Projection type magnetic particle imaging sparse visual angle reconstruction method, system and equipment

Abstract

The invention belongs to the field of medical image reconstruction, particularly relates to a projection type magnetic particle imaging sparse view angle reconstruction method, system and equipment, and aims to solve the problem of insufficient accuracy of a sparse view angle reconstruction technology. The method comprises the following steps: acquiring three-dimensional projection data of a to-be-imaged target at a sampling projection angle; segmenting the three-dimensional projection data into a plurality of plane slices, and obtaining a space coordinate vector and a sampling projection angle of any plane slice; generating input data based on the neighborhood set of the sampling projection angle of each plane slice, the space coordinate vector and the sampling projection angle, and constructing a projection completion network by taking the magnetic particle concentration signal as output data; optimizing the projection completion network by obtaining an uncertainty index; generating a complementation projection based on the optimized projection complementation network; and performing back projection reconstruction on the complemented projection to obtain a reconstructed image of the to-be-imaged target. According to the invention, high-accuracy magnetic particle imaging can be realized through less data.

Description

Technical Field

[0001] The present application belongs to the field of medical image reconstruction, and specifically relates to a method, system and equipment for sparse perspective reconstruction of projected magnetic particle imaging. Background Art

[0002] Magnetic Particle Imaging (MPI) is a tomographic imaging technology based on the nonlinear magnetic response characteristics of superparamagnetic nanoparticles. It has the advantages of no ionizing radiation, high sensitivity (detection of microgram-level iron concentrations) and real-time imaging potential. It has shown important application value in the fields of dynamic monitoring of cardiovascular disease and tracking of tumor-targeted drug release.

[0003] Current MPI scanning modes are mainly divided into two categories: one is magnetic field-free point scanning, which achieves point-by-point imaging through three-dimensional magnetic field focusing, with high spatial resolution but low acquisition efficiency; the other is magnetic field-free line scanning, which generates linear field-free regions through one-dimensional magnetic field gradients and combines them with projection rotation to achieve rapid imaging. It has significant advantages in temporal resolution but is limited by the projection angle sampling density. In projection-based magnetic particle imaging, the traditional filtered back projection (FBP) algorithm requires that the Nyquist sampling theorem be met (typically, the sampling projection angles must be greater than 120 projection angles), otherwise the reconstructed image will exhibit severe stripe artifacts.

[0004] To improve temporal resolution, existing research mainly reduces the number of projections through the following methods: data-driven deep learning: using networks such as U-Net to learn the mapping relationship from sparse to dense projections, but relying on large-scale annotated datasets, while MPI clinical data are scarce; compressed sensing theory: using signal sparsity constraints to optimize weights, but lacks adaptability to dynamic scenarios (such as drug diffusion processes). Summary of the Invention

[0005] In order to solve the above-mentioned problem in the prior art, namely, the problem of insufficient accuracy of sparse perspective reconstruction technology, the first aspect of the present application proposes a sparse perspective reconstruction method for projected magnetic particle imaging, comprising:

[0006] Step S10, obtaining three-dimensional projection data of the target to be imaged at a sampling projection angle based on a projection magnetic particle imaging device;

[0007] Step S20, dividing the three-dimensional projection data into a plurality of plane slices, and obtaining the spatial coordinate vector and sampling projection angle of any plane slice;

[0008] In step S30, input data is generated based on the neighborhood set of the sampling projection angle, the spatial coordinate vector and the sampling projection angle of each plane slice, and a magnetic particle concentration signal is taken as output data to construct a projection completion network, wherein the projection completion network is a fully connected network trained in a self-supervised manner.

[0009] In step S40, an uncertainty index is obtained by feature dropping of the projection completion network, wherein the uncertainty index is used for optimizing the projection completion network.

[0010] In step S50, the projection of the unsampled projection angle is completed based on the optimized projection completion network to generate a completed projection.

[0011] In step S60, the completed projection is back-projected and reconstructed to obtain a reconstructed image of the target to be imaged.

[0012] As a preferred embodiment, the projection completion network is constructed, comprising:

[0013] The spatial coordinate vector, the sampling projection angle and the neighborhood set of each plane slice are respectively encoded by Fourier feature mapping with a preset encoding order, and the encoding results are spliced to generate input encoding;

[0014] The input encoding is input into the projection completion network, and a magnetic particle concentration signal is output.

[0015] The projection completion network is trained based on the difference between the magnetic particle concentration signal and the true magnetic particle concentration under the sampling projection angle.

