Synchrotron radiation microscopic CT reconstruction method and device based on decomposition diffusion sampling

Abstract

The invention discloses a synchrotron radiation micro-CT reconstruction method and device based on decomposition diffusion sampling, and the method comprises the steps: collecting synchrotron radiation micro-CT images which are reconstructed at a full projection angle and do not contain ring artifacts, and taking the synchrotron radiation micro-CT images as a training data set; applying stochastic differential equation disturbance at a set moment to the training image, and inputting the training image and the disturbed training image into a score matching network model for training to obtain a trained score matching network model; performing image sampling through the trained score matching network model; solving the sampled image through an augmented Lagrange multiplier algorithm to obtain a synchrotron radiation microscopic CT image; performing de-noising diffusion processing on the sample through a de-noising diffusion implicit model to obtain an initial value of the next round of sampling; carrying out iterative sampling and solving until a set iteration round is completed, and outputting a last round of synchrotron radiation microscopic CT image; and carrying out de-noising processing through a Tweedie equation to obtain a final synchrotron radiation microscopic CT image. According to the invention, the accuracy of the reconstructed image is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of radiological medical image processing, and in particular to a synchrotron radiation micro-CT reconstruction method and device based on decomposition diffusion sampling. Background Art

[0002] Synchrotron radiation micro-CT may introduce ring artifacts into the reconstructed image due to its high detector resolution and sensitivity. At the same time, due to its long scanning time and high radiation dose, sparse angular sampling is often required, which may introduce sparse angular artifacts and noise into the reconstructed image.

[0003] At present, traditional methods solve the problem of ring artifacts and the problem of sparse angle reconstruction separately. For the problem of ring artifacts, the ring artifacts that appear as vertical stripes are removed by using filtering methods such as wavelet Fourier filtering or model-based iterative methods such as low-rank tensor decomposition methods in the projection domain. However, the stripe removal ability of such methods is limited, and the residual stripes will cause great interference to the reconstructed image. For the problem of sparse angle reconstruction, traditional methods use model-based iterative methods such as SART-TV and end-to-end deep learning methods to solve it. The iterative methods of the model rely on the preset model prior and have limited artifact removal ability. Deep learning algorithms rely on a large amount of paired annotated data, and it takes a lot of time and effort to obtain these data in practice. Generative diffusion model-based methods such as decomposed diffusion sampling methods can achieve high-quality sparse angle CT reconstruction without paired data, but they do not consider the impact of ring artifacts. The traditional methods do not remove the ring artifacts cleanly, which will cause problems such as pseudo structures in the reconstructed image. Currently, the algorithms for simultaneously removing sparse corner artifacts and ring artifacts are deep learning algorithms for image post-processing, which require paired data, which is difficult to obtain in practice.

[0004] Therefore, in view of the shortcomings of the existing technology, it is necessary to provide a sparse angle synchrotron radiation micro-CT reconstruction method based on decomposed diffusion sampling to solve the shortcomings of the existing technology in X-ray synchrotron radiation micro-CT image reconstruction. Summary of the invention

[0005] To this end, the present invention provides a synchrotron radiation micro-CT reconstruction method and device based on decomposition diffusion sampling. Through data prior and model prior, the CT image is updated on a clean CT image distribution manifold during the decomposition diffusion sampling constraint reconstruction process. The reconstructed image can effectively remove ring artifacts and sparse angular artifacts, and the model prior of the low rank of the tensor decomposes the vertical stripes and the projection image, thereby ensuring the accuracy of the reconstructed image.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, comprising:

[0007] Collecting synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and using the collected synchrotron radiation micro-CT images as a training data set;

[0008] Obtaining a disturbed training image by applying a stochastic differential equation perturbation at a set time to the training image in the training data set;

[0009] Inputting the training images in the training data set and the perturbed training images into a score matching network model; the score matching network model is trained according to the training images in the training data set and the perturbed training images, learning the corresponding scores under the perturbation of the stochastic differential equation at a set time, and obtaining the trained score matching network model;

[0010] Perform image sampling through the trained score matching network model; calculate the sampled image through the Tweedie equation to obtain a manifold distribution image of the training image that satisfies the disturbance;

[0011] The sampled image is solved by the augmented Lagrange multiplier algorithm to obtain the synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information.

