X-ray high-resolution image noise reduction method, electronic equipment and storage medium

Abstract

The invention provides an X-ray high-resolution image noise reduction method, electronic equipment and a storage medium, and the method comprises the steps: obtaining a local diffusion model through reserving a plurality of previous steps of a diffusion model; constructing an iterative local diffusion framework based on the local diffusion model; loading a high-resolution degraded image data set to be processed and a pre-trained local diffusion model; and iteratively executing forward noise addition and reverse processes, and outputting a final noise reduction result. According to the invention, the noise reduction requirement of large-size and high-resolution detector data can be met, so that the X-ray imaging quality is improved; according to the method, an iterative local diffusion framework is deduced theoretically, and the representation of complex data is realized by using extremely small model resources, so that an unsupervised noise reduction method independent of label training data is developed; the iterative local diffusion noise reduction framework can reconstruct details of lesions and tissues under the condition of reducing more than 75% of radiation dose damage to a patient.

Description

Technical Field

[0001] The present invention relates to the technical field of X-ray imaging, and in particular to an X-ray high-resolution image noise reduction method, electronic equipment and storage medium. Background Art

[0002] Large-size, high-resolution flat-panel detectors are mainly used for X-ray imaging detection such as digital subtraction angiography (DSA), bone micro-CT, and dental CT, while wide-body detectors are mainly used for X-ray imaging of high-end spiral CT. The former uses a small pixel detection unit (100μm to 200μm) design, which usually has significant noise in its detection data under conventional exposure conditions. Under low-dose observation conditions, the noise problem will be further aggravated, seriously affecting the imaging quality. In contrast, wide-body detectors use a millimeter-level pixel design, and their detection data noise is low under conventional exposure conditions. However, under low-dose acquisition conditions (less than 50mAs), the data acquisition process usually introduces a lot of noise, thereby reducing image quality. Therefore, studying X-ray image noise reduction methods has become one of the key technologies in the field of radiographic imaging.

[0003] At present, deep learning denoising methods mainly adopt fully supervised training strategies on simulated data. However, in real clinical environments, it is extremely difficult to obtain paired training data due to the limitations of patients' respiratory motion and radiation dose, which leads to its poor applicability in clinical applications. Although the diffusion model can avoid the use of paired data for network training and achieves a level of denoising comparable to that of the fully supervised method, it cannot sample images with a pixel matrix larger than 256×256. For example, the pixel unit matrix of a 3030-size CMOS flat-panel detector is ≥3000×3000, which makes the existing diffusion model unable to be directly applied. Therefore, it is urgent to develop unsupervised denoising methods that do not rely on labeled training data to meet the needs of large-size, high-resolution detectors for high-quality imaging. Summary of the invention

[0004] In order to achieve the above-mentioned purpose and other advantages of the present invention, the first purpose of the present invention is to provide a method for denoising an X-ray high-resolution image, comprising the following steps:

[0005] By retaining the first several steps of the diffusion model, a local diffusion model is obtained;

[0006] constructing an iterative local diffusion framework based on the local diffusion model;

[0007] Load the high-resolution degraded image dataset to be processed and the pre-trained local diffusion model;

[0008] The forward noise addition and reverse processes are iteratively performed to output the final denoising result.

[0009] Furthermore, the potential state at any time step t=1:T of the diffusion model is a direct sampling of the initial state x0, resulting in:

[0010]

[0011] in, α t =1-β t , β t is the noise variance;

[0012] By maximizing the lower bound of evidence to optimize the diffusion model, the transfer distribution of its latent variable decoder is parameterized as:

[0013]

[0014] Among them, ∈ θ is a noise predictor approximated by a neural network that is used to calculate the noise from x at each time step t. t prediction∈;

[0015] Using the reparameterization technique, from p θ (x t-1 |x t ) The sampling formula is:

[0016]

[0017] Furthermore, the local diffusion model is expressed as It only trains and uses time steps 1 to K. The iterative form of the local diffusion model is recorded as an iterative local diffusion model; wherein T is the total time step and K is the local time step.

