An X-ray high-resolution image denoising method, an electronic device, and a storage medium

Abstract

The application provides an X-ray high-resolution image denoising method, an electronic device and a storage medium, comprising the following steps: obtaining a local diffusion model by retaining the first several steps of a diffusion model; constructing an iterative local diffusion framework based on the local diffusion model; loading a high-resolution to-be-processed degraded image dataset and a pre-trained local diffusion model; iteratively performing forward noise addition and reverse processes, and outputting a final denoising result. The application can meet the denoising needs of large-size and high-resolution detector data, thereby improving the quality of X-ray imaging; the application theoretically deduces an iterative local diffusion framework, uses a small amount of model resources to realize the representation of complex data, thereby developing an unsupervised denoising method which does not depend on labeled training data; and the iterative local diffusion denoising framework can reconstruct the details of lesions and tissues while reducing the radiation dose damage to patients by more than 75%.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of X-ray imaging, in particular to an X-ray high-resolution image denoising method, an electronic device and a storage medium. BACKGROUND

[0002] Large-size high-resolution flat panel detectors are mainly applied to 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 usually has significant noise in the detection data under normal exposure conditions due to the design of small pixel detection units (100-200 μm), and the noise problem will be further aggravated under low-dose observation conditions, which seriously affects the imaging quality. In contrast, the wide-body detector adopts a millimeter-level pixel design, and its detection data noise is relatively low under normal exposure conditions. However, under low-dose acquisition conditions (less than 50 mAs), the data acquisition process usually introduces a large amount of noise, thereby reducing the image quality. Therefore, the study of X-ray image denoising methods has become one of the key technologies in the field of X-ray imaging.

[0003] At present, the denoising method of deep learning mainly uses a full-supervised training strategy on simulated data. However, in real clinical environments, it is extremely difficult to obtain paired training data due to patient respiratory motion and radiation dose limitations, which leads to poor applicability in clinical applications. Although the diffusion model can avoid using paired data for network training, and its denoising effect reaches a level comparable to that of the full-supervised method, it cannot sample images with a pixel matrix larger than 256x256. For example, the pixel unit matrix of a 3030-size CMOS flat panel detector is ≥3000x3000, which makes the existing diffusion model unable to be directly applied. Therefore, there is an urgent need to develop an unsupervised denoising method that does not rely on labeled training data to meet the demand for high-quality imaging of large-size, high-resolution detectors. SUMMARY

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

[0005] obtaining a local diffusion model by retaining the first several steps of the diffusion model;

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

[0007] loading a high-resolution image data set to be processed and a pre-trained local diffusion model;

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

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

[0010]

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

[0012] The diffusion model is optimized by maximizing the lower bound of the evidence, with the transition distribution of its latent variable decoder parameterized as:

[0013]

[0014] where, ∈ θ is a noise predictor approximated by a neural network to predict ∈ from x t at each time step t;

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

[0016]

[0017] Further, the local diffusion model is represented as which only trains and uses time steps 1 to K, and its iterative form is denoted as the iterative local diffusion model; where T is the total time step, and K is the local time step.

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

[0019] The degraded image is represented in the form of the sum of the original image x0 and the degradation component s n :

[0020]

[0021] Define to represent the forward noise adding process, to represent the backward denoising process;

[0022] Suppose a local diffusion model is trained on a standard image dataset , the distribution modeled by it is represented as p θ (x0), and the distribution of the standard image dataset is represented as q(x0);

[0023] An intermediate state x is generated from a sample x0~q(x0) using a forward noise adding process K , which is given by:

[0024]

[0025] The intermediate state x is mapped back to the original image x0 using a backward noise removing process K , which is given by:

[0026]

[0027] The same noise adding strategy is performed on the degraded image , and the forward noise adding formula for the degraded image is:

[0028]

[0029] Combining the representation of the degraded image , we have:

[0030]

[0031] where denotes the approximate intermediate state;

[0032] The approximate intermediate state is processed by a backward noise removing process, so that the coefficient of s n tends to 0, and the iterative formula is given by:

[0033]

[0034] where Q i is the decay coefficient of s n in the i-th iteration, and M is the total number of iterations, is the result after M iterations, and the degradation term decreases to zero as M increases.

