A subspace-aided multi-prior photon-counting spectral CT reconstruction method and device

Abstract

The application provides a subspace-assisted multi-prior photon counting spectral CT reconstruction method and device, and relates to the technical field of CT reconstruction. The method first obtains a noise feature map tensor and an orthogonal basis through subspace decomposition on a spectral CT image; then a reference image block is selected on the noise feature map tensor, and similar image blocks to the reference image block are extracted from all channels of the spectral CT image to generate a non-local similar full-channel tensor group; a global sparse regularization term about the non-local similar full-channel tensor group is constructed by using dictionary learning to obtain a denoised non-local similar full-channel tensor group, and the denoised non-local similar full-channel tensor group is multiplied by the orthogonal basis to obtain a clean tensor image; then a double regularization constraint reconstruction model is constructed according to TV regularization constraints of each channel of the clean tensor image and the spectral CT image; finally, the augmented Lagrange multiplier method is used to update the double regularization constraint reconstruction model until a preset reconstruction quality requirement is reached.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of CT reconstruction, and particularly relates to a subspace-assisted multi-prior photon counting spectral CT reconstruction method and device. BACKGROUND

[0002] Computed Tomography (CT) as a modern imaging technology has been widely used in medical, industrial and other fields. However, the traditional CT technology uses X-rays at a single energy for imaging, and the ability to distinguish the composition of the material is limited. In order to break through the imaging limitations of the traditional CT technology, spectral CT uses the difference of the attenuation coefficient of the material at different X-ray energies to obtain more abundant material information, has the material identification and analysis ability which the traditional CT technology does not have, and is being gradually researched and used as a new type of imaging technology.

[0003] The key step of the spectral CT reconstruction problem is to use the X-ray projection information collected by the detector at multiple energies to restore the distribution of different material information inside the scanned object, which directly determines the imaging quality of the spectral CT and affects the analysis, processing and application of the subsequent spectral CT reconstruction image. In order to obtain high-quality spectral CT imaging results, how to develop and design an efficient spectral CT reconstruction algorithm has become a hot and difficult problem in the field of X-ray imaging research, and has important research and application value.

[0004] In the field of spectral CT reconstruction, the traditional CT sparse regularization reconstruction method based on single energy assumption can be directly applied to spectral CT reconstruction. For example, in 2012, Xu et al. used total variation (TV) technology to reconstruct the region of interest (ROI) of spectral CT. In 2013, Zhao et al. proposed an iterative method based on tight frame, aiming to improve the image quality of spectral CT reconstruction; in 2016, Zeng et al. introduced the concept of structure tensor, and converted the TV regularization into a weighted least squares method, which achieved better results than traditional methods. These technologies achieve strong denoising effect by processing each channel image separately. However, they fail to fully utilize the high correlation between different channel images in spectral CT, resulting in loss of image texture and edge information.

[0005] To further exploit the prior information in images, tensor-based methods have started to play an increasingly important role in spectral CT reconstruction. In 2019, Xia et al. treated similar image patches in spectral CT images as third-order tensors and decomposed them into low-rank and sparse components through principal component analysis (PCA), providing an intuitive method to describe the non-local similarity features in spectral CT images. In 2021, Chen et al. further proposed a fourth-order non-local tensor decomposition model for spectral CT reconstruction. In 2022, Yu et al. proposed a framework set tensor nuclear norm to constrain the non-local similarity of spectral CT images, significantly improving the image quality. Despite this, these non-local-based methods largely rely on block matching operations to identify similar image patches, which not only makes them sensitive to noise but also increases significant computational burden. In the implementation of these algorithms, it is crucial to strike a balance between reconstruction quality and computational efficiency.

