Low-dose ct chord diagram restoration method and system based on multi-gaussian electronic noise modeling

Abstract

The application discloses a low-dose CT chord diagram recovery method and system based on multi-Gaussian electronic noise modeling, and the method comprises the following steps: according to a CT projection noise generation model, a low-dose CT chord diagram recovery model based on multi-Gaussian noise modeling is established, and an alternating projection method is used to decompose the chord diagram recovery model into a sub-problem about a plurality of to-be-estimated parameters; multi-Gaussian parameters are updated and solved according to an EM algorithm, the number of photons is iteratively solved and updated, and chord diagram data is iteratively solved according to a proximal operator, and the proximal operator in the chord diagram data solving is calculated by using a Resnet network model. The application breaks through the limitation of traditional technologies in processing electronic noise distribution, and can accurately estimate more complex electronic noise distribution. Through the alternating projection algorithm, the application not only has excellent ability in chord diagram recovery, but also realizes significant performance improvement in structural similarity and visual consistency.

Description

Technical Field

[0001] This invention belongs to the field of low-dose CT projection image restoration technology, specifically relating to a low-dose CT chordal image restoration method that introduces multi-Gaussian electronic noise. Background Technology

[0002] Computed tomography (CT) has become one of the most common diagnostic and detection methods in modern medical imaging. However, the widespread use of CT has raised concerns about the potential for X-rays to cause cancer or genetic damage, leading to widespread interest in low-dose CT imaging algorithms that reduce X-ray radiation during scans. The main approach to reducing radiation dose is to lower the X-ray tube current or voltage, thereby reducing the X-ray hazards to patients. However, lowering the tube current and voltage can increase photon and electronic noise, making the background noise in chordogram data (chordogram) more significant, severely impacting image quality and diagnostic accuracy.

[0003] Chordogram data refers to CT scan data before logarithmic transformation. It is the foundation of all CT imaging algorithms, and its quality sensitively affects the image quality of FBP (Filtered Back Projection) imaging algorithms or other imaging algorithms. Therefore, a good chordogram data recovery method can very effectively enable other CT imaging algorithms to recover high-quality CT images while reducing the radiation dose of CT scans.

[0004] Current chordogram data recovery methods model photon noise and electronic noise using Poisson and single Gaussian distributions, respectively, before calculating the chordogram data. Chinese patent CN16312983A proposes a hidden regularized low-dose CT image reconstruction method based on noise generation mechanisms, solving the problem of high-quality CT image recovery under ultra-low-dose conditions. However, it heavily relies on the assumption that the system's inherent electronic noise follows a single Gaussian distribution. CT imaging literature research shows that a single Gaussian distribution model cannot accurately characterize electronic noise. Theoretical studies have demonstrated that a multi-Gaussian mixture distribution can approximate any probability distribution. Summary of the Invention

[0005] To better recover chordogram data, this invention employs a low-dose CT chordogram recovery method based on multi-Gaussian electronic noise modeling, which can more accurately characterize the true distribution characteristics of electronic noise in the CT system and quickly and effectively remove noise from the chordogram data.

[0006] To achieve the above objectives, the technical solution adopted in this invention is: a low-dose CT chordal reconstruction method based on multi-Gaussian electronic noise modeling. First, an optimization model is established that introduces electronic noise as a multi-Gaussian electronic distribution. Then, an alternating projection iterative algorithm is designed to solve the model to obtain chordal data that is closer to the real one.

[0007] Specifically, the following steps are included:

[0008] A low-dose CT chordogram data recovery model that incorporates Gaussian electronic noise is input into low-dose CT chordogram data;

[0009] An alternating projection iterative algorithm is used to solve the low-dose CT chordal data restoration model with Gaussian electron noise. The solution of the low-dose CT chordal data restoration model with Gaussian electron noise is decomposed into solving the problem of photon number T and solving the chordal data Y. The proximal gradient descent method is used to solve Y to obtain the restored chordal data Y.

[0010] A CT imaging algorithm is used to convert the chordal data Y into interpretable CT images.