[0012] The obtained synchrotron radiation micro-CT image is subjected to denoising diffusion processing by a denoising diffusion implicit model to obtain a starting value for the next round of sampling; according to the starting value for the next round of sampling, iterative sampling and solving are performed until the set iteration rounds are completed, and the last round of synchrotron radiation micro-CT image is output;

[0013] The final round of synchrotron radiation micro-CT images are denoised using the Tweedie equation to obtain a final synchrotron radiation micro-CT image.

[0014] As a preferred solution of a synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, in the process of obtaining the perturbed training image by applying the stochastic differential equation perturbation at the set time to the training image in the training data set, the stochastic differential equation expression at the set time is:

[0015]

[0016] Where x0 is the training image; s t is the drift function; σ t is the diffusion function; z is the Gaussian distribution.

[0017] As a preferred solution of a synchrotron radiation micro-CT reconstruction method based on decomposed diffusion sampling, the loss function expression of the fractional matching network model is:

[0018]

[0019] In the formula, For all possible samples,,x t ..,,x0.’s expectation is minimized; is the score matching network model; It means that under the given x0 condition, about x t The gradient of the log-conditional probability density .

[0020] As a preferred solution of a synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, in the process of the score matching network model learning the corresponding score under the perturbation of the stochastic differential equation at the set time, the calculation formula of the corresponding score under the perturbation of the stochastic differential equation at the set time is:

[0021]

[0022] In the formula, s θ is the score matching network model; For x t The corresponding score.

[0023] As a preferred solution of a synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, in the process of calculating the sampled image through the Tweedie equation to obtain the training image manifold distribution image satisfying the perturbation, the training image manifold distribution image satisfying the perturbation is expressed as:

[0024]

[0025] In the formula, To satisfy the manifold distribution image of the training image after the disturbance.

[0026] As a preferred solution of the synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, in the process of solving the sampled image by the augmented Lagrangian multiplier algorithm to obtain the synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information, the expression solved by the conjugate gradient method according to the augmented Lagrangian multiplier is:

[0027]

[0028] Where A is the CT projection matrix; A T is the transposed projection operator; ρ is the regularization parameter; D z is the total variation operator in the Z direction; D z The transpose operator; y is the projection information; S t+1 is the stripe item of the previous iteration; R1t+1 is an auxiliary variable; w1 t+1 is the Lagrange multiplier.

[0029] As a preferred solution of the synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, in the process of denoising the last round of synchrotron radiation micro-CT image through the Tweedie equation to obtain the final synchrotron radiation micro-CT image, the expression of the final synchrotron radiation micro-CT image is:

[0030]

[0031] Where x0 is the final synchrotron radiation micro-CT image.

[0032] The present invention also provides a synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, which adopts the above-mentioned synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, including:

[0033] A training data set acquisition module, used to collect synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and use the collected synchrotron radiation micro-CT images as training data sets;

[0034] A training image perturbation processing module, used for obtaining a perturbed training image by applying a stochastic differential equation perturbation at a set time to the training image in the training data set;

[0035] A score matching network model training module is used to input the training images in the training data set and the perturbed training images into the score matching network model; the score matching network model is trained according to the training images in the training data set and the perturbed training images, learns the corresponding scores under the perturbation of the stochastic differential equation at a set time, and obtains the trained score matching network model;

[0036] An image sampling module is used to perform image sampling through the trained score matching network model; the sampled image is calculated through the Tweedie equation to obtain a training image manifold distribution image that satisfies the disturbance;

[0037] A sampling image solving module is used to solve the sampled image by using an augmented Lagrange multiplier algorithm to obtain a synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information;

[0038] An iterative sampling and solving processing module is used to perform denoising and diffusion processing on the obtained synchrotron radiation micro-CT image through a denoising and diffusion implicit model to obtain a starting value for the next round of sampling; iterative sampling and solving are performed according to the starting value for the next round of sampling until the set iteration rounds are completed, and the last round of synchrotron radiation micro-CT images are output;

[0039] The final synchrotron radiation micro-CT image acquisition module is used to perform denoising on the final round of synchrotron radiation micro-CT images through the Tweedie equation to obtain the final synchrotron radiation micro-CT images.