[0006] Meanwhile, the rapid development of deep learning technology has brought new vitality to data-driven spectral CT reconstruction methods. In 2021, Wu et al. first introduced a U-NET-based network for sparse spectral CT reconstruction. In 2023, Chen et al. developed SOUL-NET, a network that combines sparse and low-rank characteristics specifically for spectral CT reconstruction. By 2024, Wang et al. proposed the HITI-NET model, an interpretable hybrid domain integration transform iterative network that can simultaneously optimize the spectral CT reconstruction and material decomposition process. While these methods have made significant progress in image reconstruction performance, they often require processing a large number of parameters and have a high dependence on the characteristics of the training data set. In addition, deep learning technology also faces challenges such as data overfitting, computational resource limitations, insufficient model interpretability, and data privacy and security. SUMMARY

[0007] In view of the problems that the photon counting spectral CT image reconstruction time consumption is large, the reconstruction quality is poor, the decomposition noise is increased, and the quality of material decomposition is seriously affected, the present application provides a subspace-assisted multi-prior photon counting spectral CT reconstruction method and device.

[0008] The present application provides a subspace-assisted multi-prior photon counting spectral CT reconstruction method, comprising:

[0009] Step 1: generating a spectral CT image according to the projection data of spectral CT imaging, the spectral CT image comprising S-channel images;

[0010] Step 2: subspace decomposition of the spectral CT image to obtain a noise feature map tensor and an orthogonal basis;

[0011] Step 3: selecting a reference image block on the noise feature map tensor, extracting image blocks similar to the reference image block from all channels of the spectral CT image to generate a non-local similar full-channel tensor group;

[0012] Step 4: constructing a global sparse regularization term about the non-local similar full-channel tensor group by dictionary learning to obtain a denoised non-local similar full-channel tensor group, and multiplying the orthogonal basis to obtain a clean tensor image;

[0013] Step 5: constructing a double regularization constraint reconstruction model according to the clean tensor image and the TV regularization constraint of each channel of the spectral CT image;

[0014] Step 6: updating the double regularization constraint reconstruction model by using the augmented Lagrange multiplier method until the preset reconstruction quality requirement is reached.

[0015] Further, in step 1, a simultaneous algebraic reconstruction algorithm SART is used to generate the spectral CT image.

[0016] Further, step 3 specifically includes:

[0017] Step 3.1: selecting a reference image block on the noise feature map tensor, and extracting image blocks similar to the reference image block from one channel of the spectral CT image by using a block matching algorithm based on a graph domain distance, and grouping all the extracted image blocks to form a similar image block set;

[0018] Step 3.2: changing the position of the reference image block, repeating step 3.1 to obtain multiple different similar image block sets;

[0019] Step 3.3: performing steps 3.1 and 3.2 for all channels of the spectral CT image;

[0020] Step 3.4: vectorizing all the similar image block sets, and then stacking the vectorization results according to the channel order to obtain the non-local similar full-channel tensor group.

[0021] Further, the global sparse regularization term is:

[0022]

[0023] wherein, represents the i-th non-local similar full-channel tensor group in the expanded matrix form, represents the Frobenius norm, represents an adaptive dictionary, represents a group sparse coefficient, and ||·|| represents an L p represents Lp norm, 0 < p ≤ 1, denotes a weight matrix, denotes the Hadamard product of two matrices, τ denotes a regularization parameter.

[0024] Further, the reconstruction model with the double regularization constraints is:

[0025]

[0026] where A denotes a system matrix, x s denotes a spectral CT image of the s-th channel, p s denotes the projection data corresponding to the s-th channel, ||x s || TV denotes the TV norm of the s-th channel, the Frobenius norm term denotes a clean tensor image and the difference between the spectral CT image , ρ, λ are regularization parameters, denotes a global sparse regularization term with respect to the non-local similar full-channel tensor group .

[0027] Further, the step 6 specifically comprises:

[0028] introducing an auxiliary variable a Lagrange multiplier Λ and a non-negative penalty parameter β, the reconstruction model with the double regularization constraints is rewritten as:

[0029]

[0030] the above-mentioned rewritten model is solved by using an alternating direction multiplier method; wherein, the iteration formula is:

[0031]

[0032] wherein, denotes the spectral CT reconstruction image of this round.

[0033] In a second aspect, the present application provides a subspace-assisted multi-prior photon counting spectral CT reconstruction device, comprising:

[0034] an image generation module, configured to generate a spectral CT image according to the projection data of the spectral CT imaging, wherein the spectral CT image comprises S-channel images;

[0035] a subspace decomposition module, configured to perform subspace decomposition on the spectral CT image to obtain a noise feature map tensor and an orthogonal basis;

[0036] The tensor group generation module is configured to select a reference image block from the noise feature map tensor, extract similar image blocks from all channels of the spectral CT image to generate a non-local similar full-channel tensor group, and the like.