[0011] Furthermore, the low-dose CT chordogram data recovery model incorporating Gaussian electronic noise is as follows:

[0012]

[0013] in The value represents the number of photons received by the detector, where N is the total number of X-ray incident paths in the scan, and i represents the i-th incident X-ray path, i = 1, 2, ..., N. K represents the number of multi-Gaussian beams, k = 1, 2, ..., K, σ k μ k Let π represent the variance and mean of the k-th Gaussian distribution, respectively. k I represents the weight of the k-th Gaussian distribution. 0i Let T be the intensity of the i-th incident X-ray. i Let Y be the number of photons received by the detector when incident along the projection path i, and let Y be the chord diagram data. i Let g1(Y) be the chord map data on the detector when incident along the projection path i, g1(Y) be the regularization function with respect to Y, and λ1 be the penalty coefficient of the regularization function.

[0014] It should be noted that, for the sake of algorithm design, we will temporarily assume that K multi-Gaussian parameters are given.

[0015] Furthermore, training the low-dose CT chordogram data recovery model with introduced Gaussian electronic noise includes the following steps:

[0016] Several conventional dose CT images are acquired, and the conventional dose CT images are converted into chordal data through inverse Lardon transform. A mixed distribution composed of K Gaussian distributions is superimposed in the projection domain to simulate a more realistic electronic noise distribution, forming low dose projection domain data.

[0017] The alternating projection optimization algorithm is used to decompose the chord graph data recovery model into subproblems with respect to several parameters to be estimated. The EM algorithm and the proximal gradient descent algorithm are used to iteratively update each parameter. The convergence of the algorithm yields the chord graph data Y and the chord graph data recovery model.

[0018] Normal-dose CT images are transformed into chordal data using inverse Lardon transform. Then, mixed multi-Gaussian noise is introduced into the chordal data to simulate the data characteristics under real low-dose CT imaging conditions. The noise addition step in the chordal data considers not only the influence of photon noise but also the diversity of electronic noise.

[0019] Furthermore, the alternating projection optimization algorithm is used to decompose the chordogram data recovery model into sub-problems concerning several parameters to be estimated. The EM algorithm and the proximal gradient descent algorithm are then used to iteratively update each parameter, including:

[0020] The alternating projection iterative algorithm is used to solve model (1).

[0021] S3.1, Fixed multi-Gaussian parameter σ k μ k Using the weighting coefficients of the k-th Gaussian distribution and the chord graph data Y, a sub-optimization model is established to solve for the photon number T. i By separating the variables, the original problem is transformed into a single-variable optimization problem. i Optimization issues:

[0022]

[0023] The objective function is updated using Newton's method. The first and second derivatives f′(T) are calculated. i ), f″(T i )

[0024]

[0025]

[0026]

[0027]

[0028] Here, R is the floor function.

[0029] S3.2. Update the chord graph data Y using the proximal gradient method. With the chord graph data Y fixed, solve for Y using the proximal gradient descent method. The update iterative formula for Y is as follows:

[0030]

[0031] Among them, the proximal operator The Residual Network (ResNet) can be used to learn the proximal operator determined by the regularization term g1(·) using the ResNet model.

[0032] Step 3.3: Calculate the multi-Gaussian parameters Π,Λ,Σ. Using the obtained photon number T and the original data S, and with the help of the residual term data, i.e., the difference between the low-dose chordogram data S and the photon number T received by the detector, the EM algorithm is used to estimate the multi-Gaussian distribution parameters. The EM algorithm iteration process is set to 10 rounds, and the formula for iteratively updating the multi-Gaussian parameters Π,Λ,Σ is as follows:

[0033]

[0034]

[0035]

[0036]

[0037] After the iteration is complete, the estimated multi-Gaussian parameters will be treated as fixed values ​​and used in subsequent algorithm processing.

[0038] The iteration stops after a fixed number of iterations in T and Y; otherwise, the obtained multi-Gaussian parameters Π,Λ,Σ, T, and Y are substituted into S3.1 to continue execution.