[0040] As a preferred solution of the synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, in the training image perturbation processing module, in the process of obtaining the perturbed training image by applying the stochastic differential equation perturbation at the set time to the training image in the training data set, the stochastic differential equation expression at the set time is:

[0041]

[0042] Where x0 is the training image; s t is the drift function; σ t is the diffusion function; z is the Gaussian distribution.

[0043] As a preferred solution of the synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, in the score matching network model training module, the loss function expression of the score matching network model is:

[0044]

[0045] In the formula, For all possible samples,,x t ..,,x0.’s expectation is minimized; is the score matching network model; It means that under the given x0 condition, about x t The gradient of the log-conditional probability density .

[0046] As a preferred solution of the synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, in the score matching network model training module, in the process of the score matching network model learning the corresponding score under the random differential equation perturbation at the set time, the calculation formula of the corresponding score under the random differential equation perturbation at the set time is:

[0047]

[0048] In the formula, s θ is the score matching network model; For x t The corresponding score.

[0049] As a preferred solution of the synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, in the image sampling module, in the process of calculating the sampled image through the Tweedie equation to obtain the training image manifold distribution image satisfying the disturbance, the training image manifold distribution image satisfying the disturbance is expressed as:

[0050]

[0051] In the formula, To satisfy the manifold distribution image of the training image after the disturbance.

[0052] As a preferred solution of the synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, in the sampled image solving module, in the process of solving the sampled image by the augmented Lagrangian multiplier algorithm to obtain the synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information, the expression solved by the conjugate gradient method according to the augmented Lagrangian multiplier is:

[0053]

[0054]

[0055] Where A is the CT projection matrix; A T is the transposed projection operator; ρ is the regularization parameter; D z is the total variation operator in the Z direction; D z The transpose operator; y is the projection information; S t+1 is the stripe item of the previous iteration; R1 t+1 is an auxiliary variable; w1 t+1 is the Lagrange multiplier.

[0056] As a preferred solution of the synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, in the final synchrotron radiation micro-CT image acquisition module, in the process of denoising the last round of synchrotron radiation micro-CT image through the Tweedie equation to obtain the final synchrotron radiation micro-CT image, the expression of the final synchrotron radiation micro-CT image is:

[0057]

[0058] Where x0 is the final synchrotron radiation micro-CT image.

[0059] The present invention has the following advantages: collecting synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and using the collected synchrotron radiation micro-CT images as training data sets; obtaining perturbed training images by applying random differential equation perturbations at set times to training images in the training data sets; inputting the training images in the training data sets and the perturbed training images into a score matching network model; the score matching network model is trained according to the training images in the training data sets and the perturbed training images, learning the scores corresponding to the perturbations at set times, and obtaining the trained score matching network model; and performing image matching using the trained score matching network model. Image sampling; the sampled image is calculated by the Tweedie equation to obtain a training image manifold distribution image that satisfies the disturbance; the sampled image is solved by the augmented Lagrange multiplier algorithm to obtain a synchrotron radiation micro-CT image constrained by the model and the projection information measured by the detector; the obtained synchrotron radiation micro-CT image is subjected to denoising and diffusion processing by the denoising diffusion implicit model to obtain the starting value of the next round of sampling; according to the starting value of the next round of sampling, iterative sampling and solving are performed until the set iteration rounds are completed, and the final round of synchrotron radiation micro-CT images are output; the final round of synchrotron radiation micro-CT images are denoised by the Tweedie equation to obtain the final synchrotron radiation micro-CT image. Compared with the current image post-processing deep learning algorithm that simultaneously removes sparse angular artifacts and ring artifacts, the system training proposed by the present invention is realized by learning the probability distribution of high-quality images, and the training process does not require low-quality data paired with it. In addition, the reconstruction result of this method is constrained by the model prior and the original projection information, and is more accurate than the reconstruction result of the post-processing method, and is not prone to pseudo-structures. Compared with the existing ring artifact removal algorithms, such as the model-based iterative algorithms, the system proposed in the present invention utilizes both data priors and model priors. Based on the model prior constraints, the update of the CT image in the reconstruction process is performed on the clean CT image distribution manifold by decomposing the diffusion sampling constraints, which is more excellent than the traditional algorithm in terms of ring artifact removal ability. Compared with the current sparse angle CT reconstruction algorithm, for the model-based iterative algorithm, the present invention utilizes both data priors and model priors, and the reconstruction quality is higher. For the end-to-end deep learning method, the training of the present invention is achieved by learning the probability distribution of high-quality images, without the need for paired data, and the reconstruction quality is better because the model priors and the constraints of the original projection information are taken into account. For the diffusion generation model method, the algorithm proposed in the present invention considers the ring artifact term in the objective function, combines the ring artifact removal and sparse angle reconstruction into one problem, and the network simultaneously constrains the projection and reconstructed image to prevent the residual ring artifacts of the traditional ring artifact removal algorithm from further affecting the reconstruction process, and avoid problems such as image pseudo-structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the drawings required for the implementation methods or the prior art descriptions are briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0061] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0062] Figure 1 A schematic diagram of the process flow of the synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling provided in Example 1 of the present invention;