[0018] Furthermore, the step of constructing an iterative local diffusion framework based on the local diffusion model includes:

[0019] Degrade the image Represented as the original image x0 and the degraded component s n The sum of the forms:

[0020]

[0021] definition represents the forward noise adding process, represents the reverse denoising process;

[0022] Assume that a local diffusion model is trained on a standard image dataset The distribution it models is denoted as p θ (x0), the distribution of this standard image dataset is denoted as q(x0);

[0023] Use the forward noise process to generate an intermediate state x from a sample x0~q(x0) K , the formula is:

[0024]

[0025] Use the reverse denoising process to transform the intermediate state x K Mapping back to the original image x0, the formula is:

[0026]

[0027] For degraded images Executing the same noise adding strategy, the forward noise adding formula of the degraded image is:

[0028]

[0029] Combining degraded images The representation of , we get:

[0030]

[0031] in, Indicates an approximate intermediate state;

[0032] Approximate intermediate state Perform reverse denoising to make s n The coefficient approaches 0, and the iterative formula is expressed as:

[0033]

[0034] Among them, Q i ∈(0,1) is the number of s in the i-th iteration n The attenuation coefficient, M is the total number of iterations, is the result after M iterations. As M increases, the degenerate term decreases to zero.

[0035] Furthermore, the pre-training of the local diffusion model includes:

[0036] Define the total time step T and the local time step K, and randomly initialize the noise predictor ∈ θ ;

[0037] Execute sampling x0~q(x ND ), sampling t~Uniform(1,…,K), sampling Performing Gradient Descent The network is trained until convergence to obtain the final local diffusion model.

[0038] Furthermore, the method further comprises the steps of:

[0039] Construct guidance conditions to achieve high-quality image generation. The guidance condition formula is:

[0040]

[0041] in, represents the bootstrap sampling estimate, represents the final noise predictor estimate.

[0042] Furthermore, the method further comprises the steps of:

[0043] Build on The noise intensity-dependent nonlinear attenuation strategy is as follows:

[0044]

[0045] Among them, s n (i, j) represents the estimated noise, and f(·) represents the noise intensity mapping function;

[0046] Apply the default denoising parameters once and copy the result from the noisy image Subtract from to obtain s n , and use a fourth-order polynomial function to approximate the nonlinear mapping f(·). The noise estimation formula and the noise intensity mapping function formula are:

[0047]

[0048]

[0049] Among them, a m are the polynomial coefficients.

[0050] Furthermore, the iterative execution of the forward noise addition and reverse process and the output of the final noise reduction result step include:

[0051] Performing forward noise addition according to the degraded image forward noise addition formula;

[0052] Combine the guiding condition formula and the noise intensity mapping function formula to perform the reverse process

[0053] Repeat the steps of performing the forward noise adding step according to the degraded image forward noise adding formula and performing the reverse process by combining the guiding condition formula and the noise intensity mapping function formula. Repeat the steps M times and output the final denoising result.

[0054] The second object of the present invention is to provide an electronic device, comprising: a memory on which a program code is stored; a processor connected to the memory and implementing the above method when the program code is executed by the processor.

[0055] A third object of the present invention is to provide a computer-readable storage medium having program instructions stored thereon, wherein the program instructions implement the above method when executed.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention provides an X-ray high-resolution image denoising method, an electronic device and a storage medium, which can meet the denoising requirements of large-size, high-resolution detector data, thereby improving the quality of X-ray imaging; the present invention theoretically derives an iterative local diffusion framework, uses extremely small model resources to achieve the representation of complex data, and thus develops an unsupervised denoising method that does not rely on labeled training data; the iterative local diffusion denoising framework proposed by the present invention can reconstruct the details of lesions and tissues while reducing the radiation dose damage of patients by more than 75%.

[0058] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. The specific implementation of the present invention is given in detail by the following embodiments and their accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0060] Figure 1 This is a flow chart of the X-ray high-resolution image denoising method of Example 1;

[0061] Figure 2 Schematic diagram of bone micro-CT projection noise reduction and reconstruction;

[0062] Figure 3 Schematic diagram of spiral CT low-dose data denoising and reconstruction;

[0063] Figure 4 is a schematic diagram of an electronic device of Example 2;

[0064] Figure 5 This is a schematic diagram of the storage medium of Example 3. DETAILED DESCRIPTION

[0065] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. It should be noted that, under the premise of no conflict, the embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0066] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0067] The figure numbers in this application are only used to distinguish the various steps in the scheme, and are not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0069] In the related art, the diffusion model can fit the feature distribution of highly complex data sets, but there are challenges in training high-resolution images, and the reasoning efficiency is low. Therefore, although it is theoretically feasible to apply the diffusion model to high-resolution imaging tasks, the above defects make the diffusion model unable to be truly used in actual clinical practice.

[0070] The present invention provides a new method for denoising large-size, high-resolution X-ray images. The local diffusion model proposed in the present invention can achieve the task objectives with relatively small model resources. The core of the model is to use degraded images to generate the intermediate state of diffusion, so that the diffusion and its inverse process only need to be performed locally, thereby greatly reducing the difficulty of training. In the inference stage, starting directly from the intermediate state, the efficiency is significantly improved. In addition, the present invention also constructs a reverse guidance condition to further optimize the denoising quality of X-ray high-resolution images.

