interval masks.

[0066] In the embodiments of this application, the value of the second feature map F can first be divided into K intervals (such as high response interval, medium response interval, and low response interval), and each interval corresponds to a different attention strategy:

[0067] (2);

[0068] in, This is a preset dynamic threshold.

[0069] Then, attention weights are calculated based on the feature values ​​within each interval:

[0070] (3);

[0071] in, For the Sigmoid function, To prevent the smallest positive number with a denominator of 0, This is the interval mask corresponding to the k-th interval.

[0072] An interval mask can be expressed by the following formula:

[0073] (4);

[0074] Finally, the attention weights are applied to the second feature map F to obtain the enhanced result, which is then used as the first noise feature. :

[0075] (5).

[0076] Figure 6 A block diagram of a magnetic particle imaging image reconstruction apparatus based on signal prior, implemented according to this application, is shown.

[0077] like Figure 6 As shown, the magnetic particle imaging image reconstruction device based on signal prior includes a data acquisition module 610, a signal feature extraction module 620, a noise feature extraction module 630, and an image reconstruction module 640.

[0078] The data acquisition module 610 is used to acquire magnetic particle imaging signals and their corresponding initial magnetic particle imaging images.

[0079] The signal feature extraction module 620 is used to extract features from the magnetic particle imaging signal based on the squeeze-excited attention mechanism to obtain the first signal feature.

[0080] The noise feature extraction module 630 is used to fuse the first signal features and the initial magnetic particle imaging image, and obtain the first noise features through feature encoding and decoding.

[0081] The image reconstruction module 640 is used to denoise the initial magnetic particle imaging image by using the first noise feature and through the inverse process of the Gaussian diffusion model to obtain the reconstructed magnetic particle imaging image.

[0082] The first signal feature and the initial magnetic particle imaging image are fused together, and the first noise feature is obtained through feature encoding and decoding. This process can be achieved through a noise estimation model.

[0083] Figure 7 A schematic diagram of simulation data generation for training a model, implemented according to this application, is shown.

[0084] like Figure 7 As shown, the original images of multi-concentration samples are binary images of dot-like or tumor-shaped features generated by an algorithm. These are directly input into the magnetic nanoparticle imaging reconstruction framework to obtain noise-free signals and noise-free MPI simulation images. Gaussian noise is then superimposed onto the original images of the multi-concentration samples; that is, the noise from the device acquired during the actual acquisition process is superimposed onto the original images of the multi-concentration samples, and then input together into the magnetic nanoparticle imaging reconstruction framework to obtain noisy signals and noisy MPI simulation images. The noisy signals and noisy MPI simulation images are used as inputs to the noise estimation model, and the noise-free MPI simulation images are used as labels for the noise estimation model to train the model parameters. The generated noise-free and noisy signals are one-dimensional vectors, and the noise-free and noisy MPI simulation images are grayscale images with grayscale values ​​distributed from 0 to 255. The noise level in the noisy signal can be calculated using the following formula:

[0085] (6);

[0086] in, The mean of the MPI signal. This represents the standard deviation of the noise.

[0087] In the embodiments of this application, the size, position, gray value (corresponding concentration), and number of samples in the noiseless MPI simulation image and the noisy MPI simulation image are randomly distributed to simulate the results of magnetic particle imaging in real-world situations.

[0088] Figure 8 A schematic flowchart of a magnetic particle imaging image reconstruction method based on signal prior, according to another embodiment of this application, is shown.

[0089] like Figure 8 As shown, firstly, noisy signals and noisy MPI simulation images, along with their corresponding noise-free labels, are generated according to the method described above. Using the noisy signals and noisy MPI simulation images as input, a noise estimation model is trained by combining a signal encoder and a noise estimation model. The model parameters are updated through training, and the trained noise estimation model is saved. After 1000 epochs (rounds), the optimal noise estimation model is saved. For any real noisy multi-concentration sample MPI images and signals acquired, they are used as the initial magnetic particle imaging image and magnetic particle imaging signal, respectively. Image reconstruction is performed using the trained noise estimation model to obtain an MPI image with accurately quantifiable concentrations.

[0090] Figure 9A block diagram of an electronic device according to an embodiment of this application is shown.

[0091] Figure 9 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0092] like Figure 9 As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0093] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 902 and / or RAM 903. It should be noted that the aforementioned programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in one or more of the aforementioned memories.

[0094] According to embodiments of this application, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0095] According to embodiments of this application, the method flow according to embodiments of this application can be implemented as a computer software program. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by processor 901, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0096] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0097] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0098] For example, according to embodiments of this application, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0099] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this application. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the magnetic particle imaging image reconstruction method based on signal prior provided in the embodiments of this application.

[0100] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0101] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0102] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0103] 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 a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may 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. Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations are not explicitly described in this application. In particular, without departing from the spirit and teachings of this application, the features described in the various embodiments of this application can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of this application.

