WO2022052114A1 - Ct image reconstruction method and system, and device and medium - Google Patents

Abstract

A CT image reconstruction method and system, and a device and a medium. The method comprises: performing iterative image reconstruction on an image; determining whether an iterated image meets an iteration stop condition; if so, stopping iteration; and if not, continuing to perform iteration. Each round of iteration processing successively comprises update processing, smoothing processing and fusion processing, wherein the update processing comprises solving image roughness by means of calculating an intensity difference between adjacent patches, and then acquiring a maximum penalized likelihood function. By means of the present reconstruction method, both the image quality and the image generation efficiency can be taken into consideration while ensuring security, thereby outputting an image with higher quality at a faster speed and also retaining fine features of the image, which has important significance in the field of medical images.

Description

A CT image reconstruction method, system, device and medium technical field

The present invention relates to the field of CT image processing, in particular, to a CT image reconstruction method, system, device and medium.

Background technique

CT technology has the characteristics of fast, accurate, non-invasive, and painless, and can clearly obtain the attenuation information of different tissues of the human body to X-rays on the millimeter scale, thus providing information of human organs and tissues for clinicians to diagnose and prevent. The specific process of CT imaging is as follows: the detector records the projection data of the X-ray passing through the object under different viewing angles, and by reconstructing the projection data, the linear attenuation coefficient map of the fault can be obtained. CT reconstruction algorithms are mainly divided into two categories: analytical reconstruction algorithms and iterative reconstruction algorithms.

The most widely used algorithm in analytical reconstruction algorithms is Filteredback-projection (FBP). FBP is based on Radon transform for image reconstruction, but FBP does not consider the spatiotemporal inhomogeneity of the system response, nor does it take into account the instrument The noise during measurement does not have anti-noise performance, so when the signal-to-noise ratio of the projection data is low, the reconstruction quality of the CT image cannot be guaranteed.

Iterative reconstruction algorithms include algebraic reconstruction and statistical reconstruction. Among them, algebraic reconstruction mainly includes algebraic reconstruction algorithm and some new algorithms which are further extended on this basis. The maximum likelihood-expectation maximization (ML-EM) method in statistical reconstruction is widely used in clinical and practice because of its better performance than traditional algorithms in lesion detection.

In the process of X-ray transmission, part of the energy will be transferred to the human body, causing physical damage, abnormal metabolism of the human body, and even cancer. The problem of low-dose CT has gradually become a research hotspot. Decreasing the X-ray tube amperage to reduce the exposure dose per viewing angle is one of the most common ways to reduce CT dose. Reducing the tube current intensity can reduce the radiation dose of a single scan, but it will reduce the signal-to-noise ratio of the projection data, and the noise intensity will increase exponentially. Because the filtered back projection method has no anti-noise performance, the reconstructed low-dose CT images will be severely degraded and contain a large number of stripe artifacts. Therefore, low-dose CT image reconstruction generally adopts an iterative reconstruction algorithm.

However, the traditional iterative image reconstruction method based on pixels for image reconstruction requires a lot of iterative processing, and the reconstruction process is complicated and tedious. The maximum likelihood-expectation maximization method will degrade the image quality as the number of iterations increases, resulting in "checkerboard artifacts", which seriously affect the reconstruction effect. Because it is processed based on small pixels, some fine texture features in the image will be eliminated, and these fine texture features will often affect the judgment of CT images. In addition, the large amount of noise brought about by reducing the CT dose will cause edge effects, which will deteriorate the ability of image edge retention and seriously affect the differentiation of image edges.

To sum up, the existing technology generally has defects such as complex image reconstruction process, a lot of noise and artifacts in the reconstructed image, poor edge retention ability, and the original feature texture in the image is not preserved, etc., resulting in long reconstruction time, poor reconstructed image quality, Consequences such as missing image content seriously affect the diagnosis and treatment of patients. Therefore, there is a need for an image reconstruction method that can shorten the reconstruction process, improve the image quality, and preserve the image features.