[0039] Furthermore, the residual network consists of four residual blocks, each composed of two convolutional layers, batch normalization, and the ReLU activation function. Each convolutional layer has a 3×3 filter size, and the number of channels varies depending on the dimension of the input data. After each convolution, the output is normalized, a non-linear transformation is performed using the ReLU activation function, and residual connections are used to directly pass the input data to the output layer, thereby alleviating the gradient vanishing problem in deep networks. Simultaneously, to ensure that the update of Y contains richer feature information, historical update values ​​of Y are... plus The concatenation is then applied to the current Y, and the concatenated result is used as input to the residual network for further processing.

[0040] Furthermore, the chordal data Y is converted into a viewable CT image using a CT imaging algorithm, which includes: fixing the proximal operator learned in S3.2, substituting the scanned low-dose chordal data S, incident light intensity I0, and multi-Gaussian parameters Π, Λ, Σ into the algorithm in step 3, converting the recovered chordal data Y into the final CT image using the FBP algorithm, calculating the structural similarity (SSIM) and peak signal-to-noise ratio (PSNR) compared to the CT image with conventional dose, and evaluating the chordal data recovery results.

[0041] Based on the concept of the method, the present invention also provides a low-dose CT chordogram recovery system based on multi-Gaussian electronic noise modeling, comprising a data acquisition module, a chordogram data recovery module, and an image conversion module;

[0042] The data acquisition module is used to acquire low-dose CT chordogram data and input the low-dose CT chordogram data into a low-dose CT chordogram data recovery model that incorporates Gaussian electronic noise;

[0043] The string diagram data recovery module uses an alternating projection iterative algorithm to solve the low-dose CT string diagram data recovery model with Gaussian electron noise. The solution of the low-dose CT string diagram data recovery model with Gaussian electron noise is decomposed into solving the problem of solving the photon number T and solving the string diagram data Y. The proximal gradient descent method is used to solve Y to obtain the recovered string diagram data Y.

[0044] The image conversion module uses a CT imaging algorithm to convert the chordal data Y into viewable CT images.

[0045] The present invention can also provide a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and the processor can realize the above-mentioned low-dose CT chordal recovery method based on multi-Gaussian electronic noise modeling when executing the executable program.

[0046] Simultaneously, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, it can realize the above-mentioned low-dose CT chordal recovery method based on multi-Gaussian electronic noise modeling.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] This invention introduces a more realistic multi-Gaussian distribution to the CT projection noise generation mechanism, which can more accurately capture the real distribution characteristics of electronic noise in the CT system. The constructed noise model is more in line with the actual physical laws of CT images, and breaks through the limitation of representing electronic noise with a single Gaussian distribution in the chordal plot data recovery technology.

[0049] This invention proposes a low-dose CT chordal reconstruction method based on multi-Gaussian electronic noise. It cleverly introduces the expectation-maximization algorithm to update the multi-Gaussian distribution parameters, overcoming the dilemma that the alternating projection algorithm is difficult to solve when the multi-Gaussian parameters are optimized as unknown variables in the establishment of optimization models.

[0050] This invention proposes a low-dose CT image recovery method that can be applied to the study of simulation data when it is not possible to perform low-dose scanning directly on patients; it lays the foundation for future research on low-dose CT imaging and algorithms by directly scanning patients. Attached Figure Description

[0051] Figure 1 This is represented as a framework diagram for a low-dose CT chordal recovery method based on multi-Gaussian electronic noise modeling.

[0052] Figure 2 The diagram is represented as a chord diagram at a conventional dose level (200 mAs, 120 kVp).

[0053] Figure 3 This is represented as the chord diagram recovered by the algorithm of this invention.

[0054] Figure 4 The chord diagram is represented as a low-dose level (20 mAs, 120 kVp).

[0055] Figure 5 The image is represented as an image reconstructed using FBP after chordal restoration performed by the present invention with a scan of (200 mAs, 120 kVp).

[0056] Figure 6 FBP reconstructed image at a conventional dose level (200 mAs, 120 kVp).