[0016] As a preferred embodiment, the projection completion network is trained, comprising:

[0017] A sampling projection angle mask is randomly selected in a sparse angle set to generate a sparse angle subset, wherein the sparse angle set is a set of all sampling projection angles.

[0018] A plurality of sparse angle subsets are taken as training data, and the difference between the magnetic particle concentration signal corresponding to the sampling projection angle in the sparse angle subset and the true magnetic particle concentration is taken as a loss function to train the projection completion network until the loss function converges.

[0019] As a preferred embodiment, the input encoding acquisition process further comprises:

[0020] The local spectral energy density of the target data is obtained, wherein the target data is a spatial coordinate vector or a sampling projection angle or a neighborhood set.

[0021] The encoding coefficients are generated for any order encoding based on the local spectral energy density, wherein the encoding coefficients are used to adjust the frequency components of any order encoding.

[0022] The adjusted frequency components are synthesized into a vector as the input code.

[0023] As a preferred implementation, generating a complementary projection includes:

[0024] Determine the projection weight coefficient of the projection completion network based on the uncertainty index, wherein the uncertainty index is negatively correlated with the projection weight coefficient;

[0025] The projection weight coefficient is used as the weight of the loss function in the projection completion network to obtain the optimized projection completion network;

[0026] The spatial coordinate vectors of each plane slice and the unsampled projection angle are input into the optimized projection completion network, and the completed projection is output.

[0027] As a preferred implementation, the uncertainty index is obtained by discarding features of the projection completion network, including:

[0028] Randomly discard some features of the projection completion network to generate output results after discarding multiple features;

[0029] Calculate the difference between each output result as an uncertainty indicator.

[0030] As a preferred embodiment, back-projection reconstruction is performed on the supplementary projection to obtain a reconstructed image of the target to be imaged, including:

[0031] Generate an initial reconstructed image of the complementary projection through a weighted filtered back-projection algorithm;

[0032] Based on the difference and similarity between the output result of the random noise in the image segmentation network and the initial reconstructed image, the initial reconstructed image is refined to obtain a reconstructed image.

[0033] As a preferred embodiment, the supplementary projection is a sine diagram of a dense perspective after supplementing the sine diagram of a sparse perspective, wherein the sine diagram is used to characterize the mapping relationship between the spatial coordinates of the magnetic particle signal and the projection angle.

[0034] In a second aspect of the present application, a projected magnetic particle imaging sparse perspective reconstruction system is proposed, comprising:

[0035] A three-dimensional projection data acquisition module is used to acquire three-dimensional projection data of the target to be imaged at a sampling projection angle based on a projection magnetic particle imaging device;

[0036] A segmentation module is used to segment the three-dimensional projection data into multiple plane slices and obtain the spatial coordinate vector and sampling projection angle of any plane slice;

[0037] The projection completion network construction module is configured to generate input data based on a neighborhood set of sampling projection angles, a spatial coordinate vector and the sampling projection angles of each planar slice, generate a magnetic particle concentration signal as output data, and construct a projection completion network, wherein the projection completion network is a fully connected network trained in a self-supervised manner.

[0038] The projection completion network optimization module is configured to obtain an uncertainty index by performing feature dropout on the projection completion network, and optimize the projection completion network based on the uncertainty index.

[0039] The projection completion module is configured to complete the projection of the unsampled projection angles based on the optimized projection completion network, and generate a completed projection.

[0040] The image reconstruction module is configured to perform back-projection reconstruction on the completed projection, and obtain a reconstructed image of the target to be imaged.

[0041] In a third aspect, the application provides an electronic device, comprising:

[0042] at least one processor; and

[0043] a memory in communication with the at least one processor; wherein

[0044] The memory stores instructions executable by the processor, and the instructions are used to implement the projection magnetic particle imaging sparse view reconstruction method described above.

[0045] The application has the following advantages:

[0046] (1) By combining spatial coordinates and projection angles, the consistency of sparse projection is jointly completed, which can realize accurate magnetic particle imaging with less data, eliminate the dependence on large-scale data sets, reduce the reconstruction requirements of magnetic particle images, improve the time resolution of tomographic imaging of projection magnetic particle imaging, and can be applied to different projection magnetic particle imaging devices to suppress residual artifacts in magnetic particle imaging in various scenarios.

