A CT artifact removal method and system based on deep learning

Abstract

The present invention discloses a method and system for removing CT artifacts based on deep learning, which relates to the field of medical imaging technology, including: obtaining a CT image, performing sine transform and wavelet transform processing on the CT image respectively; constructing an image enhancement model, optimizing the processed image by sequentially connected image enhancement models and random back-projection layers respectively; coupling the optimized image and the original image and inputting them into the image enhancement model for reprocessing; adding the reprocessed image and the optimized image by elements to obtain an artifact-removed CT image. The present invention introduces wavelet transform to process CT images, which is used to extract CT image context and spatial information, effectively extract feature information in the artifact removal process, and improve the performance of image enhancement; establishes a CT image resolution enhancement model based on the VMamba model, improves the long-term dependence of network training, effectively identifies and removes radioactive artifacts, and improves network training efficiency.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and more specifically to a CT artifact removal method and system based on deep learning. Background Art

[0002] CT (Computer tomography) technology has been widely used in the medical field as a non-invasive and rapid detection method. With the increasing use of metal implants in the medical field, such as dental implants, spinal fixation devices and hip replacements, CT image reconstruction often produces light and dark artifacts due to the interference of implants, which makes it difficult to clearly identify the interface between metal and normal tissue and its surrounding structures, thus affecting doctors' evaluation of surgical results and subsequent treatment decision-making.

[0003] When performing CT scanning, if there are high-attenuation materials (such as metal, high-density bones, etc.) within the scanning range, they will absorb low-energy photons in the X-rays, causing some photons to fail to reach the detector. This phenomenon causes the energy distribution of the penetrating rays to shift, namely, ray hardening. The formation of artifacts is the result of the combined effect of multiple factors, the main reasons include ray hardening and photon starvation, as well as interference from factors such as scattering effects, partial volume effects, and noise (such as metal artifacts, hardening artifacts, ring artifacts, and motion artifacts). In CT image reconstruction, metal artifacts usually appear as radial or striped brightness anomalies, which cover the real image information of the surrounding tissues, resulting in errors in the identification of materials and the division of boundaries. In the existing dual-energy CT (Dual-Energy CT) and CBCT (Cone Beam Computed Tomography) image reconstruction tasks, the completion of the processing of high-density objects (such as metal implants) will encounter serious X-ray attenuation problems. This attenuation can cause data loss or distortion, thus forming secondary artifacts.

[0004] Existing methods for solving artifact removal can be divided into the following categories: 1. Optimization of CT scanning parameters: Optimization of CT scanning parameters is a common method for removing artifacts. By increasing tube voltage, tube current, using narrow collimators, reducing scanning pitch and increasing scanning layer thickness, artifacts can be reduced in the data acquisition stage. However, simply optimizing CT scanning parameters has limited effect on reducing artifacts and is difficult to meet the requirements of clinical accurate diagnosis. 2. Hardware correction: Hardware correction is to reduce the impact of scattered rays on the detector by adding specific correction devices to each component of the X-ray imaging system, or to adjust the energy and energy spectrum distribution of the initial rays. The following are also common: ① Collimator correction: A method of absorbing scattered rays by placing a thin plate with a high absorption coefficient in front of the detector. This technology is effective in fan beam CT and can eliminate about 89% of scattered rays, but the effect is not obvious in CBCT. ② Air gap correction: Scattering correction is performed by increasing the distance between the object being measured and the detector. This method may cause image blur due to the limitation of the projection magnification ratio. ③ Anti-scatter grid correction: By setting a grid structure in front of the detector to block the scattered rays from reaching the detector, the scattering situation is effectively controlled when the X-ray scattering degree is high (for example, large-sized objects to be measured and wide irradiation fields), the imaging dose is high, and low-resolution reconstruction is required. However, under low-dose and low-scattering conditions, such as medical imaging, this method may reduce image quality and increase noise, and the suppression effect of scattered rays is also poor. 3. Software correction: Using the obtained X-ray projection image, according to the method of digital image processing, through the analysis of the image itself and the estimation of the properties of the scanned object, the scattering distribution map is obtained for correction. It is divided into the following two types: ① Monte Carlo simulation: It is a simulation technology based on statistical methods. Through mathematical calculation methods, the process of interaction between multi-energy X-rays and different substances can be simulated, and the nonlinear functional relationship between multi-energy projection and single-energy projection can be obtained, thereby completing artifact correction. Although it can improve the quality of CBCT images, it has high computational complexity and long operation time. In addition, before performing Monte Carlo simulation, the scanned object needs to be accurately modeled and analyzed, which is often difficult to achieve in medical applications. ② Deep learning: With the substantial progress in computer computing power, deep learning has developed rapidly in various fields, and has been significantly improved in image reconstruction and analysis. This technology has become a new method to reduce artifacts in CT images caused by implants. The advantage of deep learning is that it can solve the inverse problem in image processing. Using deep learning to reconstruct CT images to improve image quality has become a direction with great potential.