[0037] The sparsification module is configured to construct a global sparse regularization term about the non-local similar full-channel tensor group by dictionary learning to obtain a denoised non-local similar full-channel tensor group, and multiply the denoised non-local similar full-channel tensor group by the orthogonal basis to obtain a clean tensor image.

[0038] The reconstruction module is configured to construct a double regularization constraint reconstruction model according to the clean tensor image and the TV regularization constraint of each channel of the spectral CT image, and update the double regularization constraint reconstruction model by using a augmented Lagrange multiplier method until a preset reconstruction quality requirement is reached.

[0039] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method of the first aspect when executing the program.

[0040] In a fourth aspect, the present application provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the method of the first aspect.

[0041] The present application has the following beneficial effects:

[0042] The present application is aimed at the problems of large time consumption and poor reconstruction quality of photon counting spectral CT image reconstruction, and the problems of explosive decomposition noise and serious influence on the quality of material decomposition. Based on subspace decomposition technology, a multi-prior reconstruction method combining non-local low-rank prior of spectral feature map and spatial TV regularization constraint of spectral image is proposed to realize high-precision image reconstruction of spectral CT. Unlike existing spectral CT reconstruction algorithms based on sparse optimization, the present application combines the non-local structural sparsity of the corresponding feature map of spectral CT, selects a suitable reference image block in each iteration, extracts similar image blocks using graph domain distance, constructs a full-energy feature image tensor block group, integrates it into the reconstruction model to update the objective function, and obtains a higher quality reconstruction image by using an iterative strategy based on the alternating direction method. The present application uses the non-local structural sparsity of the subspace feature image to explore the global-non-local correlation between the channels of the spectral image, combines the sparse prior of the gradient image within the channel, designs an efficient optimization strategy, and improves the calculation efficiency and the quality of the reconstruction image. The experimental results verify the effectiveness of the proposed method in spectral CT reconstruction. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1A flowchart of a subspace-assisted multi-prior photon counting spectral CT reconstruction method provided for an embodiment of the present application is shown in the figure.

[0044] Figure 2 A digital simulation walnut body example provided for an embodiment of the present application is shown in the figure.

[0045] Figure 3 A flowchart of a single iteration of a spectral CT reconstruction based on global-nonlocal sparse constraint and spatial TV regularization constraint assisted by subspace is shown in the figure.

[0046] Figure 4 A comparison of local detail reconstruction results of a digital simulation walnut body reconstructed by the method of the present application and other existing methods provided for an embodiment of the present application is shown in the figure.

[0047] Figure 5 A comparison of material decomposition reconstruction results of actual mouse data reconstructed by the method of the present application and other existing methods provided for an embodiment of the present application is shown in the figure.

[0048] Figure 6 A structure diagram of a subspace-assisted multi-prior photon counting spectral CT reconstruction device provided for an embodiment of the present application is shown in the figure.

[0049] Figure 7 A structure block diagram of an electronic device provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0050] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0051] The present application relates to a photon counting spectral CT reconstruction method based on subspace-assisted multi-prior (including global, nonlocal, and local priori). First, the spectral CT image with global low-rank characteristics is mapped to a low-dimensional feature image using subspace decomposition. Then, similar image blocks are extracted from the tensor feature image based on graph domain distance, and stacked to form a nonlocal full-channel tensor group. Next, the nonlocal similarity of the tensor group is described by using the non-convex structure sparsity of adaptive dictionary learning. In addition, the spatial sparsity of the single-channel image is further characterized by using total variation regularization. Finally, an efficient iterative solution algorithm is designed under the alternating direction method framework.