[0047] (2) When constructing the projection completion network, by combining the neighborhood set of projection angles, the local angle context information is integrated into the data of spatial coordinates and projection angles, which can effectively capture the projection change trend between continuous angles, improve the extrapolation ability of the projection completion network for missing angles, enhance the rationality and accuracy of the completed projection, and improve the projection expression integrity under sparse angles.

[0048] (3) The uncertainty index obtained by feature discarding is used to optimize the projection completion network, a dynamic modulation mechanism is introduced, which can enhance the perception ability of high-frequency areas, improve the adaptability of Fourier position coding in different scale areas, and improve the reconstruction accuracy in detail-intensive areas. BRIEF DESCRIPTION OF DRAWINGS

[0049] Other features, objects, and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments, when read in conjunction with the accompanying drawings:

[0050] Figure 1 is a flow chart of a projection magnetic particle imaging sparse view reconstruction method provided by an embodiment of the application;

[0051] Figure 2 is a system block diagram of a projection magnetic particle imaging sparse view reconstruction system provided by an embodiment of the application;

[0052] Figure 3 is a structural schematic diagram of a computer system of a server for implementing the method, system, and device embodiments of the application. DETAILED DESCRIPTION

[0053] The application will be further described below in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related application, and not to limit the application. In addition, it should be noted that only the parts related to the application are shown in the drawings for ease of description.

[0054] It should be noted that the embodiments and features in the embodiments of the application can be combined with each other without conflict. The application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0055] The present application provides a sparse perspective reconstruction method for projection-type magnetic particle imaging. The method is based on a projection-type magnetic particle imaging device and obtains three-dimensional projection data of a target to be imaged at a sampling projection angle; divides the three-dimensional projection data into multiple plane slices, and obtains the spatial coordinate vector and sampling projection angle of any plane slice; generates input data based on a neighborhood set, spatial coordinate vector, and sampling projection angle of the sampling projection angle of each plane slice, and constructs a projection completion network using a magnetic particle concentration signal as output data, wherein the projection completion network is a fully connected network trained through self-supervision; obtains an uncertainty index by discarding features of the projection completion network, wherein the uncertainty index is used to optimize the projection completion network; based on the optimized projection completion network, completes the projection of the unsampled projection angle to generate a completed projection; and performs back-projection reconstruction on the completed projection to obtain a reconstructed image of the target to be imaged. By combining spatial coordinates and projection angles to jointly achieve consistent completion of sparse projections, it is possible to achieve highly accurate magnetic particle imaging with less data, eliminating dependence on large-scale data sets, reducing the requirements for magnetic particle image reconstruction, and improving the temporal resolution of tomographic imaging of projection-type magnetic particle imaging. Furthermore, the method can be applied to different projection-type magnetic particle imaging devices to suppress residual artifacts during magnetic particle imaging in a variety of scenarios.

[0056] In order to more clearly explain the sparse perspective reconstruction method of projection magnetic particle imaging of the present application, the following is combined with Figure 1 Each step in the embodiments of the present application is described in detail.

[0057] The projected magnetic particle imaging sparse perspective reconstruction method of the first embodiment of the present application includes steps S10 to S60, each of which is described in detail as follows:

[0058] Step S10 : obtaining three-dimensional projection data of the target to be imaged at a sampling projection angle based on a projection magnetic particle imaging device.

[0059] Optionally, a projection magnetic particle imaging device (PMPI) is a magnetic particle imaging device with higher scanning efficiency, larger coverage field of view, and higher imaging resolution. It achieves magnetic particle imaging by covering the imaging field of view through multi-angle projection. In the embodiment of the present application, the original three-dimensional projection data of the target to be imaged is collected by the projection magnetic particle imaging device, which is recorded as D(x, y, z, θ).

[0060] Step S20 , dividing the three-dimensional projection data into a plurality of plane slices, and obtaining the spatial coordinate vector and sampling projection angle of any plane slice.

[0061] Optionally, the 3D projection data can be sliced ​​along any coordinate axis to generate multiple planar slices. The slice thickness is set based on the system resolution and signal-to-noise ratio to balance the reconstruction accuracy and the number of training samples.

[0062] For example, in the embodiment of the present application, the three-dimensional projection data is cut into a plurality of two-dimensional XY plane slices along the Z axis, wherein the slice thickness Δz=50 μm.

[0063] For any plane slice, if the resolution is N×N, the plane slice includes N 2 pixel points, the pixel spacing is Δx, Δy, then each pixel point can be expressed as: P j =(x j ,y j ), where x j =j x *Δx,y j =j y *Δy,j=1,...,N 2 , represents the j-th pixel, j x 、j y is the pixel index.