[0063] Figure 2 A schematic diagram of a specific implementation process framework of the synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling provided in Example 1 of the present invention;

[0064] Figure 3 A schematic diagram of some results of the present invention and the prior art in the synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling provided in Example 1 of the present invention;

[0065] Figure 4 Schematic diagram of the architecture of a synchrotron radiation micro-CT reconstruction device based on decomposed diffusion sampling provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0066] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Example 1

[0068] See also Figure 1 and Figure 2 Embodiment 1 of the present invention provides a synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling, comprising:

[0069] S1. Collect synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and use the collected synchrotron radiation micro-CT images as a training data set;

[0070] S2, obtaining a perturbed training image by applying a stochastic differential equation perturbation at a set time to the training image in the training data set;

[0071] S3, inputting the training images in the training data set and the perturbed training images into the score matching network model; the score matching network model is trained according to the training images in the training data set and the perturbed training images, learning the corresponding scores under the perturbation of the stochastic differential equation at a set time, and obtaining the trained score matching network model;

[0072] S4, performing image sampling through the trained score matching network model; calculating the sampled image through the Tweedie equation to obtain a manifold distribution image of the training image after the disturbance;

[0073] S5, solving the sampled image by using an augmented Lagrange multiplier algorithm to obtain a synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information;

[0074] S6, performing denoising and diffusion processing on the obtained synchrotron radiation micro-CT image through a denoising and diffusion implicit model to obtain a starting value for the next round of sampling; performing iterative sampling and solving according to the starting value for the next round of sampling until the set iteration rounds are completed, and outputting the last round of synchrotron radiation micro-CT image;

[0075] S7. De-noising the last round of synchrotron radiation micro-CT images using the Tweedie equation to obtain a final synchrotron radiation micro-CT image.

[0076] In this embodiment, in step S1, synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts are collected, and the collected synchrotron radiation micro-CT images are used as a training data set.

[0077] In this embodiment, in step S2, a perturbed training image is obtained by applying a stochastic differential equation perturbation at a set time to the training image in the training data set;

[0078] Specifically, stochastic differential equation (SDE) perturbations are applied to the training images at different times, and the perturbations and the original training images are input into the convolutional neural network.

[0079] Among them, the stochastic differential equation expression at the set time is:

[0080]

[0081] Where x0 is the training image; s t is the drift function; σ t is the diffusion function; z is the Gaussian distribution.

[0082] In this embodiment, in step S3, the training images in the training data set and the perturbed training images are input into the score matching network model; the score matching network model is trained according to the training images in the training data set and the perturbed training images, and the corresponding scores under the perturbation of the stochastic differential equation at a set time are learned to obtain the trained score matching network model;

[0083] Specifically, the training image in the training data set and the perturbed training image are input into a score matching network model, and the score matching network model uses the input image to obtain a score matching network s θ Learn the corresponding scores under SDE perturbations at different times.

[0084] Among them, the loss function expression of the score matching network model is:

[0085]

[0086] In the formula, For all possible samples,,x t ..,,x0.’s expectation is minimized; is the score matching network model; It means that under the given x0 condition, about x t The gradient of the log-conditional probability density .