[0057] Figure 7 The image is represented as a low-dose (20 mAs) FBP reconstructed image. Detailed Implementation

[0058] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0059] Example 1: The low-dose CT chordogram data recovery method with the introduction of multi-Gaussian electronic noise provided by the present invention includes the following steps:

[0060] (1) Design a low-dose CT chordal reconstruction algorithm that incorporates multi-Gaussian electronic noise.

[0061] The generation model for CT projection data S can be represented as:

[0062] S=T+ε (1)

[0063] in, This indicates the number of photons received by the detector. This represents the electronic background noise, where N is the total number of X-ray incident paths in the scan.

[0064] To more accurately describe the electronic noise of a CT system, the electronic noise of the i-th incident path can be expressed as a mixture of Gaussian distributions:

[0065]

[0066] Where, ε i Let be the electron noise of the i-th incident path, i = 1, ..., N; k = 1, ..., K, where K represents the number of Gaussian distributions, π k μ represents the weight coefficient of the k-th Gaussian distribution. k ,σ k Let Π = {π1,…,π} represent the mean and variance of the k-th Gaussian distribution, respectively. k}, Λ={μ1,…,μ k},∑={σ1,…,σ k}, its k-th Gaussian component is:

[0067]

[0068] The number of photons T received by a CT detector is a result of X-ray photon fluctuations. Related research results indicate that the number of photons obtained from CT scans approximately follows a Poisson distribution.

[0069]

[0070] G i This represents the average number of photons received under ideal conditions.

[0071] Let Y represent noise-free "chord graph data". According to the Lambert-Beer theorem, Y satisfies:

[0072]

[0073] Combining (4) and (5), we can obtain:

[0074]

[0075] Combining (1) and (2), we can obtain the following conditional joint distribution:

[0076]

[0077] In summary, the posterior conditional distributions of S and T with respect to Y are:

[0078]

[0079] According to Bayes' conditional probability formula, the complete posterior distributions of the parameters T and Y to be estimated can be obtained:

[0080]

[0081] In equation (9), since S is the measured data, p(S) is a constant. Thus, the maximum a posteriori estimation model for the parameters T and Y to be estimated is obtained:

[0082]

[0083] Taking the negative logarithm of the objective function (10), we can transform it into the following equivalent optimization problem:

[0084]

[0085] Due to the convexity of the logarithmic function, according to Jensen's inequality, finding the minimum of the objective function is transformed into finding the minimum of its upper bound:

[0086]

[0087] Furthermore, a chord graph implicit regularization term g1(Y) is added to the above model to construct the final CT chord graph data reconstruction model:

[0088]

[0089] (2) Update the parameters in the CT chord diagram recovery model.

[0090] This invention employs the alternating projection method to decompose the parameter problem in the chord diagram reconstruction model into subproblems concerning the parameters T and Y to be estimated. Then, the EM algorithm is used to update the temporarily assumed multi-Gaussian parameters Π, Λ, and Σ. The specific calculation process is as follows:

[0091] Update T

[0092] Solve the following subproblem concerning T:

[0093]

[0094] Using T i Given the property of variable separation, the above N-dimensional optimization problem can essentially be transformed into N single-variable optimization problems:

[0095]

[0096] For the above N univariate optimization problems, let the integer variable T i Since relaxation is a continuous variable, equation (17) can be generalized to the following problem:

[0097]

[0098] Solve for the first and second derivatives of equation (18) respectively:

[0099]

[0100] Among them, Γ(T) i +1) is the generalized factorial, used to substitute T in continuous space. i ! . Because f″(T i If T > 0, then in a continuous space, integer programming is a strictly convex function, meaning that any local optimum is also a global optimum. Therefore, Newton's method can be applied to update T.

[0101]

[0102] Where the superscript n is the iteration index, η1 is the step size, and rounding is used to approximate T in the last iteration. i The true integer value is obtained, thus yielding an approximate integer solution to the global optimal solution.

[0103] T i n =R(T) i n ) (twenty one)

[0104] Here, R is the floor function.