[0064] It is understandable that the three-dimensional projection data is obtained by multi-angle projection. Accordingly, any plane slice corresponds to a sampling projection angle. In the embodiment of the present application, the projection angle set is represented as: Θ = {θ k |θ k =2π*k / K, k=0, 1, ..., K-1}, where K represents the number of sampled projection angles. For example, K=180, and there are 180 sampled projection angles in total, which means that projection is performed every 2 degrees, and each sampled projection angle corresponds to a set of two-dimensional projection signals.

[0065] It should be noted that the pixels of the same plane slice correspond to the same sampling projection angle.

[0066] Step S30, generating input data based on the neighborhood set of the sampling projection angle of each plane slice, the spatial coordinate vector and the sampling projection angle, and constructing a projection completion network with the magnetic particle concentration signal as output data, wherein the projection completion network is a fully connected network trained by self-supervision.

[0067] Alternatively, the neighborhood set of the k-th sampling projection angle can be expressed as: Where δ is the neighborhood window size, for example, δ can be set to 2.

[0068] The trend of the sampling projection angle is reflected by the neighborhood set, input data is generated by combining the spatial coordinate vector and the sampling projection angle for information collaborative modeling, a projection completion network is constructed based on the input data and the magnetic particle concentration signal, and the projection completion network is constructed by combining the angle and the coordinate to complete the unsampled angle.

[0069] In the embodiment of the application, the spatial coordinate vector, the sampling projection angle and the neighborhood set of each plane slice are respectively encoded by Fourier feature mapping with a preset encoding order, and the encoding results are spliced to generate input encoding; the input encoding is input into the projection completion network, and a magnetic particle concentration signal is output; the projection completion network is trained based on the difference between the magnetic particle concentration signal and the true magnetic particle concentration under the sampling projection angle.

[0070] The projection completion network learns the spatial variation of different frequencies by the preset encoding order, and high-frequency information capture is achieved.

[0071] As an example, the preset encoding order can be 10, and in the embodiment of the application, the encoding is realized by Fourier feature mapping, which maps low-dimensional coordinates to high-dimensional space through a set of sine (sin) and cosine (cos) functions of different frequencies, thereby enhancing the model's perception ability of spatial structure. The dimension of the encoded spatial coordinate vector is 20, the dimension of the encoded angle is 20, and the obtained input data is 40-dimensional data.

[0072] The encoding function can be represented by the following formula:

[0073] γ(v)=[sin(2 0 πv),cos(2 0 πv),...,sin(2 L-1 πv),cos(2 L-1 πv)]

[0074] Wherein, v represents the variable to be encoded, γ(v) represents the encoding function of the variable v, and L represents the preset encoding order.

[0075] In the embodiment of the application, the variable v can be a spatial coordinate vector or a sampling projection angle.

[0076] As a possible implementation, the spatial coordinate vector and the sampling projection angle are respectively encoded by Fourier feature mapping with a preset encoding order to generate a first encoding result of the spatial coordinate vector and a second encoding result of the sampling projection angle; each neighborhood angle in the neighborhood set is encoded by Fourier mapping, and the obtained encoding results are average-pooled to obtain a third encoding result of the neighborhood set; the first encoding result, the second encoding result and the third encoding result are spliced to obtain input encoding.

[0077] Furthermore, the input code can also be dynamically modulated. In an embodiment of the present application, the local spectral energy density of the target data is obtained, where the target data is a spatial coordinate vector or a sampling projection angle or a neighborhood set; based on the local spectral energy density, coding coefficients are generated for any order of coding, where the coding coefficients are used to adjust the frequency components of any order of coding; the adjusted frequency components are synthesized into a vector as the input code.

[0078] As an example, the local spectral energy density of the target data can be calculated using the following formula:

[0079]

[0080] Wherein, v represents the encoded variable (target data in the embodiment of the present application), and E(v) represents the local spectral energy density.

[0081] Furthermore, the coding amplitude of each order is modulated according to the local spectrum energy density:

[0082] γ dyn(v) =[α0sin(2 0 πv), β0cos(2 0 πv),...,α L-1 sin(2 L-1 πv), β L-1 cos(2 L-1 πv)]

[0083] Among them, γ dyn(v) represents the input code, α l , β l Represents the coding coefficient, l=0,1,...,L-1.