SUMMARY OF THE INVENTION

Based on the problems existing in the prior art, the present invention provides a CT image reconstruction method, system, device and medium. The specific plans are as follows:

A method for reconstructing a CT image, comprising the following steps: S1, obtaining an initialization image according to projection data; S2, taking the initialization image as an image to be reconstructed, and performing image iterative processing on the image to be reconstructed; S3, judging whether the iteration stop condition is met, if so, Then stop the iteration, and output the iteratively processed image; if not, return to S2, and use the iteratively processed image as the image to be reconstructed for iterative processing again. In the S2, the image iterative processing further includes the following steps: S21, update processing, including solving the image roughness by calculating the intensity difference between adjacent patches, and then obtain a maximum penalty likelihood function; S22, smoothing processing, Including smoothing and denoising processing of the image from S21; S23, fusion processing, including performing regional fusion processing on the image from S22, including pixel-by-pixel fusion into an image.

Further, in the update process of S21, the patch used for calculating the roughness of the image includes a square pixel area composed of multiple pixels and centered on one pixel, and is used for calculating all the roughness of the image. The dough pieces are the same size.

Further, in the update process of S21, the expression of the maximum penalty likelihood function is:

where μ is the estimated image value, Φ(μ) is the objective function, L(y|μ) is the likelihood function, β is the regularization parameter that controls data fidelity and spatial smoothness, and U(μ) is the image Roughness, the maximum penalty likelihood function is obtained by solving the image roughness, and the image roughness expression is:

where ψ(μ _p - μ _q ) is the penalty function, μ _p - μ _q is the distance between pixel p and adjacent pixel q, and ω _pq is the distance between pixel p and pixel q in the neighborhood Weighting factor, n _p is the total number of detectors and N _p is the total number of image pixels.

Further, in the update process of S21, the image roughness based on the patch is calculated by the following method;

Pixel j and pixel k are respectively the center pixels of two adjacent patches, and the image roughness is calculated by calculating the intensity difference between the two patches, and then the maximum penalty likelihood function is obtained, and the difference between the pixel j and the pixel k is The pixel distance expression is:

The image roughness expressions of the pixel j and the pixel k are:

where: f _j (μ) is the feature vector composed of all pixel intensity values of the patch centered at the pixel j, f _k (μ) is composed of all the pixel intensity values of the patch centered at the pixel k , j _l denotes the lth pixel in the patch centered on pixel j, k _l denotes the lth pixel in the patch centered on pixel k, n _j denotes the total number of detectors, N _j denotes The number of all pixels in the patch, h _l is the positive weighting factor of the normalized inverse spatial distance between pixels, ψ(||f _j (μ)-f _k (μ)‖ _h ) is the penalty function.

In particular, the penalty function expression is:

Among them, δ is the attenuation factor;

When ||f _j (μ)-f _k (μ)‖ _h |<<δ, the penalty function approximates a quadratic function;

The penalty function approximates an absolute function when ||f _j (μ)-f _k (μ)‖ _h |>>δ.

Further, an optimized transmission algorithm is also included in the update process of the S21;

The optimized transmission algorithm includes constructing a surrogate function at each iteration to minimize the objective function Φ(μ).

其中n+1为迭代次数，j表示所述面片的中心像素序号，EM表示期望最大化。 Further, the optimized transmission algorithm includes an expectation-maximization algorithm, and the algorithm is obtained according to the expectation-maximization algorithm.

where n+1 is the number of iterations, j represents the central pixel number of the patch, and EM represents the expectation maximization.

其中，Re g表示正则化方法。 In particular, the S22 includes using a regularization method for smoothing, and obtaining

Among them, Reg represents the regularization method.

In particular, the fusion processing of S23 includes fusion of pixels one by one into an image, and the next iterative image estimation is obtained according to the fusion processing:

When β=0, the above equation reduces to the maximum likelihood expectation maximization formula.

A CT image reconstruction system, comprising: an initialization unit: used to obtain an initialization image according to projection data; an iterative unit: used to take the initialization image as an image to be reconstructed, and perform iterative processing on the image to be reconstructed; a judgment output unit: used to judge Whether the iterative stop condition is met, if so, output the iteratively processed image; if not, send the iteratively processed image to the iterative unit, and perform iterative processing again as the image to be reconstructed; the iterative unit includes updating a processing unit, a smoothing processing unit and a fusion processing unit; the update processing unit further includes a roughness calculation unit, the roughness calculation unit calculates the image roughness based on the patch.