[0087] Among them, the calculation formula of the corresponding score under the perturbation of the stochastic differential equation at the set time is:

[0088]

[0089] In the formula, s θ is the score matching network model; For x t The corresponding score.

[0090] In this embodiment, in step S4, image sampling is performed using the trained score matching network model; the sampled image is calculated using the Tweedie equation to obtain a training image manifold distribution image that satisfies the disturbance;

[0091] Specifically, the reconstruction process first initializes various variables, including the reconstruction starting image Projection matrix A, iteration step N, stripes S N , auxiliary variable R1 in the ALM solution processN , Q N and the Lagrange multiplier W1 N , W2 N .

[0092] The trained score matching network model is used to perform the sampling process of the diffusion model. In each sampling process, the Tweedie equation is first used to calculate a clean image that satisfies the manifold distribution of the training image.

[0093] Among them, the image expression of the training image manifold distribution after satisfying the disturbance is:

[0094]

[0095] In the formula, To satisfy the manifold distribution image of the training image after the disturbance.

[0096] In this embodiment, in step S5, the sampled image is solved by an augmented Lagrange multiplier algorithm to obtain a synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information;

[0097] Specifically, using the CT projection matrix A, and the total variation (TV) operator D in the Z direction z Calculate the projection matrix A used to solve the CT image subproblem using the conjugate gradient method CG , the calculation formula is as follows:

[0098]

[0099] In the formula, A T is the transposed projection operator; ρ is the regularization parameter; D z The transpose operator of .

[0100] Using the projection information y and the stripe item S of the previous iteration t+1 and auxiliary variable R1 t+1 and the Lagrange multiplier w1 t+1 Calculate the b required for the conjugate gradient method to solve the CT image subproblem CG , the calculation formula is as follows:

[0101]

[0102] According to A CG and b CG , the expression for solving the CT image subproblem using the conjugate gradient method is:

[0103]

[0104] Where M represents the number of conjugate gradient method iterations.

[0105] Update auxiliary variable R1 using soft threshold method t+1 The solution expression of the subproblem is:

[0106]

[0107] In the formula, is a soft threshold operator.

[0108] Using projection information y and auxiliary variables Q t+1 , the stripes S in the projection of the tensor low rank constraint are obtained by Tucker decomposition t , and the solution process is:

[0109]

[0110] In the formula, λ t is the regularization parameter.

[0111] In this embodiment, in step S6, the obtained synchrotron radiation micro-CT image is subjected to denoising diffusion processing by using a denoising diffusion implicit model to obtain a starting value for the next round of sampling; according to the starting value for the next round of sampling, iterative sampling and solving are performed until the set iteration rounds are completed, and the last round of synchrotron radiation micro-CT images are output;

[0112] Specifically, update the constraint stripe S t , the size of the auxiliary variable Q t , the solution process is:

[0113]

[0114] In the formula, Represents the tensor Q t The expansion index values ​​in the second and third dimensions; is the regularization parameter.

[0115] Update the Lagrange multiplier w1 t , w2 t , whose expression is:

[0116]

[0117] w2 t ←w2 t+1 +β(Q t -S t )

[0118] In the formula, β is the regularization parameter.

[0119] Get the starting x of the next iteration through DDIM sampling t-1 ,

[0120]

[0121] Where η is the noise level of the control sampling; ∈ is the noise coefficient.

[0122] In this embodiment, in step S7, the final synchrotron radiation micro-CT image is denoised by the Tweedie equation to obtain a final synchrotron radiation micro-CT image. The result is as follows: Figure 3 shown.

[0123] Specifically, the final expression of the synchrotron radiation micro-CT image is:

[0124]

[0125] Where x0 is the final synchrotron radiation micro-CT image.