[0071] Example 1

[0072] A method for denoising high-resolution X-ray images, such as Figure 1 As shown, the following steps are included:

[0073] S1, obtaining a local diffusion model by retaining the first several steps of the diffusion model;

[0074] In some embodiments, the diffusion model is actually a Markov hierarchical variational autoencoder (HVAE), whose hidden encoder (forward process) is defined as a linear Gaussian model, from Start sampling and continuously generate hidden states x1 to xT , where x 1:T has the same magnitude as x0 to q(x ND ). By formulating the noise variance table β1,..., β T , the distribution of the final hidden state x T is approximately a standard Gaussian distribution. According to α t = 1 - β t and notation, the latent state at any time step t = 1:T can be a direct sample of the initial state x0, i.e.:

[0075]

[0076] As a variant of HVAE, the diffusion model can also be optimized by maximizing the evidence lower bound (ELBO). The transition distribution of the latent variable decoder (reverse process) can be parameterized as:

[0077]

[0078] where, ∈ θ is a noise predictor approximated by a neural network, which can predict ∈ from x t at each time step t. Using the reparameterization trick, sampling from p θ (x t-1 |x t ) can be written as:

[0079]

[0080] Here, only the first K steps of the diffusion model at time step T (K << T) are retained, which is called the local diffusion model (PDM). For ease of expression, it is denoted as It is defined at time steps 1 to T, but only trained and used at time steps 1 to K. Where, T is the total number of time steps and K is the local number of time steps. The iterative form of PDM is denoted as the iterative local diffusion model (IPDM).

[0081] S2. Construct an iterative local diffusion framework based on the local diffusion model;

[0082] In some embodiments, the step of constructing an iterative local diffusion framework based on the local diffusion model includes:

[0083] Represent the degraded image as the sum of the original image x0 and the degradation component s n :

[0084]

[0085] It is proved here that IPDM can be utilized to obtain from Eliminate the degenerate component s n Specifically, we define some symbols to simplify the subsequent analysis. represents the forward noise addition process of formula (2), represents the reverse denoising process of formula (4). Specifically, suppose a The distribution it models is denoted as p θ (x0). Formula (2) can be used to convert a sample x0~q(x0) into an intermediate state x K . Then, we can use the reverse process to convert x K Mapping back to x0 looks like this:

[0086]

[0087] use Replace x0 in formula (6) and prove that the result of formula (7) still converges to the undegraded image x0. First, The same noise addition strategy is implemented as follows:

[0088]

[0089] Substituting formula (5) into formula (8), we can obtain:

[0090]

[0091] in, represents the “approximate intermediate state”, which is different from the real intermediate state x K The main difference between is the presence of weighted degradation components. The focus shifts to applying the inverse process to The impact of In the first backward step, the noise predictor ∈ θ The result can be calculated by Taylor expansion:

[0092]

[0093] If the training is converged, it should satisfy:

[0094]

[0095] Substituting formula (11) and formula (10) into formula (4), we get:

[0097] (12)

[0098] Iterate the above operation until

[0099]

[0100] in, represents the result of the first iteration. have:

[0101]

[0102] This means that the above operation can be repeated so that s n The coefficient of approaches 0, thus achieving noise elimination. The iterative formula can be expressed as:

[0103]

[0104] Among them, Q i ∈(0,1) is the number of s in the i-th iteration n The attenuation coefficient, M is the total number of iterations, is the result after M iterations. As M increases, the degenerate term decreases to zero.

[0105] In some embodiments, more than 30,000 high-resolution image slices are used to pre-train the local diffusion model to ensure its accuracy in modeling high-quality image sampling. At the same time, the total time step T and the local time step K are defined, and the noise predictor ρ is randomly initialized;

[0106] Next, the sampling x0~q(x ND ), sampling t~Uniform(1,…,K), sampling Performing Gradient Descent The network is trained until convergence to obtain the final local diffusion model (PDM).

[0107] In some embodiments, in order to generate a clear image corresponding to the high-noise image to be tested and to reduce the influence of sampling errors, a guiding condition is constructed to achieve high-quality image generation. The specific guiding conditions are as follows:

[0108]

[0109] In the above formula, represents the bootstrap sampling estimate, represents the final noise predictor estimate.