[0005] Therefore, how to propose a CT artifact removal method and system based on deep learning, use deep learning to reconstruct CT images, solve the problem of CT image artifacts, reduce the interference and impact of artifacts on images, and reduce clinical misinterpretation and misjudgment of images is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0006] In view of this, the present invention provides a CT artifact removal method and system based on deep learning, which uses deep learning to reconstruct CT images, solves the problem of CT image artifacts, reduces the interference and influence of artifacts on images, and reduces clinical misreading and misjudgment of images. In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A CT artifact removal method based on deep learning, comprising:

[0008] Acquire a CT image, and perform sine transform and wavelet transform on the CT image respectively;

[0009] Construct an image enhancement model, and optimize the processed image by passing the image enhancement model and random back-projection layer connected in sequence;

[0010] The optimized image and the original image are coupled and input into the image enhancement model for further processing;

[0011] The reprocessed image and the optimized image are element-wise added to obtain an artifact-removed CT image.

[0012] Optionally, acquiring the CT image includes: collecting medical CT data of the patient, including oral, neurological and orthopedic data, establishing a CT image dataset with artifacts, and selecting images with artifacts as a verification set.

[0013] Optionally, the wavelet transform processing includes: continuous wavelet transform and discrete wavelet transform, and convolution processing is performed on the image signal and the wavelet basis function.

[0014] Optionally, constructing the image enhancement model includes: using the mamba module as a submodule to build a U-net network structure to obtain a UMamba-Net image enhancement model, and the UMamba-Net image enhancement model includes three components: an encoder, a decoder, and a jump connection.

[0015] Optionally, the mamba module includes:

[0016] Construct a continuous system by which a one-dimensional input function or sequence is mapped to an input via an intermediate hidden state;

[0017] This continuous system is discretized, a time scale parameter is introduced and the state matrix and projection parameters are converted into discrete parameters using a fixed discretization rule;

[0018] After discretization, the computation is performed via linear recursion or global convolution.

[0019] Optionally, the structure of the UMamba-Net image enhancement model is: coupling the VSS module with the Patch merging layer to obtain a first substructure, and coupling the VSS module with the Patch expansion layer to obtain a second substructure;

[0020] Placing the plurality of first substructures and the plurality of second substructures symmetrically, connecting the plurality of first substructures in sequence from top to bottom, with the topmost first substructure as input, and connecting the plurality of second substructures in sequence from bottom to top, with the topmost second substructure as output;

[0021] The bottom-level first substructure is connected to the bottom-level second substructure through two VSS modules in turn, and the output value of the first substructure and the output value of the second substructure corresponding to the lower level of the second substructure are element-wise added as the input value of the corresponding second substructure.

[0022] Optionally, the VSS module includes: after layer normalization, the input is divided into two branches, in the first branch, the input passes through a linear layer and then an activation function; in the second branch, the input is processed by a linear layer, a depth-wise separable convolution and an activation function, and then sent to the SS2D module for feature extraction; the extracted features are standardized by layer normalization and element-wise multiplied with the output of the first branch to merge the two paths, and finally the features are mixed using a linear layer, and this result is combined with a residual connection to form the output of the VSS block.