[0105] Update Y

[0106] Once the values ​​of other parameters are determined, updating parameter Y is equivalent to solving the following subproblem:

[0107]

[0108] The first term in the above equation is given at point Y( k If we approximate the problem using a second-order Taylor expansion, then the second-order approximation of the subproblem is:

[0109]

[0110] Among them, Y (n-1) It is the update result obtained after the (n-1)th iteration of the chord graph data, where η2 is the step size. Solve for (22), Y (n) The update can be represented using the proximal operator as follows:

[0111]

[0112] in It is a proximal operator determined by the regularization term g1(·). The calculation formula is as follows:

[0113]

[0114] Combining equations (23) and (24), Y (n) The complete update iteration formula is as follows:

[0115]

[0116] For equation (25), the proximal operator Residual networks (ResNet) can be used for learning.

[0117] The update of Y utilizes the "identity shortcut link" in deep residual networks, employing ResNet to expand the proximal operator. This allows the residual network to automatically learn prior information about the image from the training data and progressively optimize the objective function. The residual network consists of four residual blocks, each composed of two convolutional layers, batch normalization, and the ReLU activation function. Each convolutional layer has a 3×3 filter size, with the number of channels varying according to the dimension of the input data. After each convolution, the output is normalized, non-linearly transformed using the ReLU activation function, and the residual connections directly pass the input data to the output layer, thus mitigating the vanishing gradient problem in deep networks.

[0118] Calculate Π,Λ,Σ:

[0119] Assuming latent variables Let represent the probability that the latent class of each sample is a Gaussian distribution of the kth order, and its posterior expectation is as follows:

[0120]

[0121] Update and iterate the parameters Π, Λ, and ∑:

[0122]

[0123]

[0124]

[0125] in, It is the sum of probabilities belonging to the k-th Gaussian distribution. Repeat the E-step and M-step until the maximum number of iterations is reached or T and Y converge. The final parameters Π,Λ,Σ, which are the estimates of the multi-Gaussian model in this round, are introduced as constants in the update of the remaining parameters.

[0126] The aforementioned multi-Gaussian update module effectively suppresses noise and artifacts in chordogram data while preserving the detailed texture of the reconstructed image. Compared to deep networks that assume electronic noise follows a single Gaussian noise model, this method demonstrates better performance in terms of chordogram structural similarity and visual consistency, and more effectively suppresses noise in chordogram data.

[0127] Example

[0128] S1 converts CT images under conventional doses into chordal data through inverse Radon transform. Electronic noise following a multi-Gaussian distribution is introduced into the chordal domain to simulate the data characteristics under real low-dose CT imaging conditions.

[0129] S2, collect chest CT chordal images generated using low-dose technology and paired chest CT chordal images generated using normal dose; 700 simulated low-dose CT chordal images and 700 corresponding normal-dose CT chordal images are collected, with each image size being 736×1152, forming a total of 700 paired CT chordal image datasets.

[0130] S3, the paired CT chord image dataset is randomly divided into a training set and a test set in a 6:1 ratio, with a total of 600 pairs of CT chord image images in the training set and 100 pairs of CT chord image images in the test set.

[0131] S4. The proximal operator is trained using a chord graph data recovery algorithm that incorporates multi-Gaussian electronic noise, and the proximal operator is computed using a deep ResNet network model.

[0132] S5. Train the low-dose CT image denoising model using the Adam optimizer and an adaptive learning rate decay strategy. The hyperparameters set are: 20 training epochs, 2 data batch sizes, and an initial learning rate of 0.0001.

[0133] S6. After model training is complete, call the model weight file saved during training and input low-dose CT chordogram images from the test set to obtain denoised, high-quality CT chordograms. The chordogram at the conventional dose level (200 mAs, 120 kVp) is shown below. Figure 2 As shown, the restored chord diagram is as follows: Figure 3 As shown, the chord diagram for low dose levels (20 mAs, 120 kVp) is as follows: Figure 4 As shown; FBP reconstructed images at conventional dose levels (200 mAs, 120 kVp) are as follows. Figure 5 As shown, the FBP reconstructed image after scanning a chord graph at a low dose level (20 mAs, 120 kVp) and recovering it using multi-Gaussian error modeling is as follows. Figure 6 As shown, the FBP reconstructed image at a low dose level (20 mAs, 120 kVp) is as follows: Figure 7 As shown.