[0104] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A magnetic particle imaging image reconstruction method based on signal prior, characterized in that, include: Acquire magnetic particle imaging signals and their corresponding initial magnetic particle imaging images; The magnetic particle imaging signal is denoised based on the Doceby wavelet to obtain the first imaging signal; The first imaging signal is used to extract features by using multiple cascaded convolutional layers to obtain the second signal features; The second signal feature is enhanced based on the squeeze-excitation attention mechanism module, and a third signal feature is obtained. Perform a global average pooling operation on the third signal feature to obtain the fourth signal feature; The fourth signal feature is embedded and encoded to obtain the fifth signal feature; The fifth signal feature is decoded to obtain the first signal feature; The first signal feature and the initial magnetic particle imaging image are stitched together in the channel dimension to obtain the first feature map; The first feature map is subjected to feature reconstruction processing to obtain the second feature map; Based on the dynamic range attention mechanism, feature enhancement is performed on the second feature map to obtain the first noise feature; Using the first noise feature, the initial magnetic particle imaging image is denoised through the inverse process of the Gaussian diffusion model to obtain the reconstructed magnetic particle imaging image.

2. The method according to claim 1, characterized in that, The step of performing feature reconstruction processing on the first feature map to obtain the second feature map includes: The first feature map is downsampled and its features are extracted using N cascaded feature coding units to obtain the third feature map, where N is a positive integer greater than 1; The third feature map is downsampled to obtain the fourth feature map; The fourth feature map is upsampled to obtain the fifth feature map; The fifth feature map is upsampled using N cascaded upsampling modules to obtain the second feature map. The output of the i-th feature encoding unit is input to the (N-i+1)-th upsampling module via a skip connection, where i is a positive integer not greater than N.

3. The method according to claim 2, characterized in that, The method further includes: The input of the j-th feature encoding unit is used as the first feature sub-graph; Based on the first feature sub-map, a second feature sub-map is obtained through downsampling and feature extraction; The second feature sub-map is used as the output of the j-th feature encoding unit; The output of the j-th feature encoding unit is the input of the (j+1)-th feature encoding unit, where 0 < j ≤ N.

4. The method according to claim 3, characterized in that, The process of obtaining a second feature sub-map based on the first feature sub-map through downsampling and feature extraction includes: The first feature sub-image is downsampled to obtain the third feature sub-image; Perform a Fast Fourier Transform on the third feature sub-map to obtain a frequency domain feature map; The frequency domain feature map is divided into an initial high-frequency feature map and an initial low-frequency feature map according to a preset frequency threshold. The initial high-frequency feature map is enhanced using a Butterworth bandpass filter to obtain a high-frequency feature map. The initial low-frequency feature map is enhanced using a Gaussian low-pass filter to obtain a low-frequency feature map. The high-frequency feature map and the low-frequency feature map are concatenated along the channel dimension to obtain the second feature sub-map.

5. The method according to claim 1, characterized in that, The dynamic range attention mechanism is used to enhance the second feature map to obtain the first noise feature, including: Based on multiple preset feature value intervals, multiple interval masks are generated to identify whether the value of the second feature map is located within the feature value interval; Based on multiple interval masks, feature enhancement is performed on the second feature map to obtain the first noise feature.

6. The method according to claim 5, characterized in that, The step of enhancing the second feature map based on multiple interval masks to obtain the first noise feature includes: The second feature map and the multiple interval masks are respectively subjected to L2 norm calculation and normalization to obtain multiple attention weights; Based on the multiple attention weights and multiple interval masks, the first noise feature is obtained by weighting the second feature map.

7. The method according to claim 1, characterized in that, The method further includes: The first signal feature and the reconstructed magnetic particle imaging image are fused together, and the updated first noise feature is obtained through feature encoding and decoding. Using the updated first noise feature, the reconstructed magnetic particle imaging image is denoised through the inverse process of the Gaussian diffusion model to update the reconstructed magnetic particle imaging image.

8. A magnetic particle imaging image reconstruction device based on signal prior, characterized in that, include: The data acquisition module is used to acquire magnetic particle imaging signals and their corresponding initial magnetic particle imaging images; The signal feature extraction module is used to denoise the magnetic particle imaging signal based on Doxibe wavelets to obtain a first imaging signal; extract features from the first imaging signal using multiple cascaded convolutional layers to obtain a second signal feature; enhance the second signal feature based on a squeeze-excited attention mechanism module to obtain a third signal feature; perform global average pooling on the third signal feature to obtain a fourth signal feature; embed the fourth signal feature into an encoding to obtain a fifth signal feature; and decode the fifth signal feature to obtain a first signal feature. The noise feature extraction module is used to stitch the first signal feature and the initial magnetic particle imaging image in the channel dimension to obtain a first feature map; perform feature reconstruction processing on the first feature map to obtain a second feature map; and perform feature enhancement on the second feature map based on the dynamic range attention mechanism to obtain a first noise feature. The image reconstruction module is used to denoise the initial magnetic particle imaging image by using the first noise feature and through the inverse process of the Gaussian diffusion model to obtain the reconstructed magnetic particle imaging image.

Images