[0035] Further, the pre-training of the local diffusion model comprises:

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

[0037] Sample x0~q(x ND ), sample t~Uniform(1,…,K), and sample Perform gradient descent training on the network until it converges, and obtain the final local diffusion model.

[0038] Further, it further comprises the steps of:

[0039] The guiding condition is constructed to realize high-quality image generation, and the guiding condition formula is:

[0040]

[0041] wherein, represents a guiding sampling estimation, represents a final noise predictor estimation.

[0042] Further, the method further comprises the step of:

[0043] The noise intensity-dependent nonlinear attenuation strategy based on is constructed, and the specific strategy is as follows:

[0044]

[0045] wherein, s n (i,j) represents an estimated noise, and f(·) represents a noise intensity mapping function;

[0046] The default denoising parameter is applied once, and the result is subtracted from the noise image to obtain s n , and a fourth-order polynomial function is used to approximate the nonlinear mapping f(·), and the noise estimation formula and the noise intensity mapping function formula are:

[0047]

[0048]

[0049] wherein, a m is a polynomial coefficient.

[0050] Further, the method further comprises the step of:

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

[0052] Performing inverse process combined with the guiding condition formula and the noise intensity mapping function formula

[0053] Repeating the step of performing forward noise addition according to the forward noise addition formula of the degraded image and the step of performing inverse process combined with the guiding condition formula and the noise intensity mapping function formula M times, and outputting a final denoising result

[0054] The second object of the present application is to provide an electronic device, comprising a memory having program codes stored thereon, and a processor connected with the memory, and when the program codes are executed by the processor, the above method is realized.

[0055] The third object of the present application is to provide a computer readable storage medium having program instructions stored thereon, and when the program instructions are executed, the above method is realized.

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

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

[0058] The above description is only a summary of the technical solutions of the present application, in order to more clearly understand the technical means of the present application, and the content of the specification can be implemented, the following preferred embodiments of the present application are described in detail with the help of the accompanying drawings. The specific embodiments of the present application are given in detail by the following examples and their accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0059] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application, the schematic embodiments of the present application and their description are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0060] Figure 1 It is the X-ray high-resolution image denoising method flowchart of example 1;

[0061] Figure 2 It is the bone micro-CT projection denoising and reconstruction schematic diagram;

[0062] Figure 3 It is the spiral CT low dose data denoising and reconstruction schematic diagram;

[0063] Figure 4 It is the electronic device schematic diagram of example 2;

[0064] Figure 5 It is the storage medium schematic diagram of example 3. DETAILED DESCRIPTION

[0065] The application will be further described below in conjunction with the drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. It should be noted that, under the condition of no conflict, the embodiments described below or the technical features between the embodiments can be combined to form new embodiments.

[0066] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative labor fall within the scope of protection of the present application.

[0067] The figure numbers in the present application are only used to distinguish the steps in the scheme, and are not used to limit the execution order of the 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 commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application.

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

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

[0071] Embodiment 1

[0072] An X-ray high-resolution image denoising method, as shown in Figure 1 , includes the following steps:

[0073] S1, a local diffusion model is obtained 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), and the hidden encoder (forward process) thereof is defined as a linear Gaussian model, which is sampled from Start sampling, and continuously generate hidden states x1 to xT where x 1:T has the same size as x ND 0: q(x T t) is formulated by specifying the noise variances β T 1,..., β t T. The distribution of the final hidden state x t T is approximately a standard Gaussian distribution according to α = 1 - β

[0075]

[0076] The diffusion model as a variant of HVAE can also be optimized by maximizing the evidence lower bound (ELBO). The transition distribution of the latent variable decoder (the backward 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 with time step T (K << T) are kept, which is called the partial diffusion model (PDM). For ease of expression, it is denoted as It is defined over time steps 1 to T, but only time steps 1 to K are trained and used. Where T is the total time step and K is the partial time step. The iterative form of PDM is denoted as the iterative partial diffusion model (IPDM).