Further, the update processing unit further includes: an optimization transmission unit, configured to construct a surrogate function in each iteration of the image to minimize the objective function Φ(μ).

A computer device comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, all The one or more processors implement the CT image reconstruction method described above.

A computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the CT image reconstruction method described above.

The present invention has the following beneficial effects:

Aiming at the technical problems of imaging quality and speed of low-dose CT images, the present invention proposes an image reconstruction method, system, equipment and medium, which solves the common problems in the prior art that the image reconstruction process is complex, the reconstructed image contains a lot of noise and artifacts, It can eliminate noise, suppress artifacts, retain edges and fine features, reduce the number of iterations while ensuring image quality, and has the advantages of short reconstruction process, high image quality, and high image quality. Features such as comprehensive retention are of great significance to the development of the medical impact field.

In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below, and are described in detail as follows in conjunction with the accompanying drawings.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore do not It should be regarded as a limitation of the scope, and for those of ordinary skill in the art, other related drawings can also be obtained according to these drawings without any creative effort.

Fig. 1 is the flow chart of the CT image reconstruction method of the present invention;

Fig. 2 is the concrete flow chart of the CT image reconstruction method of the present invention;

3 is a block diagram of a CT image reconstruction system of the present invention;

Fig. 4 is the concrete block diagram of the CT image reconstruction system of the present invention;

FIG. 5 is a schematic diagram of the application of the CT image reconstruction method of the present invention to a computer device.

detailed description

The process of CT image reconstruction includes: first, the detector records the projection data of the X-ray passing through the scanned object under different viewing angles; secondly, the projection data is constructed into the required CT initial image, and then the image iterative reconstruction processing is performed. Perform multiple iterative processing on the image; then, determine whether the reconstructed image after each iteration satisfies the iterative stop condition, if so, output the reconstructed image that satisfies the iterative stop condition, and if not, continue the image iterative reconstruction.

Definition: Assuming the presence of measurement errors and noise effects, X-ray CT measurements are expressed as follows:

y=Hx+e

Among them: y=(y ₁ , y ₂ ,...,y _M ) ^T represents the projection data after system calibration and logarithmic transformation; x=(x ₁ ,x ₂ ,...,x _N ) ^T represents the to-be-estimated Attenuation coefficient; H represents the image vector to be calculated, T represents the matrix transpose, and e represents the measurement error and noise. The purpose of CT image reconstruction is to estimate the attenuation coefficient x based on the measured value y of H.

The present invention mainly performs image reconstruction through the penalized likelihood method. The penalized likelihood method estimates the reconstructed image by maximizing the penalized likelihood function, and the specific expression is as follows:

Φ(μ)=L(y|μ)-βU(μ)

Among them, L(y|μ) is the likelihood function, U(μ) is the image roughness penalty term, Φ(μ) can be understood as an objective function, and β is the regularization parameter that controls data fidelity and spatial smoothness , μ is the estimated image value to be calculated.

A likelihood function is a function of parameters in a statistical model that represents the likelihood in the parameters of the model. Likelihood functions play an important role in statistical inference, such as maximizing likelihood (ML), which is to find the maximum value of the likelihood function, indicating that the corresponding parameters can make the statistical model the most reasonable. In practical applications Generally, the logarithm of the likelihood function is taken as the function for finding the maximum value. In the above expression, when β approaches 0, the reconstructed image approaches the maximum likelihood estimation.

The general image roughness is measured based on the intensity difference between two adjacent pixels, and the expression is:

where ψ(μ _p - μ _q ) is the penalty function, μ _p - μ _q is the distance between pixel p and adjacent pixel q, and ω _pq is the distance between pixel p and pixel q in the neighborhood Weighting factor, n _p is the total number of detectors and N _p is the total number of image pixels.