[0084] As an example, α l , β l ∝E(v) η , η represents the control parameter, ∝ represents proportional to, the proportional coefficient can be obtained by random sampling, the sampling range is [0,1], and the control parameter η can be taken as 0.5.

[0085] Furthermore, each angle in the neighborhood set is encoded by Fourier mapping to obtain the encoding result γ(θ t ), in, represents the neighborhood set of the kth sampling projection angle, θ t Represents any neighborhood angle in the neighborhood angle set. Then, the coding results of each neighborhood angle in the neighborhood set are average pooled to obtain the third coding result AvgPool(γ(θ t ), Finally, the first encoding result, the second encoding result and the third encoding result are spliced to obtain input encoding of the jth pixel point under the kth sampling projection angle: wherein γ(P j ) represents the first encoding result of the jth pixel point, γ(θ k ) represents the second encoding result of the kth sampling projection angle, AvgPool(γ(θ t ) represents the third encoding result of the neighborhood set of the kth sampling projection angle. The input encoding of the pixel points included in each slice plane under the corresponding sampling projection angle constitutes the input data of the projection completion network.

[0086] The encoding can enhance the nonlinear mapping capability, maintain the sparsity of the input, and thus improve the generalization capability of the projection completion network under a small amount of angle training data.

[0087] In the embodiments of the present application, the projection completion network is a multi-layer perceptron architecture, and the specific structure configuration is as follows: input layer: accepting a vector with a dimension of 2L; hidden layer: 5 fully connected layers, each layer containing 512 neurons, and the activation function being ReLU; output layer: 1 neuron, using a Sigmoid activation function to normalize the output to [0, 1], representing the relative value of the magnetic particle concentration.

[0088] The multi-layer perceptron structure with Fourier encoding has strong function approximation capability, can effectively capture the complex change rule of the magnetic particle signal in the spatial and angle domains, and has excellent completion capability and angle generalization capability.

[0089] Further, a sampling projection angle is randomly selected for shielding in the sparse angle set to generate a sparse angle subset, wherein the sparse angle set is a set of all sampling projection angles; the difference between the magnetic particle concentration signal corresponding to the sampling projection angle in the sparse angle subset and the true magnetic particle concentration is taken as a loss function, and the projection completion network is trained until the loss function converges.

[0090] By randomly shielding a projection sampling angle in the sparse angle set in each iteration, a plurality of sparse angle subsets K train The corresponding loss function can be represented by the following formula by supervising the projection completion network learning through the sparse angle subset:

[0091]

[0092] wherein, represents the loss function value, |K train | represents the number of elements in the sparse angle subset K train ,​ Indicates input encoding Input the magnetic particle concentration signal output from the projection completion network, represents the actual magnetic particle concentration of the j-th pixel at the k-th projection sampling angle, express The square of the L2 norm of .

[0093] The projection completion network is trained using the above loss function until the loss function converges, and a trained projection completion network is obtained.

[0094] The projection completion network of the embodiment of the present application does not rely on real full-angle data. Through the construction of input coding, it can learn the intrinsic correlation between angles and achieve relatively accurate prediction of magnetic particle concentration. In addition, the projection completion network is a self-supervised fully connected network. It does not require manual labeling during the training process and has a simple structure, which further improves the accuracy of prediction results when data is scarce.

[0095] Step S40 , obtaining an uncertainty index by performing feature discarding on the projection completion network, wherein the uncertainty index is used to optimize the projection completion network.

[0096] Optionally, in order to avoid possible erroneous predictions that may cause accumulated errors in image reconstruction, the projection completion network is further optimized by calculating uncertainty indicators.

[0097] In an embodiment of the present application, some features of the projection completion network are randomly discarded to generate output results after multiple features are discarded; and the difference between each output result is calculated as an uncertainty indicator.

[0098] Among them, the projection completion network is used to discard features through multiple Dropout samplings, and the output results of the projection completion network after feature discarding of the input data are obtained, and the uncertainty index is determined based on the difference in the output results.

[0099] As an example, the difference can be the variance, and the uncertainty index can be calculated by the following formula:

[0100] t=1,...,T.

[0101] in, represents the uncertainty index of the j-th pixel at the k-th projection sampling angle, Var represents the variance function, It represents the output result of the t-th feature discard, and T represents the number of Dropout samplings.

[0102] Further, the projection weighting coefficient of the projection completion network is determined based on an uncertainty index, where the uncertainty index is negatively correlated with the projection weighting coefficient.