[0081] S2, constructing an iterative partial diffusion framework based on the partial diffusion model;

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

[0083] The degraded image is expressed in the form of the sum of the original image x n 0 and the degradation component s

[0084]

[0085] It is proved here that IPDM can be used to recover the original image s n Specifically, some notations are defined to simplify the analysis in the following, denotes the forward noise adding process of equation (2), denotes the backward noise removing process of equation (4). Specifically, it is assumed that a is trained on a standard image dataset (i.e. un-degraded images, whose distribution is denoted as q(x0)). θ (x0). One sample x0~q(x0) can be generated into an intermediate state x K using equation (2). Subsequently, x K can be mapped back to x0using the backward process as follows:

[0086]

[0087] Substitute x0in equation (6) with and prove that the result of equation (7) still converges to the un-degraded image x0. First, perform the same noise adding strategy on , which is denoted as follows:

[0088]

[0089] Substitute equation (5) into equation (8) to get:

[0090]

[0091] where denotes the "approximate intermediate state", which is different from the true intermediate state x K mainly in that there is a weighted degraded component. The focus shifts to the impact of applying the backward process on , i.e. computing The result of the noise predictor e θ in the first backward step can be computed through Taylor expansion:

[0092]

[0093] If trained to convergence, the following should be satisfied:

[0094]

[0095] Substitute equation (11) and equation (10) into equation (4) to get:

[0096] (12)

[0097] Iterate the above operation until

[0098]

[0099] wherein, denotes the result of the first iteration. Because there is:

[0100]

[0101] This means that the above operation can be repeated, making s n tend to 0, thus achieving denoising. The iterative formula can be expressed as:

[0102]

[0103] wherein, Q i ∈(0, 1) is the attenuation coefficient of s n in the i-th iteration, M is the total number of iterations, is the result after M iterations. As M increases, the degenerate term decreases to zero.

[0104] In some embodiments, more than 30,000 high-resolution image slices are used to pre-train the local diffusion model to ensure its modeling accuracy for 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 p is randomly initialized;

[0105] Next, sample x0~q(x ND ), sample t~Uniform(1,…,K), sample Perform gradient descent Train the network until it converges to obtain the final local diffusion model (PDM).

[0106] In some embodiments, in order to generate a clear image corresponding to a high-noise image to be tested, and at the same time to reduce the influence of sampling error, a guide condition is constructed to realize high-quality image generation, and the specific guide condition is as follows:

[0107]

[0108] In the above formula, denotes the guide sampling estimate, denotes the final noise predictor estimate.

[0109] In the present application, the effectiveness of as a guide gradually weakens in the sampling process. Initially, The early-to-mid stage of sampling provides relatively accurate directions; however, its inherent bias adversely affects the sampling accuracy in the late stage. A sensible approach to set τ is to adopt a larger value in the initial stage of sampling and gradually decrease it to zero. Moreover, considering the non-uniformity of the noise distribution, it is not sensible to adopt a uniform τ decay strategy across all pixels. Therefore, a noise-intensity-dependent nonlinear decay strategy, called power-weighted cosine planning, is proposed in some embodiments, as follows:

[0110]

[0111] In the above equation, K denotes the PDM local time step, s n (i,j) denotes the estimated noise, and f(·) denotes the noise intensity mapping function. Given that only an approximate noise amplitude is needed, a simple approach can be chosen: apply the default denoising parameters once and subtract the result from the noisy image to obtain s n . A fourth-order polynomial function is used to approximate the nonlinear mapping f(·). The above procedure is represented as follows:

[0112]

[0113] In the above equation, the polynomial coefficients a m need to be adjusted based on different test datasets.

[0114] 3. Load the high-resolution image dataset to be processed and the pre-trained local diffusion model;

[0115] S4. Iteratively perform the forward noise addition and inverse process, and output the final denoised result.

[0116] In combination with the above embodiments, the step of iteratively performing the forward noise addition and inverse process, and outputting the final denoised result includes:

[0117] Load the high-resolution image dataset to be processed q(x LD ) and the pre-trained local diffusion model

[0118] For the iterative local diffusion denoising process i = 1: M: perform forward noise addition according to equation (8); perform the inverse process jointly according to equations (16) and (17) Repeat the above steps M times and output the final denoised result

[0119] Bone micro-CT projection denoising and reconstruction is shown in Figure 2 , and the denoising process is shown in Figure 2 ​​As shown on the left, the large amount of noise in the projection data acquired by the large-area flat panel detector leads to noise and artifacts in the bone micro-CT reconstruction image, thus blurring the microscopic details of the bone; for example... Figure 2 As shown on the right, after using the iterative local diffusion noise reduction processing proposed in this embodiment, the noise in the reconstructed image is greatly suppressed, and the details of the bone microstructure can be clearly observed.