[0134] On the test set, the FBP reconstructed image restored using multi-Gaussian error modeling achieved a PSNR (Peak Signal-to-Noise Ratio) of 36.29 and an SSIM (Structure Similarity Index Measure) of 0.948. These results demonstrate that this method can effectively denoise low-dose CT chordograms.

[0135] Example 2: The present invention also provides a low-dose CT chordogram recovery system based on multi-Gaussian electronic noise modeling, comprising a data acquisition module, a chordogram data recovery module, and an image conversion module;

[0136] The data acquisition module is used to acquire low-dose CT chordogram data and input the low-dose CT chordogram data into a low-dose CT chordogram data recovery model that incorporates Gaussian electronic noise;

[0137] The string diagram data recovery module uses an alternating projection iterative algorithm to solve the low-dose CT string diagram data recovery model with Gaussian electron noise. The solution of the low-dose CT string diagram data recovery model with Gaussian electron noise is decomposed into solving the problem of solving the photon number T and solving the string diagram data Y. The proximal gradient descent method is used to solve Y to obtain the recovered string diagram data Y.

[0138] The image conversion module uses a CT imaging algorithm to convert the chordal data Y into viewable CT images.

[0139] It also includes a model training module, which acquires several conventional dose CT images and transforms them into chordal data using inverse Lardon transform. A mixture of K Gaussian distributions is then superimposed in the projection domain to simulate a more realistic electronic noise distribution, forming low-dose projection domain data. An alternating projection optimization algorithm is used to decompose the chordal data recovery model into sub-problems concerning several parameters to be estimated. The EM algorithm and proximal gradient descent algorithm are used to iteratively update each parameter, including:

[0140] By fixing the Gaussian parameters Π, Λ, Σ and chord diagram data Y, a sub-optimization model is established to solve for the photon number T. This model utilizes T... i The variable separation property transforms the original problem into a single-variable T. i Optimization issues:

[0141]

[0142] The objective function is updated using Newton's method, and f(T) is calculated. i The first and second derivatives of f′(T) i ), f″(Ti Update T using Newton's method:

[0143]

[0144]

[0145]

[0146]

[0147] Where R is the floor function, n is the iteration index, and η1 is the step size. In the last iteration, rounding is used to approximate T. i The true integer value;

[0148] With a fixed number of photons T, Y is solved using the near-end gradient descent method. The update iterative formula for Y is as follows:

[0149]

[0150] Among them, the proximal operator Learning is achieved using residual networks;

[0151] Using the photon number T and the raw data S, and by leveraging the difference between the raw chord diagram data and the photon number received by the detector, the EM algorithm is used to estimate the multi-Gaussian distribution parameters, obtaining and fixing the multi-Gaussian parameters Π, Λ, and Σ.

[0152] Repeat the above steps until T and Y converge.

[0153] In summary, this invention provides a low-dose CT chordogram restoration method based on multi-Gaussian electronic noise modeling. Based on the CT projection noise generation model, a low-dose CT chordogram restoration model based on multi-Gaussian noise modeling is established. A maximum a posteriori distribution is constructed under known chordogram data conditions. The alternating projection method is used to decompose the chordogram restoration model into sub-problems concerning several parameters to be estimated. The multi-Gaussian parameters are updated and solved using the EM algorithm, and the photon number is iteratively solved. The chordogram data is iteratively solved using a proximal operator, which is calculated using a ResNet network model. This invention overcomes the limitations of traditional techniques that are restricted to assuming the distribution of electronic noise in the image data. While electronic noise follows a single Gaussian distribution, this invention addresses the limitations of more complex electronic noise distributions by providing accurate estimations. It proposes a Poisson distribution + multi-Gaussian mixed noise model, cleverly incorporating multi-Gaussian parameters without modeling them as variables. This overcomes the difficulty in designing solution algorithms when multi-Gaussian parameters are modeled as variables. By approximating real electronic noise with a multi-Gaussian mixed distribution, it better preserves the true structure of chordogram data, enabling the reconstruction of clearer CT images using FBP imaging algorithms or other novel imaging algorithms. This demonstrates superior capabilities in chordogram restoration and significant performance improvements in structural similarity and visual consistency. Furthermore, this invention offers better adaptability and ease of operation, making it particularly suitable for processing large-scale datasets, especially in the field of clinical computed tomography, effectively meeting the urgent need for rapid imaging processing.