[0023] Optionally, the activation function is a SiLU activation function.

[0024] Optionally, the SS2D module is a 2D-selective scanning module, and the SS2D module consists of three parts: a scan expansion operation, an S6 module, and a scan merging operation. The scan expansion operation expands the input image along four different directions to form a sequence, and the sequence is processed by the S6 module to extract features, and the information from each direction is scanned. The scan merging operation sums and merges the sequences from different directions. The S6 module is mainly composed of a series of linear transformations and discretization processes, which are used to process the input feature sequence. The S6 module is a complex component in the Mamba architecture, responsible for processing the input feature sequence through a series of linear transformations and discretization processes. It plays a key role in capturing the temporal dynamics of the sequence, which is a key aspect of the sequence modeling task, including tensor operations and custom discretization methods to handle the complex requirements of sequence data.

[0025] Optionally, a deep learning-based CT artifact removal system, comprising:

[0026] Processing module: used to obtain CT images and perform sine transform and wavelet transform on the CT images;

[0027] Model building and optimization module: used to build an image enhancement model, and optimize the processed image through the image enhancement model and random back-projection layer connected in sequence;

[0028] Reprocessing module: used to couple the optimized image and the original image and input them into the image enhancement model for reprocessing;

[0029] Artifact removal module: used to add the reprocessed image and the optimized image element by element to obtain an artifact-removed CT image.

[0030] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a CT artifact removal method and system based on deep learning, which has the following beneficial effects:

[0031] The present invention proposes a CT artifact removal method based on deep learning, including: obtaining a CT image, performing sine transform and wavelet transform processing on the CT image respectively; constructing an image enhancement model, optimizing the processed image through the image enhancement model and the random back-projection layer connected in sequence respectively; coupling the optimized image and the original image and inputting them into the image enhancement model for reprocessing; adding the reprocessed image and the optimized image element by element to obtain an artifact-removed CT image. The present invention achieves the improvement of image quality under the same radiation dose; reduces the interference and influence of artifacts on the image, reduces clinical misreading and misjudgment of images; and improves the performance and efficiency of network training. By collecting the medical CT data of patients and establishing a CT image data set with artifacts, some artifact images are selected for network model verification; introducing wavelet transform to process the CT image, which is used to extract CT image context and spatial information, and effectively extracting feature information during the artifact removal training process, thereby improving the performance of image enhancement; establishing a CT image resolution enhancement model based on the VMamba model, improving the long-term dependence of network training, effectively identifying and removing radioactive artifacts, and improving network training efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0033] Figure 1 A schematic flow chart of a CT artifact removal method based on deep learning provided by the present invention.

[0034] Figure 2 This is a schematic diagram of the UMamba-Net image enhancement model structure provided by the present invention.

[0035] Figure 3 This is a schematic diagram of the VSS module structure provided by the present invention.

[0036] Figure 4 A schematic diagram of image comparison with and without artifacts provided by the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] Deep learning is an important branch of machine learning. It processes complex data patterns by simulating the structure and function of human brain neural networks. The core of deep learning is the use of multi-layer neural networks, which can learn useful feature representations from large amounts of data without artificially designing feature extraction algorithms. Deep learning is the study of the inherent laws and representation levels of sample data. The information obtained during the learning process is of great help in the interpretation of data such as text, images, and sounds.

[0039] Deep learning has two major applications in the field of medical imaging. The first application is intelligent imaging systems, including fast reconstruction, noise reduction, and super-resolution imaging; the second is intelligent analysis of medical image big data, including classification, detection, segmentation, and registration of medical images. Through deep neural network learning, artificial intelligence-assisted medical image processing technology has been developed, which can restore image data under high radiation dose conditions at low doses and improve image quality; at the same time, it can reduce the impact of artifacts on surrounding tissues, and compare images with and without artifacts. Figure 4 shown.