[0126] In summary, the present invention collects synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and uses the collected synchrotron radiation micro-CT images as a training data set; obtains a perturbed training image by applying a random differential equation perturbation at a set time to the training image in the training data set; inputs the training image in the training data set and the perturbed training image into a score matching network model; the score matching network model is trained according to the training image in the training data set and the perturbed training image, learns the corresponding score under the perturbation of the random differential equation at the set time, and obtains the trained score matching network model; and performs image matching through the trained score matching network model. Sampling; calculating the sampled image through the Tweedie equation to obtain a training image manifold distribution image that satisfies the disturbance; solving the sampled image through the augmented Lagrange multiplier algorithm to obtain a synchrotron radiation micro-CT image constrained by the model and the projection information measured by the detector; denoising and diffusing the obtained synchrotron radiation micro-CT image through the denoising diffusion implicit model to obtain the starting value of the next round of sampling; iterative sampling and solving according to the starting value of the next round of sampling until the set iteration rounds are completed, and the final round of synchrotron radiation micro-CT images are output; denoising the final round of synchrotron radiation micro-CT images through the Tweedie equation to obtain the final synchrotron radiation micro-CT image. Compared with the current image post-processing deep learning algorithm that simultaneously removes sparse angular artifacts and ring artifacts, the system training proposed by the present invention is realized by learning the probability distribution of high-quality images, and the training process does not require low-quality data paired with it. And the reconstruction result of this method is constrained by the model prior and the original projection information, and the reconstruction result is more accurate than the post-processing method, and it is not easy to produce pseudo-structures and the like. Compared with the existing ring artifact removal algorithms, such as the model-based iterative algorithms, the system proposed in the present invention utilizes both data priors and model priors. Based on the model prior constraints, the update of the CT image in the reconstruction process is performed on the clean CT image distribution manifold by decomposing the diffusion sampling constraints, which is more excellent than the traditional algorithm in terms of ring artifact removal ability. Compared with the current sparse angle CT reconstruction algorithm, for the model-based iterative algorithm, the present invention utilizes both data priors and model priors, and the reconstruction quality is higher. For the end-to-end deep learning method, the training of the present invention is achieved by learning the probability distribution of high-quality images, without the need for paired data, and the reconstruction quality is better because the model priors and the constraints of the original projection information are taken into account. For the diffusion generation model method, the algorithm proposed in the present invention considers the ring artifact term in the objective function, combines the ring artifact removal and sparse angle reconstruction into one problem, and the network simultaneously constrains the projection and reconstructed image to prevent the residual ring artifacts of the traditional ring artifact removal algorithm from further affecting the reconstruction process, and avoid problems such as image pseudo-structure.

[0127] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0128] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0129] Example 2

[0130] See also Figure 4 Embodiment 2 of the present invention provides a synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, comprising:

[0131] A training data set acquisition module 001 is used to collect synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and use the collected synchrotron radiation micro-CT images as training data sets;

[0132] The training image perturbation processing module 002 is used to obtain a perturbed training image by applying a stochastic differential equation perturbation at a set time to the training image in the training data set;

[0133] The score matching network model training module 003 is used to input the training images in the training data set and the perturbed training images into the score matching network model; the score matching network model is trained according to the training images in the training data set and the perturbed training images, learns the corresponding scores under the perturbation of the stochastic differential equation at a set time, and obtains the trained score matching network model;

[0134] The image sampling module 004 is used to perform image sampling through the trained score matching network model; the sampled image is calculated through the Tweedie equation to obtain a training image manifold distribution image that satisfies the disturbance;

[0135] The sampling image solving module 005 is used to solve the sampled image by using an augmented Lagrange multiplier algorithm to obtain a synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information;

[0136] Iterative sampling and solving processing module 006 is used to perform denoising and diffusion processing on the obtained synchrotron radiation micro-CT image through a denoising and diffusion implicit model to obtain a starting value for the next round of sampling; iterative sampling and solving are performed according to the starting value for the next round of sampling until the set iteration rounds are completed, and the last round of synchrotron radiation micro-CT images are output;

[0137] The final synchrotron radiation micro-CT image acquisition module 007 is used to perform denoising on the final round of synchrotron radiation micro-CT images through the Tweedie equation to obtain the final synchrotron radiation micro-CT images.

[0138] In this embodiment, in the training image perturbation processing module 002, in the process of obtaining the perturbed training image by applying the stochastic differential equation perturbation at the set time to the training image in the training data set, the stochastic differential equation expression at the set time is:

[0139]

[0140] Where x0 is the training image; s t is the drift function; σ t is the diffusion function; z is the Gaussian distribution.