[0120] Denoising and reconstruction of low-dose data from spiral CT Figure 3 As shown. Figure 3 As shown in the second row of #1 and #2, when the radiation dose is reduced by more than 75%, the reconstruction results contain a large amount of image noise, resulting in blurred lesions and tissue details; as... Figure 3 As shown in the first row of #1 and #2, after using the iterative local diffusion noise reduction process proposed in this embodiment, the noise in the reconstructed image is greatly suppressed, and the lesions and tissue details can be clearly observed.

[0121] This embodiment provides a high-resolution X-ray image denoising method applicable to various applications such as denoising of large-size, high-resolution detectors, denoising of low-dose spiral CT data, and denoising of bone micro-CT data. The method provided in this embodiment can meet the denoising requirements of large-size, high-resolution detector data, thereby improving the quality of X-ray imaging. This embodiment theoretically derives an iterative local diffusion framework, which uses minimal model resources to represent complex data, thereby developing an unsupervised denoising method that does not rely on labeled training data. The iterative local diffusion denoising framework proposed in this embodiment can reconstruct the details of lesions and tissues while reducing radiation dose damage to patients by more than 75%.

[0122] Example 2

[0123] An electronic device, such as Figure 4 As shown, the method includes: a memory storing program code; and a processor connected to the memory, which, when executed by the processor, implements a method for denoising high-resolution X-ray images. For a detailed description of the method, please refer to the corresponding description in the above method embodiments, which will not be repeated here.

[0124] Example 3

[0125] A computer-readable storage medium, such as Figure 5 As shown, it stores program instructions, which, when executed, implement a method for denoising high-resolution X-ray images. For a detailed description of the method, please refer to the corresponding description in the above method embodiments; it will not be repeated here.

[0126] The number of devices and the processing size described herein are used to simplify the description of the present application. Applications, modifications and variations of the present application will be apparent to those skilled in the art.

[0127] Although embodiments of the present application have been disclosed in connection with the explicitly described embodiments, it should be understood that many modifications, substitutions and changes can be made by those skilled in the art to the disclosed embodiments without departing from the spirit and scope of the application as defined by the appended claims. Therefore, the present application is not intended to be limited to the particular embodiments described.

[0128] The apparatus, electronic device, non-volatile computer storage medium and method provided by the embodiments of the present application are corresponding, therefore, the apparatus, electronic device, 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, therefore, the beneficial technical effects of the corresponding apparatus, electronic device, non-volatile computer storage medium will not be described here.

[0129] Those skilled in the art will also appreciate that, in addition to being implemented in purely computer readable program code means, the controller can also be implemented by means of logic programmed into a logic gate array, switch array, application specific integrated circuit, programmable logic controller and embedded microcontroller, etc. to perform the same functions. Such a controller can therefore be considered as a hardware component, and the means included therein for performing the various functions can be considered as structures within the hardware component. Alternatively, the means for performing the various functions can even be considered as both software elements implementing the method and structures within the hardware component.

[0130] The system, apparatus or unit illustrated by the above embodiments can be specifically implemented by a computer chip or an entity, or by a product with certain functions. For the convenience of description, the above apparatus is described by dividing into various units according to functions during description. Of course, the functions of each unit can be implemented in the same or multiple software and / or hardware when implementing one or more embodiments of the present application.

[0131] Those skilled in the art will appreciate that the embodiments of the present application can be provided as a method, system or computer program product. Therefore, the embodiments of the present application can adopt a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present application can adopt a computer program product in the form of being implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0132] The specification is presented with reference to flow diagrams and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing system or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagram and / or block diagram in the flow diagrams and / or block diagrams can also be implemented by the functional blocks of a computer system. Figure 1 The flow diagram and / or block diagram in the flow diagrams and / or block diagrams can also be implemented by the functional blocks of a computer system.

[0133] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagram and / or block diagram in the flow diagrams and / or block diagrams can also be implemented by the functional blocks of a computer system. Figure 1 The flow diagram and / or block diagram in the flow diagrams and / or block diagrams can also be implemented by the functional blocks of a computer system.

[0134] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagram and / or block diagram in the flow diagrams and / or block diagrams can also be implemented by the functional blocks of a computer system. Figure 1 The flow diagram and / or block diagram in the flow diagrams and / or block diagrams can also be implemented by the functional blocks of a computer system.