[0040] Supervised networks, including convolutional neural networks (CNNs) and U-Nets, are widely used in medical image processing. In addition, unsupervised methods have also been introduced to address artifacts in the latent space of generative adversarial networks (GANs). Typically, processing damaged sinusoids can address divergent artifacts caused by metal, and the projection data outside the metal artifacts is treated as clean data. Among them, one of the challenges of sinusoidal restoration is the secondary artifacts caused by discontinuities after the restoration process. Since each layer of a CNN has only a limited receptive field, such as a 3×3 convolution kernel, deep layers can only partially capture long-range discontinuities. In order to fully mine the local information of surrounding angles and detectors, additional prior information needs to be utilized for further improvement. For example, metal mask projection and adaptive scale can mine more information in the damaged area, resulting in better performance.

[0041] Furthermore, the dual-domain network can solve the problem of artifact removal. By using two or more rough enhanced images, the dual-domain architecture greatly improves the performance of the single-domain method. However, in the field of sinusoids and images, the fundamental problem of limited receptive field remains unsolved. In addition to these two domains, the use of wavelet transform to extract time-frequency localization characteristics can provide high resolution in both time and frequency domains, and can easily provide context and spatial information for global recovery. In the backbone network, since the artifacts are radioactive, the UMamba module is used to build an image enhancement network UMamba-Net. UMamba-Net can improve the long-term memory of the network during training, effectively improve the effect of artifact removal, and the network can be quickly trained in the RNN mode, thereby improving the model artifact removal training efficiency.

[0042] The embodiment of the present invention discloses a CT artifact removal method based on deep learning, such as Figure 1 As shown, including:

[0043] Acquire a CT image, and perform sine transform and wavelet transform on the CT image respectively;

[0044] Construct an image enhancement model, and optimize the processed image by passing the image enhancement model and random back-projection layer connected in sequence;

[0045] The optimized image and the original image are coupled and input into the image enhancement model for further processing;

[0046] The reprocessed image and the optimized image are element-wise added to obtain an artifact-removed CT image.

[0047] Furthermore, the acquisition of CT images includes: collecting medical CT data of patients, more than 2,000 cases, including oral, neurological and orthopedic, and establishing a CT image data set with artifacts, among which some images with artifacts are selected for verification of the network model.

[0048] Furthermore, the wavelet transform processing includes: continuous wavelet transform and discrete wavelet transform, and the image signal is convolved with the wavelet basis function.

[0049] Specifically, wavelet transform has good time-frequency localization characteristics and can provide high resolution in both the time domain and the frequency domain. It can accurately locate the change points in the image, such as edges and textures. Wavelet transform is achieved by convolving the image signal with the wavelet basis function. The formula for continuous wavelet transform (CWT) is:

[0050]

[0051] Where x(t) is the original signal, ψ(t) is the mother wavelet function, a and b are the scale and translation parameters respectively, and * represents the complex conjugate.

[0052] The formula for discrete wavelet transform (DWT) is:

[0053] W ψ [j,k]=∑ n x[n]ψ j,k [n];

[0054] in is the discrete wavelet transform, j and k represent the scale and displacement parameters respectively, and n represents the discrete time index.

[0055] Wavelet transform is introduced to process CT images to extract CT image context and spatial information, and effectively extract feature information during artifact removal training, thereby improving image enhancement (resolution) performance.

[0056] Furthermore, the image enhancement model is constructed by using the mamba module as a submodule to build a U-net network structure to obtain a UMamba-Net image enhancement model. The UMamba-Net image enhancement model includes three components: an encoder, a decoder, and a skip connection, such as Figure 2 shown.

[0057] Furthermore, the mamba module includes:

[0058] Construct a continuous system by which a one-dimensional input function or sequence is mapped to an input via an intermediate hidden state;

[0059] This continuous system is discretized, a time scale parameter is introduced and the state matrix and projection parameters are converted into discrete parameters using a fixed discretization rule;

[0060] After discretization, the computation is performed via linear recursion or global convolution.