[0141] In this embodiment, in the score matching network model training module 003, the loss function expression of the score matching network model is:

[0142]

[0143] In the formula, For all possible samples,,x t ..,,x0.’s expectation is minimized; is the score matching network model; It means that under the given x0 condition, about x t The gradient of the logarithmic conditional probability density. In this embodiment, in the score matching network model training module 003, in the process of the score matching network model learning the corresponding score under the random differential equation perturbation at the set time, the calculation formula of the corresponding score under the random differential equation perturbation at the set time is:

[0144]

[0145] In the formula, s θ is the score matching network model; For xt The corresponding score.

[0146] In this embodiment, in the image sampling module 004, in the process of calculating the sampled image through the Tweedie equation to obtain the training image manifold distribution image satisfying the disturbance, the training image manifold distribution image expression satisfying the disturbance is:

[0147]

[0148] In the formula, To satisfy the manifold distribution image of the training image after the disturbance.

[0149] In this embodiment, in the sampled image solving module 005, in the process of solving the sampled image by the augmented Lagrangian multiplier algorithm to obtain the synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information, the expression solved by the conjugate gradient method according to the augmented Lagrangian multiplier is:

[0150]

[0151] Where A is the CT projection matrix; A T is the transposed projection operator; ρ is the regularization parameter; D z is the total variation operator in the Z direction; D z The transpose operator; y is the projection information; S t+1 is the stripe item of the previous iteration; R1 t+1 is an auxiliary variable; w1 t+1 is the Lagrange multiplier.

[0152] In this embodiment, in the final synchrotron radiation micro-CT image acquisition module 007, in the process of performing denoising processing on the last round of synchrotron radiation micro-CT image through the Tweedie equation to obtain the final synchrotron radiation micro-CT image, the expression of the final synchrotron radiation micro-CT image is:

[0153]

[0154] Where x0 is the final synchrotron radiation micro-CT image.

[0155] It should be noted that the information interaction, execution process, etc. between the modules / units of the above-mentioned system are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown in the previous part of the present application, and will not be repeated here.

[0156] Example 3

[0157] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which a program code of a synchrotron radiation micro-CT reconstruction method based on decomposed diffusion sampling is stored, and the program code includes instructions for executing the synchrotron radiation micro-CT reconstruction method based on decomposed diffusion sampling of Embodiment 1 or any possible implementation thereof.

[0158] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0159] Example 4

[0160] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0161] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling of Example 1 or any possible implementation thereof.

[0162] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor implemented by reading software codes stored in a memory. The memory can be integrated in the processor or can be located outside the processor and exist independently.

[0163] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0164] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0165] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A synchrotron radiation micro-CT reconstruction method based on decomposed diffusion sampling, characterized in that: include: Collecting synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and using the collected synchrotron radiation micro-CT images as a training data set; Obtaining a disturbed training image by applying a stochastic differential equation perturbation at a set time to the training image in the training data set; Inputting the training images in the training data set and the perturbed training images into a score matching network model; the score matching network model is trained according to the training images in the training data set and the perturbed training images, learning the corresponding scores under the perturbation of the stochastic differential equation at a set time, and obtaining the trained score matching network model; Perform image sampling through the trained score matching network model; calculate the sampled image through the Tweedie equation to obtain a manifold distribution image of the training image that satisfies the disturbance; The sampled image is solved by the augmented Lagrange multiplier algorithm to obtain the synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information. Performing denoising and diffusion processing on the obtained synchrotron radiation micro-CT image through a denoising and diffusion implicit model to obtain a starting value for the next round of sampling; According to the starting value of the next round of sampling, iterative sampling and solving are performed until the set iteration rounds are completed, and the final round of synchrotron radiation micro-CT images are output; The final round of synchrotron radiation micro-CT images are denoised using the Tweedie equation to obtain a final synchrotron radiation micro-CT image.

2. The synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling according to claim 1, characterized in that: In the process of obtaining the perturbed training image by applying the stochastic differential equation perturbation at the set time to the training image in the training data set, the stochastic differential equation expression at the set time is: Where x0 is the training image; s t is the drift function; σ t is the diffusion function; z is the Gaussian distribution.