[0135] It should also be noted that the terms "comprising," "including," and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements in the list, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0136] This specification may be described in the general context of computer-executable instructions, such as program units, that are executed by a computer. Generally, program units include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification may also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program units may reside in local and remote computer storage media, including storage devices.

[0137] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0138] The above description is merely an embodiment of this specification and is not intended to limit the scope of one or more embodiments of this specification. Various modifications and variations can be made to one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of one or more embodiments of this specification.

Claims

1. A method for X-ray high-resolution image denoising, characterized in that, The method comprises the following steps: K steps of a diffusion model with a time step T are reserved to obtain a local diffusion model, wherein K << T; Any time step of the diffusion model The potential state at the initial state Direct sampling of the initial state, yielding: ； wherein , , is the noise variance; The diffusion model is optimized by maximizing the lower bound of evidence, and the transition distribution of a latent variable decoder of the diffusion model is parameterized as: ； wherein, is a noise predictor approximated by a neural network to predict from at each time step t; Using the reparameterization trick, from The formula for sampling is: ； The local diffusion model is represented as which only trains and uses time steps 1 to K, an iterative form of the local diffusion model is denoted as an iterative local diffusion model; wherein T is the total time step, and K is the local time step. An iterative local diffusion framework is constructed based on the local diffusion model; The step of constructing the iterative local diffusion framework based on the local diffusion model comprises: degraded image is represented as an original image and a degraded component in the form of a sum: ； Definitions denotes a forward noise adding process, denotes a reverse noise removing process; Assume a local diffusion model is trained on a standard image dataset , the distribution of which is represented as , the distribution of which is represented as ; using a forward noise adding process to one sample generating an intermediate state , which is given by ； using a reverse denoising process map back to the original image , which is ； Degrading images The same noise addition strategy is performed, and the degrading image forward noise formula is: ； Degraded image Combining the representation of the degraded image ； wherein represents an approximate intermediate state; to an intermediate state The iteration formula is expressed as:​ ； in, In the i-th iteration The attenuation coefficient, This represents the total number of iterations. for The results after iteration, as As the value increases, the degradation term decreases to zero; A high-resolution to-be-processed degraded image dataset and a pre-trained local diffusion model are loaded; A forward noise addition and an inverse process are iteratively performed to output a final denoising result.

2. The method of claim 1, wherein: The pre-training of the local diffusion model comprises: define a total time step T and a local time step K, and randomly initialize the noise predictor ; Sequentially performing sampling , Sampling , Sampling , Performing gradient descent , Training the network until convergence, obtaining the final local diffusion model.

3. The X-ray high-resolution image noise reduction method as described in claim 1, characterized in that, The method further comprises the following steps: A guide condition is constructed to achieve high-quality image generation, and a guide condition formula is: ； wherein, denotes the pilot-sampling estimate, denotes the final noise predictor estimate.

4. The X-ray high-resolution image denoising method of claim 3, wherein, The method further comprises the following steps: Constructing a noise intensity-dependent nonlinear attenuation strategy based on The specific implementation is as follows: ； wherein, represents an estimated noise, represents a noise intensity mapping function; The default denoising parameters are applied once and the result is subtracted from the noisy image to obtain and a fourth order polynomial function is used to approximate the non-linear mapping The noise estimation formula and the noise intensity mapping function formula are:​ ； ； wherein are polynomial coefficients.

5. The method of claim 4, wherein, The step of iteratively performing the forward noise addition and the inverse process to output the final denoising result comprises: The forward noise addition is performed according to the forward noise addition formula of the degraded image; performing the reverse procedure in combination with the guidance condition formula and the noise intensity mapping function formula ; The step of performing forward noise addition according to the degradation image forward noise adding formula and the step of performing inverse process according to the guide condition formula and the noise intensity mapping function formula are repeatedly performed Step M times, and output the final denoising result .

6. An electronic device, comprising: It comprises: A memory having program codes stored thereon; A processor connected with the memory, and when the program codes are executed by the processor, the method according to any one of claims 1-5 is implemented.

7. A computer-readable storage medium, characterized in that, A memory having program codes stored thereon, and when the program codes are executed, the method according to any one of claims 1-5 is implemented.

Images