[0052] As Figure 1As shown, this embodiment of the invention provides a subspace-assisted multi-priority photon counting spectral CT reconstruction method, comprising:

[0053] S101: Generate a spectral CT image based on the projection data of the spectral CT imaging, wherein the spectral CT image includes images of S channels;

[0054] Specifically, after acquiring the projection data of the image, the parameter values ​​required for the reconstruction algorithm are selected, and the initial image can be an all-zero or all-one matrix; the projection data, algorithm parameters, and initial image are then processed by the Simultaneous Algebraic Reconstruction Algorithm (SART) to generate an energy spectrum CT image. It consists of images with S channels, where the image of the s-th channel is denoted as x. s (1≤s≤S).

[0055] S102: For the energy spectrum CT image Subspace decomposition yields the noise feature map tensor and orthonormal basis E;

[0056] S103: In the noise feature map tensor Select reference image block p i,j From energy spectrum CT images Image patches similar to the reference image patch are extracted from all channels to generate a nonlocal similarity full-channel tensor set;

[0057] S104: Construct a global sparse regularization term for the nonlocal similar full-channel tensor set using dictionary learning to obtain a denoised nonlocal similar full-channel tensor set, and multiply it with the orthogonal basis to obtain a clean tensor image;

[0058] Specifically, dictionary learning is used to characterize the structural sparsity of nonlocally similar full-channel tensors, and sparse optimization theory is used to construct a global sparse regularization term.

[0059] S105: Based on the TV regularization constraints of each channel of the clean tensor image and the energy spectrum CT image, construct a reconstruction model with dual regularization constraints.

[0060] S106: The augmented Lagrange multiplier method is used to update the reconstruction model with the double regularization constraint until the preset reconstruction quality requirements are met, such as no significant change between the current reconstructed image and the previous reconstructed image. It should be noted that, for the first reconstruction, the previous reconstructed image refers to the energy spectrum CT image generated in step S101.

[0061] Traditional spectral CT reconstruction algorithms are plagued by strong noise artifacts, which severely affect the identification of valuable information. The spectral CT reconstruction method provided in this invention starts from the initial image, characterizes the nonlocal correlation of feature images through subspace decomposition during the reconstruction process, and improves the reconstruction model using a constructed nonlocally similar full-channel tensor set, establishing a new reconstruction model. Furthermore, by employing the augmented Lagrange multiplier method—a continuous iterative and cyclic mechanism—reconstruction accuracy is improved and computational efficiency is optimized.

[0062] Figure 2 This is an example of a digitally simulated walnut shape. Combined with... Figure 3 As shown, based on the above embodiments, the embodiments of the present invention use the following process to generate a nonlocally similar full-channel tensor set: First, in the noise feature map tensor... Select a reference image block p i,j The graph-based block matching (GBM) algorithm is used in energy dispersive CT images. Extraction within the first channel and p i,j There are N similar image patches in total. p The reference image block is then used to form a set of similar image blocks. Next, the position of the reference image block is changed to obtain different sets of similar image blocks. Then, the same process is repeated in different channels, referring to the process for the first channel, until image block matching is completed for all channels. Finally, the different sets of similar image blocks are vectorized, and the vectorized results are stacked in the order of the channels to form a nonlocal similarity full-channel tensor group.

[0063] Based on the above embodiments, the global sparse regularization term constructed in this embodiment of the invention is:

[0064]

[0065] in, Represents the i-th nonlocally similar full-channel tensor set Expanded matrix form, Describing the Frobenius norm, This indicates an adaptive dictionary. Representative group sparsity coefficients, ||·|| p L represents p Norm, 0 <p≤1, Represents the weight matrix. This represents the Hadamard product of two matrices, and τ represents the regularization parameter.

[0066] Based on the above embodiments, the reconstruction model of the dual regularization constraint is as follows:

[0067]

[0068] wherein A represents a system matrix for converting a CT image into projection data; x s represents a spectral CT image of the s th channel, p s represents projection data corresponding to the s th channel, i.e. a CT image under the s th energy, and there are S energies in total.||x s || TV represents a TV norm of the s th channel, Frob enius norm term represents a clean tensor image and a spectral CT image , ρ, λ are regularization parameters, represents a global sparse regularization term about a non-local similar full-channel tensor group is a non-convex low-rank regularization term.