[0154] In addition, the present invention can also provide a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads part or all of the computer executable program from the memory and executes it, and the processor can realize the low-dose CT chordal recovery method based on multi-Gaussian electronic noise modeling described in the present invention when executing part or all of the computer executable program.

[0155] On the other hand, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the implementation of the low-dose CT chordal recovery method based on multi-Gaussian electronic noise modeling described in the present invention.

[0156] The computer device may be a laptop, a desktop computer, or a workstation.

[0157] The processor described in this invention may be a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or an off-the-shelf programmable gate array (FPGA).

[0158] The memory described in this invention can be an internal storage unit of a laptop, desktop computer, or workstation, such as memory or hard disk; or it can be an external storage unit, such as a portable hard disk or flash memory card.

[0159] Computer-readable storage media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer-readable storage media can include: read-only memory (ROM), random access memory (RAM), solid-state drives (SSDs), or optical discs, etc. Random access memory can include resistive random access memory (ReRAM) and dynamic random access memory (DRAM).

[0160] The foregoing description presents and describes several preferred embodiments of the invention. However, as previously indicated, it should be understood that the invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the invention should be within the protection scope of the appended claims.

Claims

1. A low-dose CT sinogram restoration method based on modeling of multi-Gaussian electronic noise, characterized in that, The method comprises the following steps: The low-dose CT chord data is input into a low-dose CT chord data restoration model with introduced Gaussian electronic noise; the low-dose CT chord data restoration model with introduced Gaussian electronic noise is: (1) wherein, represents the number of photons received by the detector, N is the total number of X-ray incident paths in the scan, represents the number of photons received by the detector when the i-th incident X-ray path is incident along the projection path, , K represents the number of multi-Gaussians, , , represents the variance and mean of the i-th Gaussian distribution, respectively, k represents the weight of the i-th Gaussian distribution, k is the intensity of the i-th incident X-ray, is the number of photons received by the detector when the i-th incident X-ray is incident along the projection path, is the sinogram data, i is the sinogram data on the detector when the i-th incident X-ray is incident along the projection path, is the regularization function with respect to the i-th Gaussian distribution, and is the penalty coefficient of the regularization function;​​​​​​​ The training of the low-dose CT chord data restoration model with introduced Gaussian electronic noise comprises the following steps: A plurality of regular-dose CT images are acquired, the regular-dose CT images are converted into chord data through inverse Radon transformation, a mixed distribution composed of K Gaussian distributions is superimposed in the projection domain to simulate a more real electronic noise distribution, and low-dose projection domain data are formed; The string diagram data recovery model is decomposed into sub-problems about several to-be-estimated parameters by using an alternating projection optimization algorithm, and each parameter is iteratively updated by using an EM algorithm and a proximal gradient descent algorithm respectively, and the string diagram data after convergence of the algorithm Y and the string diagram data recovery model. An alternating projection iterative algorithm is used to solve a low-dose CT sinogram data restoration model with Gaussian electronic noise. The low-dose CT sinogram data restoration model with Gaussian electronic noise is decomposed into a problem of solving photon number and a problem of solving sinogram data Y . A proximal gradient descent method is used to solve Y , and the restored sinogram data Y is obtained. CT imaging algorithms are used to convert chord diagram data Y into viewable CT images.