[0110] In the present invention, use Its effectiveness as a guide gradually decreases during the sampling process. Initially, It provides relatively accurate directions in the early to middle stages of sampling; however, its inherent bias adversely affects sampling accuracy in the later stages. A sensible approach to setting τ is to use a large value in the initial stages of sampling and gradually reduce it to zero. In addition, considering The noise distribution is uneven, and it is unwise to adopt a uniform τ attenuation strategy on all pixels. Therefore, in some embodiments, a method based on The noise intensity-dependent nonlinear attenuation strategy is called the power-weighted cosine scheme, which is as follows:

[0111]

[0112] In the above formula, K represents the local time step of PDM, s n (i, j) represents the estimated noise, and f(·) represents the noise intensity mapping function. Given that only an approximate noise amplitude is required, a simple approach can be chosen: apply the default denoising parameters once and pass the result from the noisy image Subtract from to obtain s n . And use the fourth-order polynomial function to approximate the nonlinear mapping f(·). The above process is expressed as follows:

[0113]

[0114] In the above formula, the polynomial coefficient a m Specific adjustments are required based on different test data sets.

[0115] 3. Load the high-resolution degraded image dataset to be processed and the pre-trained local diffusion model;

[0116] S4, iteratively perform the forward noise addition and reverse process, and output the final noise reduction result.

[0117] In combination with the above embodiment, the iterative execution of the forward noise addition and reverse process and the step of outputting the final noise reduction result include:

[0118] Load the high-resolution degraded image dataset q(x LD ) and the pre-trained local diffusion model

[0119] For the iterative local diffusion denoising process i=1:M: perform forward noise addition according to formula (8); perform the reverse process by combining formula (16) and formula (17) Repeat the above steps M times and output the final denoising result

[0120] Bone micro-CT projection noise reduction and reconstruction Figure 2 As shown. Figure 2As shown on the left side of the figure, due to the large amount of noise in the projection data collected by the large-array flat-panel detector, there are noise and artifacts in the image of the bone micro-CT reconstruction result, which blurs the microscopic details of the bone; Figure 2 As shown on the right side of , after the iterative local diffusion denoising process proposed in this embodiment is adopted, the noise in the reconstructed image is greatly suppressed, and the bone microstructure details can be clearly observed.

[0121] Denoising and reconstruction of low-dose spiral CT data Figure 3 As shown. Figure 3 As shown in the second row of #1 and #2 in , when the radiation dose is reduced by more than 75%, the reconstruction results contain a lot of image noise, resulting in blurred lesions and tissue details; Figure 3 As shown in the first row of #1 and #2 in FIG. 1 , after the iterative local diffusion noise reduction processing proposed in this embodiment is adopted, the noise in the reconstructed image is greatly suppressed, and the lesions and tissue details can be clearly observed.

[0122] The present embodiment provides a method for denoising an X-ray high-resolution image, which is suitable for a variety of applications such as denoising of large-size, high-resolution detectors, denoising of spiral CT low-dose data, and denoising of bone micro-CT data; the method provided by the present embodiment can meet the denoising requirements of large-size, high-resolution detector data, thereby improving the quality of X-ray imaging; the present embodiment theoretically derives an iterative local diffusion framework, which uses extremely small model resources to achieve the representation of complex data, thereby developing an unsupervised denoising method that does not rely on labeled training data; the iterative local diffusion denoising framework proposed in the present embodiment can reconstruct the details of lesions and tissues while reducing the radiation dose damage to patients by more than 75%.

[0123] Example 2

[0124] An electronic device, such as Figure 4 As shown, it includes: a memory on which program codes are stored; a processor connected to the memory, and when the program codes are executed by the processor, a method for reducing noise of X-ray high-resolution images is implemented. For a detailed description of the method, reference can be made to the corresponding description in the above method embodiment, which will not be repeated here.

[0125] Example 3

[0126] A computer readable storage medium such as Figure 5 As shown, program instructions are stored thereon, and when the program instructions are executed, a method for reducing noise of X-ray high-resolution images is implemented. For a detailed description of the method, reference may be made to the corresponding description in the above method embodiment, and no further description is given here.

[0127] The number of devices and processing scales described here are used to simplify the description of the present invention. Applications, modifications and variations of the present invention will be obvious to those skilled in the art.

[0128] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and implementation modes. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

[0129] The apparatus, electronic device, non-volatile computer storage medium and method provided in the embodiments of this specification correspond to each other, and therefore, the apparatus, electronic device and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, electronic device and non-volatile computer storage medium will not be repeated here.

[0130] Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices for implementing various functions included therein can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software units for implementing the method and structures within the hardware component.

[0131] The systems, devices or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described separately by functions in various units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or more software and / or hardware.