[0069] Further, the model shown in the above formula (2) is updated by using an augmented Lagrange multiplier method, and a specific formula is as follows:

[0070]

[0071] An auxiliary variable a Lagrange multiplier Λ and a non-negative penalty parameter β are introduced, and the above formula (3) is rewritten into an unconstrained optimization model:

[0072]

[0073] The reconstruction model is decomposed into four sub-problems, and the above unconstrained optimization model is minimized by using an alternating direction multiplier method, and an iteration formula is as follows:

[0074]

[0075] wherein represents a spectral CT reconstruction image of this round. Figure 3 A flowchart of a single iteration of spectral CT reconstruction based on global-non-local sparse constraint and spatial TV regularization constraint of a subspace auxiliary provided by the embodiment of the present application.

[0076] To address the issues of high reconstruction time, poor reconstruction quality, and a surge in decomposition noise in photon-counted spectral CT images, which severely impacts the quality of material decomposition, this invention proposes a multi-priority reconstruction method based on subspace decomposition technology. This method combines the nonlocal low-rank prior of the spectral feature map with the spatial TV regularization constraint of the spectral image to achieve high-precision spectral CT image reconstruction. Unlike existing spectral CT reconstruction algorithms based on sparse optimization, this invention leverages the nonlocal structural sparsity of the feature map corresponding to the spectral CT image. In each iteration, a suitable reference image block is selected, and similar image blocks are extracted using graph-domain distance. A group of full-energy feature image tensors is constructed and integrated into the reconstruction model to update the objective function. The updated objective function is then iteratively updated using an alternating direction method to obtain a higher-quality reconstructed image. This invention utilizes the nonlocal structural sparsity of the subspace feature image to explore the global-nonlocal correlation between spectral image channels. Combined with the sparse prior of the gradient image within each channel, an efficient optimization strategy is designed, improving computational efficiency and the quality of the reconstructed image.

[0077] Figure 4 The results show the reconstruction of local details on a digitally simulated walnut body using the method of this invention and other existing methods. Figure 5 This diagram illustrates a comparison of material decomposition and reconstruction results using the proposed method with other existing methods on actual mouse data. Other existing methods mainly include Filtered Back Projection (FBP), Total Variational (TV) method, SBM_L0, FTNN, and ITS_TV. FBP is a traditional CT image analytical reconstruction algorithm; TV is an iterative reconstruction technique that establishes image priors through gradient constraints on each channel image; SBM_L0 is an energy spectrum reconstruction method that integrates subspace (BM3D) denoising technology and in-channel gradient L0 norm constraints; FTNN is an energy spectrum CT reconstruction technique based on tensor frame set norm; and ITS_TV is a multi-prior energy spectrum reconstruction method that combines tensor CANDECOMP / PARAFAC (CP) decomposition with tensor sparse regularization and in-channel total variational (TV) regularization. The relevant parameters involved in each method are shown in Table 1. Experimental results verify the effectiveness of the proposed method in energy spectrum CT reconstruction.

[0078] Table 1. Relevant parameters for each algorithm.

[0079]

[0080] Corresponding to the methods mentioned above, such as Figure 6 As shown, this embodiment of the invention also provides a subspace-assisted multi-priority photon counting spectral CT reconstruction device, including: an image generation module, a subspace decomposition module, a tensor group generation module, a sparsification module, and a reconstruction module.

[0081] The image generation module is configured to generate a spectral CT image including S-channel images according to projection data of spectral CT imaging; the subspace decomposition module is configured to perform subspace decomposition on the spectral CT image to obtain a noise feature map tensor and an orthogonal basis; the tensor group generation module is configured to select a reference image block on the noise feature map tensor, extract image blocks similar to the reference image block from all channels of the spectral CT image to generate a non-local similar full-channel tensor group; the sparsification module is configured to construct a global sparse regularization term about the non-local similar full-channel tensor group by using dictionary learning to obtain a denoised non-local similar full-channel tensor group, and multiply the denoised non-local similar full-channel tensor group by the orthogonal basis to obtain a clean tensor image; and the reconstruction module is configured to construct a reconstruction model with double regularization constraints according to the clean tensor image and TV regularization constraints of each channel of the spectral CT image, and update the reconstruction model with double regularization constraints by using a augmented Lagrange multiplier method until a preset reconstruction quality requirement is reached.