Example 1

In view of the deficiencies of the prior art, this embodiment provides a CT image reconstruction method. The specific scheme of this embodiment is as follows:

An image reconstruction method, the method comprising: S1, obtaining an initialization image according to projection data; S2, taking the initialization image as the image to be reconstructed, and performing image iterative processing on the image to be reconstructed; S3, judging whether the iteration stop condition is met, if so, then Stop the iteration, and output the iteratively processed image; if not, return to S2, and use the iteratively processed image as the image to be reconstructed for iterative processing again. The specific flow chart is shown in Figure 1 of the description. Wherein, in S2, the image iterative processing includes performing: S21, updating processing on the image, including solving the image roughness by calculating the intensity difference of adjacent patches, and then obtaining the maximum penalty likelihood function; S22, smoothing processing, on the The image in S21 is subjected to smoothing and denoising processing; and S23, fusion processing, performs regional fusion processing on the image from S22, including pixel-by-pixel fusion into an image. The specific process is shown in Figure 2 of the description.

Iteratively process the image to be reconstructed. The image to be reconstructed includes an initialization image and an iteratively processed image that does not satisfy the iterative stop condition, that is, the initialized image of S1 and an image that has been iteratively processed in S3 but does not satisfy the iteration stop condition. Specifically, image update processing is performed on the image. The image update processing part uses the penalized likelihood function to estimate the image, and more specifically, estimates the reconstructed image by maximizing the penalized likelihood function, and the specific expression is as follows:

Φ(μ)=L(y|μ)-βU(μ)

Among them, L(y|μ) is the likelihood function, U(μ) is the image roughness penalty term, Φ(μ) can be understood as an objective function, and β is the regularization parameter that controls data fidelity and spatial smoothness , μ is the estimated image value. The maximum penalty likelihood function is obtained by solving the image roughness, and the image roughness expression is:

where ψ(μ _p - μ _q ) is the penalty function, μ _p - μ _q is the distance between pixel p and adjacent pixel q, and ω _pq is the distance between pixel p and pixel q in the neighborhood weight factor.

Further, the S21 image update process also includes calculating the image roughness based on the patch. A patch includes a pixel block formed by a plurality of adjacent pixels, and the shape of the pixel block can be regular, such as a square, or irregular. Preferably, in this embodiment, a regular pixel block is selected, specifically, a patch with one pixel as the center and other pixels surrounding it is selected. In 2D modeling, patches include 3×3, 5×5, etc. patches, and can also be a single pixel. The patch may include complete pixels or incomplete pixels, that is, a square area formed by cutting pixels. Preferably, complete pixels are selected in this embodiment, that is, only complete pixels are included in the patch. There are 9 complete pixels in the patch, centered on one pixel, and the other 8 pixels around the center pixel. Patches of different sizes affect the efficiency of image processing. Compared with a single pixel, a patch has a larger area, and more texture features on the pixel are preserved, and it will not cause too many fine features to be eliminated due to smoothing. Preferably, the patches in the images are the same size. According to the requirements of the penalized likelihood method, pixel j and pixel k are selected to calculate the image roughness. Pixel j and pixel k are the center pixels of two adjacent patches respectively, and the image roughness is calculated by calculating the intensity difference between the two patches. degree, and then obtain the maximum penalty likelihood function. The distance expression between pixel j and adjacent pixel k is:

Where: j _l represents the lth pixel in the patch centered on pixel j, k _l represents the lth pixel in the patch centered on pixel k, n _j identifies the total number of detectors, and N _j represents the patch The number of all pixels in , h _l is a positive weighting factor of the normalized inverse spatial distance between pixels:

According to the calculation expression of image roughness, the image roughness based on patch is obtained as:

Among them, f _j (μ) is the feature vector composed of all pixel intensity values of the patch centered at the pixel j, and f _k (μ) is composed of all the pixel intensity values of the patch centered at the pixel k , ψ(||f _j (μ)-f _k (μ)‖ _h ) is the penalty function. ψ(||f _j (μ)-f _k (μ)‖ _h ) penalty function includes quadratic function and hyperbolic function. Preferably, the penalty function expression selected in this embodiment is:

When |||f _j (μ)-f _k (μ)‖ _h |<<δ, the penalty function approximates a quadratic function;

When |||f _j (μ)-f _k (μ)‖ _h |>>δ, the penalty function approximates an absolute function.