[0103] As an example, the projection weighting coefficient can be determined by the following formula

[0104]

[0105] Wherein, μ is an adjustment factor, which is preset according to actual conditions, for example, in the embodiment of the present application, it can be taken as 5.

[0106] Further, the projection weighting coefficient is taken as the weight of the loss function in the projection completion network, and an optimized projection completion network is obtained.

[0107] Through the projection weighting coefficient, the automatic weight reduction of low confidence can be realized, and the robustness of the quality of the back projection image is effectively improved.

[0108] Step S50, based on the optimized projection completion network, the projection of the unsampled projection angle is completed, and the completed projection is generated.

[0109] Optionally, the projection completion network is trained by sparse angles, learns the internal relationship between the sampling projection angles and the relationship between the spatial coordinates and the sampling projection angles and the magnetic particle concentration, and when the unsampled angle is input, the magnetic particle concentration of the unsampled angle can be predicted.

[0110] In the embodiment of the present application, the spatial coordinate vector of each plane slice and the unsampled projection angle are input into the optimized projection completion network, and the completed projection is output.

[0111] It can be understood that the completed projection is a dense view angle sinogram after completing the sparse view angle sinogram, where the sinogram is used to represent the mapping relationship between the spatial coordinates of the magnetic particle signal and the projection angle.

[0112] The embodiment of the present application combines the spatial coordinates and the projection angle, and jointly realizes the consistency completion of the sparse projection, which can realize the magnetic particle imaging with high accuracy through less data, and get rid of the dependence on large-scale data sets.

[0113] Step S60, back projection reconstruction is performed on the completed projection, and the reconstructed image of the target to be imaged is obtained.

[0114] Optionally, after obtaining the accurate completed projection, the target to be imaged can be back projected and reconstructed based on the completed projection to generate the final reconstructed image.

[0115] In some embodiments, the reconstructed image can be generated by weighted filter back projection algorithm.

[0116] In an embodiment of the present application, an initial reconstructed image of the completed projection is generated by a weighted filtered back projection algorithm; based on the difference and similarity between the output result of the random noise in the image segmentation network and the initial reconstructed image, the initial reconstructed image is refined to obtain a reconstructed image.

[0117] Among them, the weighted filtering back projection algorithm is a reconstruction algorithm commonly used in this field and will not be described in detail in the embodiments of this application.

[0118] As a possible implementation method, an image segmentation network G with a U-Net architecture is constructed. ψ , used to reconstruct the initial image f o Perform image refinement, where the loss function of the image segmentation network is:

[0119]

[0120] in, represents the loss function value; λ1, λ2 and λ3 are weight parameters respectively; G ψ (z) represents the output result of random noise z in the image segmentation network; f0 represents the initial reconstructed image; Represents G ψ The similarity loss between (z) and f0 is achieved through the structural similarity loss function (SSIM); Represents the marginal loss, which is achieved by calculating the gradient difference.

[0121] As an example, the edge loss can be calculated as follows:

[0122]

[0123] in, is the image gradient operator (Sobel), and || ||1 represents the L1 norm.

[0124] By setting weight parameters, we can improve edge restoration and spatial consistency, avoid over-smoothing of weak texture areas by U-Net, and improve image spatial coherence and edge clarity by introducing edge loss.

[0125] As an example, the weight parameters may be set to: λ1 = 1.0, λ2 = 0.2, λ3 = 0.1.

[0126] Furthermore, by monitoring PSNR (G ψ (z), f0), determine whether the loss function converges, for example, in PSNR (G ψ When the improvement of (z), f0) is less than 0.1dB for 10 consecutive iterations, the loss function is determined to have converged and the optimization is terminated.

[0127] Among them, the process of using PSNR (peak signal-to-noise ratio) to optimize model convergence is an existing technology and will not be repeated in the embodiments of this application.

[0128] The embodiment of the present application further eliminates residual structural artifacts through image refinement to generate a final reconstructed image.

[0129] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of this application.

[0130] See also Figure 2 The projection magnetic particle imaging sparse perspective reconstruction system of the second embodiment of the present application includes: a three-dimensional projection data acquisition module 100, a segmentation module 200, a projection completion network construction module 300, a projection completion network optimization module 400, a projection completion module 500 and an image reconstruction module 600.