[0082] It should be noted that the device provided by the embodiment of the present application is to realize the above-mentioned method, and the functions can be referred to the above-mentioned method embodiment, which will not be repeated here.

[0083] Figure 7 An example of a schematic diagram of a physical structure of an electronic device is shown in Figure 7 As shown, the electronic device can include a processor 701, a communications interface 702, a memory 703, and a communications bus 704, wherein the processor 701, the communications interface 702, and the memory 703 complete mutual communication through the communications bus 704. The processor 701 can invoke a logical instruction in the memory 703 to execute a subspace-assisted multi-prior photon counting spectral CT reconstruction method, which includes: generating a spectral CT image including S-channel images according to projection data of spectral CT imaging; performing subspace decomposition on the spectral CT image to obtain a noise feature map tensor and an orthogonal basis; selecting a reference image block on the noise feature map tensor, extracting image blocks similar to the reference image block from all channels of the spectral CT image to generate a non-local similar full-channel tensor group; constructing a global sparse regularization term about the non-local similar full-channel tensor group by using dictionary learning to obtain a denoised non-local similar full-channel tensor group, and multiplying the denoised non-local similar full-channel tensor group by the orthogonal basis to obtain a clean tensor image; constructing a reconstruction model with double regularization constraints according to the clean tensor image and TV regularization constraints of each channel of the spectral CT image; and updating the reconstruction model with double regularization constraints by using an augmented Lagrange multiplier method until a preset reconstruction quality requirement is reached.

[0084] In addition, the logic instructions in the memory 703 described above are implemented in the form of software function units and sold or used as independent products, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0085] The embodiments of the present application also provide a computer program product, which comprises a computer program stored on a non-transitory computer readable storage medium, and the computer program comprises program instructions, and when the program instructions are executed by a computer, the computer can execute the method provided by the above-mentioned method embodiments, for example, comprising: generating a spectral CT image according to projection data of spectral CT imaging, the spectral CT image comprising S-channel images; performing subspace decomposition on the spectral CT image to obtain a noise feature map tensor and an orthogonal basis; selecting a reference image block on the noise feature map tensor, extracting image blocks similar to the reference image block from all channels of the spectral CT image to generate a non-local similar full-channel tensor group; constructing a global sparse regularization term about the non-local similar full-channel tensor group by using dictionary learning to obtain a denoised non-local similar full-channel tensor group, and multiplying the denoised non-local similar full-channel tensor group by the orthogonal basis to obtain a clean tensor image; constructing a double regularization constraint reconstruction model according to the clean tensor image and the TV regularization constraint of each channel of the spectral CT image; and updating the double regularization constraint reconstruction model by using the augmented Lagrange multiplier method until a preset reconstruction quality requirement is reached.

[0086] The embodiment of the present application also provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method provided by each method embodiment, for example, comprising: generating a spectral CT image according to projection data of spectral CT imaging, the spectral CT image comprising S-channel images; performing subspace decomposition on the spectral CT image to obtain a noise feature map tensor and an orthogonal basis; selecting a reference image block on the noise feature map tensor, extracting image blocks similar to the reference image block from all channels of the spectral CT image to generate a non-local similar full-channel tensor group; constructing a global sparse regularization term about the non-local similar full-channel tensor group by using dictionary learning to obtain a denoised non-local similar full-channel tensor group, and multiplying the denoised non-local similar full-channel tensor group by the orthogonal basis to obtain a clean tensor image; constructing a reconstruction model with double regularization constraints according to the clean tensor image and the TV regularization constraint of each channel of the spectral CT image; and updating the reconstruction model with double regularization constraints by using the augmented Lagrange multiplier method until a preset reconstruction quality requirement is reached.