Among them, δ is the penalty factor, the penalty function is non-differentiable at zero and has a continuous second-order derivative, which can ensure that the edge of the reconstructed image is preserved and a clear edge is formed. The patch-based regularization method calculates the image roughness by comparing the similarity between patches, which has stronger edge retention ability, reduces the number of iterations, shortens the image reconstruction time, and preserves the fine texture features of the image. . Compared with traditional single-pixel-based priors, it has better robustness in distinguishing edges, and can also reduce noise in reconstructed images while retaining more detailed information.

Further, the S21 image update process also includes optimizing the transmission algorithm, and the basic idea of optimizing the transmission algorithm includes constructing a surrogate function at each iteration, that is, constructing the surrogate function Q(μ; μ ⁿ ) of the image μ at the nth image iteration , the objective function Φ(μ) is minimized by satisfying the following two conditions:

Q(μ; μ ⁿ )-Q(μ ⁿ ; μ ⁿ )≤Φ(μ)-Φ(μ ⁿ )

表示相对于μ的梯度，目标函数Φ(μ)＝L(y|μ)-βU(μ)； in,

Represents the gradient relative to μ, the objective function Φ(μ)=L(y|μ)-βU(μ);

Convert the maximum value of Φ(μ) to the maximum value of Q(μ; μ ⁿ ) and bring it into the maximum penalty likelihood function to get:

Φ(μ) increases monotonically with updated μ ⁿ⁺¹ : Φ(μ ⁿ⁺¹ )≥Φ(μ ⁿ ).

Each image iteration is updated as follows:

其中n+1为迭代次数，j表示所述面片的中心像素序号，EM表示期望最大化。 The well-known expectation maximization (EM) algorithm is a special case of the optimal transmission algorithm. Based on the existing research, the present invention selects the expectation maximization (EM) algorithm to perform image update processing, and obtains the estimated image value based on the expectation maximization processing, that is,

where n+1 is the number of iterations, j represents the central pixel number of the patch, and EM represents the expectation maximization.

其中n为迭代次数，j表示所述面片的中心像素序号，Re g表示正则化方法。 Specifically, in S22, image smoothing processing is performed on the image from S21. Specifically, the regularization method is used to smooth the image, a regular term is added to the likelihood function, and the weight of the parameters is controlled by the size of the threshold (or the regular term coefficient), so that the solution of the problem is in the model. There is a trade-off between fitting accuracy and sparsity metric. The core purpose of the regularization method is to limit the size of the parameter space to reduce the complexity of the model, and to limit the value range of the solution by adding a regular term to ensure good properties of the solution, such as uniqueness. In the smoothing process of S22, the estimated image value is obtained based on the regularization method, namely

where n is the number of iterations, j represents the central pixel number of the patch, and Reg represents the regularization method.

和S22的平滑处理中得到的

得到下一次迭代图像估计，预估图像值表达式为： Specifically, in S23, image fusion processing is performed on the image from S22. Specifically, pixel-by-pixel image fusion is performed on the updated and smoothed image, and the pixels in the image are fused to form a patch, such as a patch with a 3×3 and 5×5 structure. Obtained according to the update process of S21

and obtained in the smoothing process of S22

The next iteration image estimation is obtained, and the estimated image value expression is:

Among them, β is a regularization parameter that controls data fidelity and spatial smoothness. When β=0, the above update equation is simplified to the maximum likelihood-expectation maximization formula. The maximum likelihood-expectation maximization method degrades image quality with increasing number of iterations, producing "checkerboard artifacts". Terminating the iteration early and integrating a penalty term or some kind of prior knowledge in the likelihood function overcomes the problems of the maximum likelihood-expectation maximization method to a certain extent. When the iterations do not converge, the present invention requires fewer iterations to achieve the same effect of reconstructed images, thereby avoiding the degradation of image quality. Through repeated image iterative reconstruction, the projection data is converted into an image, the number of iterations is small, the image quality is high, and the iteration efficiency is high.