3. The synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling according to claim 2 is characterized in that: The loss function expression of the score matching network model is: In the formula, For all possible samples,,x t .., the expectation of x0. is minimized; is the score matching network model; It means that under the given x0 condition, about x t The gradient of the log-conditional probability density .

4. The synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling according to claim 3 is characterized in that: In the process of the score matching network model learning the score corresponding to the perturbation of the stochastic differential equation at the set time, the calculation formula of the score corresponding to the perturbation of the stochastic differential equation at the set time is: In the formula, s θ is the score matching network model; For x t The corresponding score.

5. The synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling according to claim 4 is characterized in that: In the process of calculating the sampled image through the Tweedie equation to obtain the training image manifold distribution image satisfying the disturbance, the training image manifold distribution image expression satisfying the disturbance is: In the formula, To satisfy the manifold distribution image of the training image after the disturbance.

6. The synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling according to claim 5, characterized in that: In the process of solving the sampled image by the augmented Lagrangian multiplier algorithm to obtain the synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information, the expression solved by the conjugate gradient method according to the augmented Lagrangian multiplier is: Where A is the CT projection matrix; A T is the CT transpose projection operator; ρ represents the regularization parameter D z is the total variation operator in the Z direction; D z The transpose operator; y is the projection information; S t+1 is the stripe item of the previous iteration; R1 t+1 is an auxiliary variable; w1 t+1 is the Lagrange multiplier.

7. The synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling according to claim 6, characterized in that: In the process of performing denoising on the last round of synchrotron radiation micro-CT images through the Tweedie equation to obtain the final synchrotron radiation micro-CT images, the expression of the final synchrotron radiation micro-CT images is: Where x0 is the final synchrotron radiation micro-CT image.

8. A synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling, using the synchrotron radiation micro-CT reconstruction method based on decomposition diffusion sampling according to any one of claims 1 to 7, characterized in that: include: A training data set acquisition module, used to collect synchrotron radiation micro-CT images reconstructed at all projection angles and free of ring artifacts, and use the collected synchrotron radiation micro-CT images as training data sets; A training image perturbation processing module, used for obtaining a perturbed training image by applying a stochastic differential equation perturbation at a set time to the training image in the training data set; A score matching network model training module is used to input the training images in the training data set and the perturbed training images into the score matching network model; the score matching network model is trained according to the training images in the training data set and the perturbed training images, learns the corresponding scores under the perturbation of the stochastic differential equation at a set time, and obtains the trained score matching network model; An image sampling module is used to perform image sampling through the trained score matching network model; the sampled image is calculated through the Tweedie equation to obtain a training image manifold distribution image that satisfies the disturbance; A sampling image solving module is used to solve the sampled image by using an augmented Lagrange multiplier algorithm to obtain a synchrotron radiation micro-CT image constrained by the model and the detector measurement projection information; An iterative sampling and solving processing module is used to perform denoising and diffusion processing on the obtained synchrotron radiation micro-CT image through a denoising and diffusion implicit model to obtain a starting value for the next round of sampling; iterative sampling and solving are performed according to the starting value for the next round of sampling until the set iteration rounds are completed, and the last round of synchrotron radiation micro-CT images are output; The final synchrotron radiation micro-CT image acquisition module is used to perform denoising on the final round of synchrotron radiation micro-CT images through the Tweedie equation to obtain the final synchrotron radiation micro-CT images.

9. The synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling according to claim 8, characterized in that: In the training image perturbation processing module, in the process of obtaining the perturbed training image by applying the stochastic differential equation perturbation at the set time to the training image in the training data set, the stochastic differential equation expression at the set time is: Where x0 is the training image; s t is the drift function; σ t is the diffusion function; z is the Gaussian distribution.

10. The synchrotron radiation micro-CT reconstruction device based on decomposition diffusion sampling according to claim 9, characterized in that: In the score matching network model training module, the loss function expression of the score matching network model is: In the formula, For all possible samples,,x t ..,,x0.’s expectation is minimized; is the score matching network model; It means that under the given x0 condition, about x t The gradient of the log-conditional probability density .

Images