[0131] The three-dimensional projection data acquisition module 100 is used to acquire three-dimensional projection data of the target to be imaged at a sampling projection angle based on a projection magnetic particle imaging device;

[0132] A segmentation module 200 is used to segment the three-dimensional projection data into a plurality of plane slices and obtain the spatial coordinate vector and sampling projection angle of any plane slice;

[0133] A projection completion network construction module 300 is used to generate input data based on a neighborhood set of sampling projection angles of each plane slice, a spatial coordinate vector, and the sampling projection angle, and to construct a projection completion network using a magnetic particle concentration signal as output data, wherein the projection completion network is a fully connected network trained through self-supervision;

[0134] A projection completion network optimization module 400 is configured to obtain an uncertainty index by discarding features of the projection completion network, wherein the uncertainty index is used to optimize the projection completion network;

[0135] The projection completion module 500 is used to complete the projection of the unsampled projection angle based on the optimized projection completion network to generate a completed projection;

[0136] The image reconstruction module 600 is used to perform back-projection reconstruction on the supplementary projection to obtain a reconstructed image of the target to be imaged.

[0137] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0138] It should be noted that the projected magnetic particle imaging sparse perspective reconstruction system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present application can be further decomposed or combined. For example, the modules in the above embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present application are only for distinguishing the modules or steps and are not considered to be improper limitations of the present application.

[0139] An electronic device according to a third embodiment of the present application includes:

[0140] at least one processor; and

[0141] a memory communicatively connected to at least one of the processors; wherein,

[0142] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned projection magnetic particle imaging sparse perspective reconstruction method.

[0143] A fourth embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned projected magnetic particle imaging sparse perspective reconstruction method.

[0144] A computer program product according to a fifth embodiment of the present application, when running on an electronic device, enables the electronic device to execute the above-mentioned method for sparse perspective reconstruction of projected magnetic particle imaging.

[0145] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes and related instructions of the electronic device, computer-readable storage medium, and computer program product described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0146] Those skilled in the art should be able to appreciate that, in conjunction with the modules and method steps of each example described in the embodiments disclosed herein, it is possible to implement them with electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0147] Reference below Figure 3 , which shows a structural diagram of a computer system of a server for implementing the method, system, and device embodiments of the present application. Figure 3 The server shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0148] like Figure 3 As shown, the computer system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 302 or the program loaded from the storage part 308 to the random access memory (RAM) 303. Various programs and data required for system operation are also stored in the RAM 303. The CPU 301, ROM 302 and RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0149] The following components are connected to the I / O interface 305: an input section 306 including a keyboard, a mouse, and the like; an output section 307 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 308 including a hard disk and the like; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. Removable media 311, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 310 as needed, so that computer programs read therefrom can be installed into the storage section 308 as needed.

[0150] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 309, and / or installed from the removable medium 311. When the computer program is executed by the central processing unit (CPU) 301, the above-described functions defined in the methods of the present application are executed. Note that the computer readable medium described above in the present application can be either a computer readable signal medium or a computer readable storage medium or any combination of these two. The computer readable storage medium may, for example, be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of these. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of these. In the present application, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device. In the present application, the computer readable signal medium can include a data signal carried in a baseband or as part of a carrier wave, in which the computer readable program code is carried. Such a propagated data signal can take any of a variety of forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device. Program code contained on a computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of these.

[0151] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone 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 a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0152] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0153] The terms "first", "second", etc. are used to distinguish similar objects, rather than to describe or indicate a particular order or sequence.

[0154] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0155] Thus far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of the present application is obviously not limited to these specific embodiments. Without departing from the principles of the present application, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present application.

Claims

1. A method for sparse perspective reconstruction of projected magnetic particle imaging, characterized in that: include: Step S10, obtaining three-dimensional projection data of the target to be imaged at a sampling projection angle based on a projection magnetic particle imaging device; Step S20, dividing the three-dimensional projection data into a plurality of plane slices, and obtaining the spatial coordinate vector and sampling projection angle of any plane slice; Step S30, generating input data based on the neighborhood set of the sampling projection angle of each plane slice, the spatial coordinate vector and the sampling projection angle, and constructing a projection completion network with the magnetic particle concentration signal as output data, wherein the projection completion network is a fully connected network trained by self-supervision; Step S40, obtaining an uncertainty index by discarding features of the projection completion network, wherein the uncertainty index is used to optimize the projection completion network; Step S50, based on the optimized projection completion network, completing the projection of the unsampled projection angle to generate a completed projection; Step S60 , performing back-projection reconstruction on the supplementary projection to obtain a reconstructed image of the target to be imaged.