Example 2

In view of the shortcomings of the prior art, a CT image reconstruction system is proposed in this embodiment, and the step modules in the CT image reconstruction method of Embodiment 1 are embodied to form a system. The system block diagram is shown in FIG. 3 of the specification. , including:

Initialization unit: used to obtain the initialization image according to the projection data;

Iterative unit: used to take the initialization image as the image to be reconstructed, and perform image iterative processing on the image to be reconstructed;

Judgment output unit: used to judge whether the iteration stop condition is met, if so, output the iteratively processed image; if not, send the iteratively processed image to the iterative unit, and perform iterative processing again.

The iterative unit includes an update processing unit, a smoothing processing unit and a fusion processing unit; the image passes through the update processing unit, the smoothing processing unit and the fusion processing unit in sequence; the update processing unit is used to solve the image by calculating the intensity difference of adjacent patches roughness, and then obtain the maximum penalty likelihood function; smoothing processing unit, used for smoothing and denoising processing of the image from the updating processing unit; fusion processing unit, performing regional fusion processing on the image from the smoothing processing unit, including pixel-by-pixel processing merged into an image.

Wherein, the update processing unit further includes a roughness calculation unit, and the roughness calculation unit is configured to store an algorithm for calculating the image roughness based on the patch, and perform the image roughness calculation based on the patch on the image.

The initialization unit establishes an image model according to the acquired projection data of the projection object, forms an initialization image, and sends it to the iterative unit; the update processing unit of the iterative unit receives the initialization image sent by the initialization unit, and takes the initialization image as the image to be reconstructed. The image is subjected to image update processing, and sent to the smoothing processing unit after the update processing; the smoothing processing unit receives the image sent by the update processing unit, performs smoothing processing on the image based on the regularization method, and sends the processed image to the fusion processing unit; fusion; The processing unit receives the image sent by the smoothing processing unit, performs regional fusion processing on the image, including pixel-by-pixel fusion into an image, and sends the processed image to the judgment output unit to complete a round of iterative processing; the judgment output unit receives the image sent by the iterative unit. image, to determine whether the iterative stop condition is met, if so, output the iteratively processed image, if not, return the iteratively processed image to the iterative unit for a new round of iterative processing. The specific system block diagram is shown in Figure 4 of the description.

Wherein, the update processing unit further includes an optimization transmission unit, which is provided with an optimization transmission algorithm for constructing a surrogate function in each iteration of the image to minimize the objective function Φ(μ).

In order to solve the technical problems of low-dose CT image imaging quality and speed, the present invention proposes a CT image reconstruction system based on existing research, which can take into account the image quality and image generation efficiency under the premise of ensuring safety, and can be more efficient Faster speeds output (eg display) higher quality images. The image is iteratively reconstructed by an iterative unit, and the image passes through an update processing unit, a smoothing processing unit and a fusion processing unit in turn, and finally an output unit judges whether the iterative stop condition is met. In this embodiment, the optimized transmission algorithm is used to optimize the penalized likelihood reconstruction, which solves the artifacts and noise caused by low dose, avoids the degradation of image quality, and improves the image quality; the image roughness calculation method based on the patch is adopted. The image roughness is calculated to ensure the image quality, while shortening the reconstruction time, retaining the fine features and edge features of the image, and has better robustness in distinguishing edges than the traditional single pixel-based prior.

Example 3

FIG. 5 is a schematic structural diagram of a computer device according to Embodiment 3 of the present invention. The computer device 12 shown in FIG. 5 is only an example, and should not impose any limitation on the function and scope of use of the embodiments of the present invention.

As shown in FIG. 5, computer device 12 takes the form of a general-purpose computing device. Components of computer device 12 may include, but are not limited to, one or more processors or processing units 16 , system memory 28 , and a bus 18 connecting various system components including system memory 28 and processing unit 16 . Computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by device computer 12, including both volatile and nonvolatile media, removable and non-removable media. System memory 28 may include computer system readable media in the form of volatile memory.

Computer device 12 may also communicate with one or more external devices 14 (eg, keyboards, pointing devices, displays, etc.), may also communicate with one or more devices that enable a user to interact with computer device 12, and/or communicate with The computer device 12 is capable of communicating with any device that communicates with one or more other computing devices.