[0132] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification may be in the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the embodiments of this specification may be in the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0133] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0134] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0136] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0137] The specification may be described in the general context of computer-executable instructions executed by a computer, such as program units. Generally, program units include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program units may be located in local and remote computer storage media, including storage devices.

[0138] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0139] The above description is only an embodiment of this specification and is not intended to limit one or more embodiments of this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of one or more embodiments of this specification.

Claims

1. A method for denoising an X-ray high-resolution image, characterized in that: The following steps are involved: By retaining the first several steps of the diffusion model, a local diffusion model is obtained; constructing an iterative local diffusion framework based on the local diffusion model; Load the high-resolution degraded image dataset to be processed and the pre-trained local diffusion model; The forward noise addition and reverse processes are iteratively performed to output the final denoising result.

2. The X-ray high-resolution image denoising method according to claim 1, characterized in that: The potential state at any time step t=1:T of the diffusion model is a direct sampling of the initial state x0, yielding: in, α t =1-β t , β t is the noise variance; By maximizing the lower bound of evidence to optimize the diffusion model, the transfer distribution of its latent variable decoder is parameterized as: where α is a noise predictor approximated by a neural network, which is used to calculate the noise from x at each time step t. t prediction∈; Using the reparameterization technique, from p θ (x t-1 |x t ) The sampling formula is:

3. The X-ray high-resolution image denoising method according to claim 1, characterized in that: The local diffusion model is expressed as It only trains and uses time steps 1 to K. The iterative form of the local diffusion model is recorded as an iterative local diffusion model; wherein T is the total time step and K is the local time step.

4. The method for reducing noise of an X-ray high-resolution image according to claim 3, characterized in that: The step of constructing an iterative local diffusion framework based on the local diffusion model comprises: Degrade the image Represented as the original image x0 and the degraded component s n The sum of the forms: definition represents the forward noise adding process, represents the reverse denoising process; Assume that a local diffusion model is trained on a standard image dataset The distribution it models is denoted as p θ (x0), the distribution of this standard image dataset is denoted as q(x0); Use the forward noise process to generate an intermediate state x from a sample x0~q(x0) K , the formula is: Use the reverse denoising process to transform the intermediate state x K Mapping back to the original image x0, the formula is: For degraded images Executing the same noise adding strategy, the forward noise adding formula of the degraded image is: Combining degraded images The representation form is: in, Indicates an approximate intermediate state; Approximate intermediate state Perform reverse denoising to make s n The coefficient approaches 0, and the iterative formula is expressed as: Among them, Q i ∈(0,1) is the number of s in the i-th iteration n The attenuation coefficient, M is the total number of iterations, is the result after M iterations. As M increases, the degenerate term decreases to zero.

5. The method for reducing noise of an X-ray high-resolution image according to claim 4, characterized in that: The pre-training of the local diffusion model includes: Define the total time step T and the local time step K, and randomly initialize the noise predictor ∈ θ ; Execute sampling x0~q(x ND ), sampling t~Uniform(1,…,K), sampling Performing Gradient Descent The network is trained until convergence to obtain the final local diffusion model.

6. The X-ray high-resolution image denoising method according to claim 4, characterized in that: Also includes the steps: Construct guidance conditions to achieve high-quality image generation. The guidance condition formula is: in, represents the bootstrap sampling estimate, represents the final noise predictor estimate.

7. The method for denoising an X-ray high-resolution image according to claim 6, characterized in that: Also includes the steps: Build on The noise intensity-dependent nonlinear attenuation strategy is as follows: Among them, s n (i, j) represents the estimated noise, and f(·) represents the noise intensity mapping function; Apply the default denoising parameters once and copy the result from the noisy image Subtract from to obtain s n , and use a fourth-order polynomial function to approximate the nonlinear mapping f(·). The noise estimation formula and the noise intensity mapping function formula are: Among them, a m are the polynomial coefficients.

8. The X-ray high-resolution image denoising method according to claim 7, characterized in that: The iterative execution of the forward noise addition and reverse process and the output of the final noise reduction result step include: Performing forward noise addition according to the degraded image forward noise addition formula; Combine the guiding condition formula and the noise intensity mapping function formula to perform the reverse process Repeat the steps of performing the forward noise adding step according to the degraded image forward noise adding formula and performing the reverse process by combining the guiding condition formula and the noise intensity mapping function formula. Repeat the steps M times and output the final denoising result.

9. An electronic device, characterized in that: include: A memory having program code stored therein; A processor is connected to the memory, and when the program code is executed by the processor, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that: Program instructions are stored thereon, and when the program instructions are executed, the method according to any one of claims 1 to 8 is implemented.

Images