2. The method for sparse perspective reconstruction of projected magnetic particle imaging according to claim 1, characterized in that: The construction of the projection completion network includes: Encoding the spatial coordinate vector, the sampling projection angle, and the neighborhood set of each plane slice with a preset encoding order through Fourier eigenmapping, and concatenating the encoding results to generate input code; Inputting the input code into the projection completion network to output a magnetic particle concentration signal; The projection completion network is trained based on the difference between the magnetic particle concentration signal and the true magnetic particle concentration at the sampling projection angle.

3. The method for sparse perspective reconstruction of projected magnetic particle imaging according to claim 2, characterized in that: The training of the projection completion network includes: Randomly selecting a sampled projection angle mask from the sparse angle set to generate a sparse angle subset, wherein the sparse angle set is a set of all sampled projection angles; The projection completion network is trained using the plurality of sparse angle subsets as training data and the difference between the magnetic particle concentration signal corresponding to the sampled projection angle in the sparse angle subset and the actual magnetic particle concentration as the loss function until the loss function converges.

4. The method for sparse perspective reconstruction of projected magnetic particle imaging according to claim 2, characterized in that: The process of obtaining the input code further includes: Acquire local spectral energy density of target data, wherein the target data is the spatial coordinate vector or the sampling projection angle or the neighborhood set; generating a coding coefficient for any order coding based on the local spectral energy density, wherein the coding coefficient is used to adjust the frequency component of the any order coding; The adjusted frequency components are synthesized into a vector as the input code.

5. The method for sparse perspective reconstruction of projected magnetic particle imaging according to claim 1, characterized in that: Generating the completed projection includes: Determining a projection weighting coefficient of the projection completion network based on the uncertainty indicator, wherein the uncertainty indicator is negatively correlated with the projection weighting coefficient; Using the projection weight coefficient as the weight of the loss function in the projection completion network to obtain an optimized projection completion network; The spatial coordinate vectors and unsampled projection angles of each planar slice are input into the optimized projection completion network, and the completed projection is output.

6. The method for sparse perspective reconstruction of projected magnetic particle imaging according to claim 1 or 5, characterized in that: The step of discarding features from the projection completion network to obtain an uncertainty indicator includes: Randomly discarding some features of the projection completion network to generate output results after discarding multiple features; The difference between each output result is calculated as the uncertainty index.

7. The method for sparse perspective reconstruction of projected magnetic particle imaging according to claim 1, characterized in that: The back-projection reconstruction of the supplementary projection to obtain a reconstructed image of the target to be imaged includes: generating an initial reconstructed image of the supplementary projection by a weighted filtered back-projection algorithm; Based on the difference and similarity between the output result of the random noise in the image segmentation network and the initial reconstructed image, image thinning is performed on the initial reconstructed image to obtain the reconstructed image.

8. The method for sparse perspective reconstruction of projected magnetic particle imaging according to claim 1, characterized in that: The supplementary projection is a sinogram of a dense perspective after supplementing a sinogram of a sparse perspective, wherein the sinogram is used to represent the mapping relationship between the spatial coordinates of the magnetic particle signal and the projection angle.

9. A projection-type magnetic particle imaging sparse perspective reconstruction system, characterized in that: include: A three-dimensional projection data acquisition module is used to acquire three-dimensional projection data of the target to be imaged at a sampling projection angle based on a projection magnetic particle imaging device; A segmentation module, configured to segment the three-dimensional projection data into a plurality of planar slices and obtain a spatial coordinate vector and a sampling projection angle of any planar slice; a projection completion network construction module, configured to generate input data based on a neighborhood set of the sampling projection angles of each plane slice, the spatial coordinate vector, and the sampling projection angles, and to construct a projection completion network using a magnetic particle concentration signal as output data, wherein the projection completion network is a fully connected network trained through self-supervision; a projection completion network optimization module, configured to obtain an uncertainty index by discarding features of the projection completion network, wherein the uncertainty index is used to optimize the projection completion network; The projection completion module is used to complete the projection of unsampled projection angles based on the optimized projection completion network and generate completed projections; An image reconstruction module is used to perform back-projection reconstruction on the supplementary projection to obtain a reconstructed image of the target to be imaged.

10. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the sparse perspective reconstruction method for projected magnetic particle imaging according to any one of claims 1 to 8.

Images