The processing unit 16 executes various functional applications and data processing by running the programs stored in the system memory 28, for example, implementing a CT image reconstruction method provided in Embodiment 1 of the present invention, and the method includes:

S1, obtaining an initialization image according to the projection data; S2, taking the initialization image as the image to be reconstructed, and performing image iterative processing on the image to be reconstructed; S3, judging whether the iteration stop condition is met, if so, stopping the iteration, and outputting the image after iterative processing; If not, return to S2, and use the iteratively processed image as the image to be reconstructed to perform iterative processing again. Among them, the image iterative processing in S2 includes:

S21, update processing, including obtaining the maximum penalty likelihood function by calculating the roughness of the image based on the patch; S22, smoothing processing, performing smoothing and denoising processing on the image from S21; S23, fusion processing, performing regional processing on the image from S22 Fusion processing, including pixel-by-pixel fusion into an image.

In this embodiment, a CT image reconstruction method is applied to a specific computer device, and the method is stored in the memory. When the actuator executes the memory, the method will be run to reconstruct the CT image, which is fast and convenient to use and has a wide range of applications. .

Of course, those skilled in the art can understand that the processor can also implement the technical solution of the CT image reconstruction method provided by any embodiment of the present invention.

Example 4

This embodiment 4 provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of the CT image reconstruction method provided by any embodiment of the present invention, and the method includes:

S1, obtaining an initialization image according to the projection data; S2, taking the initialization image as the image to be reconstructed, and performing image iterative processing on the image to be reconstructed; S3, judging whether the iteration stop condition is met, if so, stopping the iteration, and outputting the image after iterative processing; If not, return to S2, and use the iteratively processed image as the image to be reconstructed to perform iterative processing again. Among them, the image iterative processing in S2 includes:

S21, update processing, including obtaining a maximum penalty likelihood function by calculating the image roughness based on the patch; S22, smoothing processing, performing smoothing and denoising processing on the image from S21; S23, fusion processing, performing on the image from S22 Region fusion processing, including pixel-by-pixel fusion into an image.

The computer storage medium in the embodiments of the present invention may adopt any combination of one or more computer-readable mediums. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination of the above. More specific examples (a non-exhaustive list) of computer readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), Erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

This embodiment applies a CT image reconstruction method to a computer-readable storage medium, which stores a computer program, and when the program is executed by a processor, implements the steps of the CT image reconstruction method provided by the present invention, which is simple, fast, and easy to store , not easy to lose.

To sum up, the present invention proposes an image reconstruction method, system, device and medium, which solves the common problems in the prior art that the image reconstruction process is complex, there are a lot of noise and artifacts in the reconstructed image, the edge retention ability is poor, and the original features in the image are solved. Defects such as texture not being preserved can eliminate noise, suppress artifacts, preserve edges and fine features, reduce the number of iterations while ensuring image quality, and have the characteristics of short reconstruction process, high image quality, and comprehensive image feature retention. Applying the CT reconstruction method to specific systems, computer equipment and computer storage media, and embodying the method is of great significance to the development of the medical impact field.

Those of ordinary skill in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device, and they can be centralized on a single computing device, or distributed on a network composed of multiple computing devices. Optionally, they may be implemented in program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be fabricated separately into individual integrated circuit modules, or a plurality of modules of them Or the steps are made into a single integrated circuit module to realize. As such, the present invention is not limited to any specific combination of hardware and software.

Note that the above are only preferred embodiments of the present invention and applied technical principles. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and can also include more other equivalent embodiments without departing from the concept of the present invention. The scope is determined by the scope of the appended claims.

The above disclosures are only a few specific implementation scenarios of the present invention, however, the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims (13)

A CT image reconstruction method, comprising the following steps: S1. Obtain an initialization image according to the projection data; S2, take the initialization image as the image to be reconstructed, and perform image iterative processing on the image to be reconstructed; S3, judge whether the iteration stop condition is met, if so, stop the iteration, and output the iteratively processed image; if not, return to S2, and use the iteratively processed image as the image to be reconstructed for iterative processing again; It is characterized in that, In the S2, the image iterative processing further includes the following steps: S21, update processing, including solving the image roughness by calculating the intensity difference of adjacent patches, and then obtaining the maximum penalty likelihood function; S22, smoothing, including performing smoothing and denoising on the image from S21; S23. Fusion processing, including performing regional fusion processing on the image from S22. The method according to claim 1, wherein, in the update process of S21, the patch used for calculating the image roughness comprises a square pixel area composed of multiple pixels and centered on one pixel , all the patches used to calculate the image roughness are the same size. The method according to claim 1, wherein, in the update process of S21, the expression of the maximum penalty likelihood function is:

where μ is the estimated image value, Φ(μ) is the objective function, L(y|μ) is the likelihood function, β is the regularization parameter that controls data fidelity and spatial smoothness, and U(μ) is the image Roughness, the maximum penalty likelihood function is obtained by solving the image roughness, and the image roughness expression is:

where ψ(μ _p - μ _q ) is the penalty function, μ _p - μ _q is the distance between pixel p and adjacent pixel q, and ω _pq is the distance between pixel p and pixel q in the neighborhood Weighting factor, n _p is the total number of detectors and N _p is the total number of image pixels. The method according to claim 3, wherein, in the update process of S21, the image roughness based on the patch is calculated by the following method; Pixel j and pixel k are respectively the center pixels of two adjacent patches, and the image roughness is calculated by calculating the intensity difference between the two patches, and then the maximum penalty likelihood function is obtained, and the difference between the pixel j and the pixel k is The pixel distance expression is:

The image roughness expressions of the pixel j and the pixel k are:

where: f _j (μ) is the feature vector composed of all pixel intensity values of the patch centered at the pixel j, f _k (μ) is composed of all the pixel intensity values of the patch centered at the pixel k , j _l denotes the lth pixel in the patch centered on pixel j, k _l denotes the lth pixel in the patch centered on pixel k, n _j denotes the total number of detectors, N _j denotes The number of all pixels in the patch, h _l is the positive weighting factor of the normalized inverse spatial distance between pixels, ψ(||f _j (μ)-f _k (μ)‖ _h ) is the penalty function. The method according to claim 4, wherein the penalty function expression is:

Among them, δ is the attenuation factor; When |||f _j (μ)-f _k (μ)‖ _h |<<δ, the penalty function approximates a quadratic function; When |||f _j (μ)-f _k (μ)‖ _h |<<δ, the penalty function approximates an absolute function. The method according to any one of claims 1-5, characterized in that, in the update process of S21, an optimized transmission algorithm is further included; The optimized transmission algorithm includes constructing a surrogate function at each iteration to minimize the objective function Φ(μ). The method according to claim 6, wherein the optimized transmission algorithm comprises an expectation maximization algorithm, and the data obtained according to the expectation maximization algorithm

where n+1 is the number of iterations, j represents the central pixel number of the patch, and EM represents the expectation maximization. The method according to claim 7, characterized in that in said S22, the image from S21 is smoothed and denoised using a regularization method, and all items involving the patch μj centered on the pixel _j are combined, obtained according to the regularization method

where Reg represents the regularization method. The method according to claim 8, characterized in that, in the fusion process of S23, the image from S22 is fused pixel by pixel into an image, and the estimation at the n+1th iteration is obtained after pixel-by-pixel fusion. Image value:

When β=0, the above equation reduces to the maximum likelihood expectation maximization formula. A CT image reconstruction system, comprising: Initialization unit: used to obtain the initialization image according to the projection data; Iterative unit: used for taking the initialization image as the image to be reconstructed, and performing iterative processing on the image to be reconstructed; Judgment output unit: used to judge whether the iterative stop condition is met, if so, output the image after iterative processing; if not, send the iteratively processed image to the iterative unit, and perform iterative processing again as the image to be reconstructed ; It is characterized in that, The iteration unit includes an update processing unit, a smoothing processing unit and a fusion processing unit; The update processing unit further includes a roughness calculation unit that calculates image roughness based on the patch. The system according to claim 10, wherein the update processing unit further comprises: The optimized transmission unit is used to construct a surrogate function at each iteration of the image to minimize the objective function Φ(μ). A computer device, characterized in that the computer device comprises: one or more processors; memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the CT image reconstruction method according to any one of claims 1-9. A computer-readable storage medium on which a computer program is stored, characterized in that, when the program is executed by a processor, the CT image reconstruction method according to any one of claims 1-9 is implemented.

Images