[0087] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platforms, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in the form of a software product, can be embodied in a computer software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0088] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features thereof; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A subspace-assisted multi-priority photon counting spectral CT reconstruction method, characterized in that, include: Step 1: Generate a spectral CT image based on the projection data of the spectral CT imaging, wherein the spectral CT image includes images of S channels; Step 2: Perform subspace decomposition on the energy spectrum CT image to obtain the noise feature map tensor and orthogonal basis; Step 3: Select a reference image block on the noise feature map tensor, and extract image blocks similar to the reference image block from all channels of the energy spectrum CT image to generate a non-local similarity full-channel tensor group; Step 4: Construct a global sparse regularization term for the nonlocal similar full-channel tensor set using dictionary learning to obtain the denoised nonlocal similar full-channel tensor set, and multiply it with the orthogonal basis to obtain a clean tensor image; Step 5: Based on the TV regularization constraints of each channel of the clean tensor image and the energy spectrum CT image, construct a reconstruction model with double regularization constraints; Step 6: Update the reconstruction model with the double regularization constraint using the augmented Lagrange multiplier method until the preset reconstruction quality requirements are met.

2. The subspace-assisted multi-priority photon counting spectral CT reconstruction method according to claim 1, characterized in that, In step 1, the simultaneous algebraic reconstruction algorithm SART is used to generate energy spectrum CT images.

3. The subspace-assisted multi-priority photon counting spectral CT reconstruction method according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Select a reference image block on the noise feature map tensor, and use a block matching algorithm based on map distance to extract image blocks similar to the reference image block from one of the channels of the energy spectrum CT image, and form a set of similar image blocks from all the extracted image blocks; Step 3.2: Change the position of the reference image patch and repeat step 3.1 to obtain multiple different sets of similar image patches; Step 3.3: Perform steps 3.1 and 3.2 for all channels of the energy dispersive CT image; Step 3.4: Vectorize all similar image patch sets, and then stack the vectorized results in channel order to obtain the non-local similarity full-channel tensor group.

4. The subspace-assisted multi-priority photon counting spectral CT reconstruction method according to claim 1, characterized in that, The global sparse regularization term is: in, Represents the i-th nonlocally similar full-channel tensor set Expanded matrix form, Denotes the Frobenius norm. This indicates an adaptive dictionary. Representative group sparsity coefficients, ||·|| p L represents p Norm, 0 <p≤1, Represents the weight matrix. This represents the Hadamard product of two matrices, and τ represents the regularization parameter.

5. The subspace-assisted multi-priority photon counting spectral CT reconstruction method according to claim 1, characterized in that, The reconstruction model of the double regularization constraint is as follows: Where A represents the system matrix, x s Representing spectral CT images The s-th channel, p s This represents the projection data corresponding to the s-th channel, ||x s || TV Let the TV norm of the s-th channel be denoted by the Frobenius norm term. Representing a clean tensor image With spectral CT images The difference between them, where ρ and λ are regularization parameters. Represents a set of nonlocally similar full-channel tensors The global sparse regularization term.

6. The subspace-assisted multi-priority photon counting spectral CT reconstruction method according to claim 1, characterized in that, Step 6 specifically includes: Introducing auxiliary variables The reconstruction model with the double regularization constraint is rewritten using the Lagrange multiplier Λ and the nonnegative penalty parameter β as follows: The modified model described above is solved using the alternating direction multiplier method; the iterative formula is as follows: in, This represents the reconstructed spectral CT image from this round.

7. A subspace-assisted multi-priority photon counting spectral CT reconstruction device, characterized in that, include: The image generation module is used to generate a spectral CT image based on the projection data of the spectral CT imaging, wherein the spectral CT image includes an image with S channels; The subspace decomposition module is used to perform subspace decomposition on the energy spectrum CT image to obtain noise feature map tensors and orthogonal bases; Tensor group generation module is used to select a reference image block on the noise feature map tensor and extract image blocks similar to the reference image block from all channels of the energy spectrum CT image to generate a non-local similar full-channel tensor group. The sparsification module is used to construct a global sparse regularization term for the nonlocally similar full-channel tensor set using dictionary learning, to obtain a denoised nonlocally similar full-channel tensor set, which is then multiplied by the orthogonal basis to obtain a clean tensor image. The reconstruction module is used to construct a reconstruction model with double regularization constraints based on the TV regularization constraints of each channel of the clean tensor image and the energy spectrum CT image, and to update the reconstruction model with double regularization constraints using the augmented Lagrange multiplier method until the preset reconstruction quality requirements are met.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images