2. The low-dose CT sinogram restoration method based on modeling of multi-Gaussian electronic noise according to claim 1, characterized in that, The chord data restoration model is decomposed into sub-problems about a plurality of to-be-estimated parameters by using an alternating projection optimization algorithm, and each parameter is iteratively updated by using an EM algorithm and a proximal gradient descent algorithm, which comprises: Fixed polygamma parameters and chord diagram data Y, set up a sub-optimization model to solve the number of photons , using T i the variable separation property, convert the original problem into a single variable T i optimization problem: The objective function is updated using Newton's method, and the calculation is performed. First and second derivatives , Using Newton's method T Update: where R is the rounding function, the superscript n is the iteration index, is the step size, in the last step iteration, the rounding operation is used to approximate the real integer value of . Fixed number of photons , the proximal gradient descent method is used to solve Y , Y The update iteration formula of is as follows: wherein the proximal operator Residual networks are learned. with the number of photons T and the original data S , with the difference between the original chord diagram data and the number of photons received by the detector, the multi-Gaussian distribution parameters are estimated by the EM algorithm to obtain the multi-Gaussian parameters and fixed, Repeat the above steps until T and Y converge or iterate to a maximum number of times.

3. The low-dose CT sinogram restoration method based on modeling of multi-Gaussian electronic noise according to claim 2, characterized in that, The residual network is composed of four residual blocks, each of which is composed of two layers of convolution, batch normalization and nonlinear activation function ReLU. After each convolution, the output is normalized, nonlinearly transformed by ReLU activation function, and the input data is directly transmitted to the output layer by using residual connection. The historical update value is spliced to the current , and the spliced result is input into the residual network for further processing.

4. The low-dose CT sinogram restoration method based on modeling of multi-Gaussian electronic noise according to claim 1, characterized in that, The CT imaging algorithm is converted into a final CT image by using a FBP algorithm.

5. The low-dose CT sinogram restoration method based on modeling of multi-Gaussian electronic noise according to claim 1, characterized in that, An Adam optimizer is used to train the low-dose CT image denoising model in cooperation with an adaptive learning rate decay strategy.

6. A low-dose CT sinogram restoration system based on modeling of multi-Gaussian electronic noise, characterized by, A data acquisition module, a chord data restoration module and an image conversion module; The data acquisition module is used to acquire low-dose CT chord data and input the low-dose CT chord data into a low-dose CT chord data restoration model with introduced Gaussian electronic noise; the low-dose CT chord data restoration model with introduced Gaussian electronic noise is: (1) in, This represents the number of photons received by the detector, where N is the total number of X-ray incident paths during the scan. Indicates entering the first The incident X-ray path, , K Indicates the number of Gaussians. , , They represent the first k The variance and mean of a Gaussian distribution. Indicates the first k The weights of a Gaussian distribution For the first The intensity of the incident X-ray, For along the projection path The number of photons received by the detector at the time of incidence. For chord diagram data, For along the projection path i The chord diagram data on the detector at the time of incidence, For about The regularization function is This is the penalty coefficient of the regularization function; The training of the low-dose CT chord data restoration model with introduced Gaussian electronic noise comprises the following steps: A plurality of regular-dose CT images are acquired, the regular-dose CT images are converted into chord data through inverse Radon transformation, a mixed distribution composed of K Gaussian distributions is superimposed in the projection domain to simulate a more real electronic noise distribution, and low-dose projection domain data are formed; The string diagram data recovery model is decomposed into sub-problems about several to-be-estimated parameters by using an alternating projection optimization algorithm, and each parameter is iteratively updated by using an EM algorithm and a proximal gradient descent algorithm respectively, and the string diagram data after convergence of the algorithm Y and the string diagram data recovery model. The chord data recovery module solves the low-dose CT chord data recovery model with introduced Gaussian electronic noise by using an alternating projection iteration algorithm, decomposes the low-dose CT chord data recovery model with introduced Gaussian electronic noise into solving the photon number problem and solving the chord data Y problem, solves Y by using a proximal gradient descent method, and obtains recovered chord data Y . The image conversion module employs CT imaging algorithms to convert chord diagram data Y into viewable CT images.

7. A computer device, comprising: The processor and the memory are included, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes, and the processor can realize the low-dose CT chord restoration method based on multi-Gaussian electronic noise modeling according to any one of claims 1-5 when executing the computer executable program.

8. A computer-readable storage medium, characterized in that, A computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the low-dose CT chord restoration method based on multi-Gaussian electronic noise modeling according to any one of claims 1-5.

Images