[0061] Furthermore, the structure of the UMamba-Net image enhancement model is: coupling the VSS module with the Patch merging layer to obtain a first substructure, and coupling the VSS module with the Patch expansion layer to obtain a second substructure;

[0062] Placing the plurality of first substructures and the plurality of second substructures symmetrically, connecting the plurality of first substructures in sequence from top to bottom, with the topmost first substructure as input, and connecting the plurality of second substructures in sequence from bottom to top, with the topmost second substructure as output;

[0063] The bottom-level first substructure is connected to the bottom-level second substructure through two VSS modules in turn, and the output value of the first substructure and the output value of the second substructure corresponding to the lower level of the second substructure are element-wise added as the input value of the corresponding second substructure.

[0064] In a specific embodiment, the UMamba-Net image enhancement model specifically includes:

[0065] Modern SSMs (such as Mamba) not only establish long-distance dependencies, but also show linear complexity with respect to the input size. Taking Mamba as a submodule, we build a UMamba-Net image enhancement model with a U-net network structure. UMamba-Net mainly consists of three main parts: encoder, decoder, and skip connection.

[0066] Step 1: In modern SSM-based models, namely structured state-space sequence models (S4) and Mamba, we rely on a continuous system that transforms a one-dimensional input function or sequence (denoted as x(t)∈R) through an intermediate hidden state h(t)∈R. N Mapped to an input y(t)∈R. The above process is expressed as a linear ordinary differential equation (ODE):

[0067] h′(t)=Ah(t)+Bx(t);

[0068] y(t) = Ch(t);

[0069] Among them, A∈R N×N Represents the state matrix, B∈R N×1 , C∈R N×1 They represent the projection parameters respectively.

[0070] Step 2: S4 and Mamba discretize this continuous system to make it more suitable for deep learning scenarios.

[0071] A time scale parameter Δ is introduced and A and B are converted into discrete parameters using a fixed discretization rule and Zero-order hold (ZHO) is used as the discretization rule and is defined as follows:

[0072]

[0073] where I is the identity matrix with the same dimensions as matrix A.

[0074] Step 3: After discretization, the SSM-based model can be calculated in two ways: linear recursion or global convolution, as follows:

[0075]

[0076] y(t) = Ch(t);

[0077]

[0078]

[0079] in, represents a structured convolution kernel, and L represents the length of the input sequence x.

[0080] A CT image resolution enhancement model based on the VMamba model is established to improve the long-term dependence of network training, effectively identify and remove radioactive artifacts, and improve network training efficiency.

[0081] Furthermore, the VSS module includes: The VSS module is derived from the structure of VMamaba and is the core module of VM-UNet. After layer normalization, the input is divided into two branches. In the first branch, the input passes through a linear layer and then an activation function; in the second branch, the input is processed by a linear layer, a depth-separable convolution and an activation function, and then sent to the SS2D module for feature extraction; the extracted features are standardized by layer normalization and element-wise multiplied with the output of the first branch to merge the two paths, and finally the linear layer is used to mix the features, and this result is combined with the residual connection to form the output of the VSS block, such as Figure 3 shown.

[0082] Furthermore, the activation function is a SiLU activation function.

[0083] Further, the SS2D module is a 2D-selective scanning module, which consists of three parts: a scan expansion operation, an S6 module, and a scan merge operation. The scan expansion operation expands the input image into a sequence along four different directions (from upper left to lower right, from lower right to upper left, from upper right to lower left, and from lower left to upper right), and the sequence is processed by the S6 module to extract features to ensure that information from all directions is thoroughly scanned, thereby capturing a variety of features. The information from all directions is scanned, and the scan merge operation sums and merges the sequences from different directions.

[0084] In a specific embodiment, a CT artifact removal system based on deep learning includes:

[0085] Processing module: used to obtain CT images and perform sine transform and wavelet transform on the CT images;

[0086] Model building and optimization module: used to build an image enhancement model, and optimize the processed image through the image enhancement model and random back-projection layer connected in sequence;

[0087] Reprocessing module: used to couple the optimized image and the original image and input them into the image enhancement model for reprocessing;

[0088] Artifact removal module: used to add the reprocessed image and the optimized image element by element to obtain an artifact-removed CT image.

[0089] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0090] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A CT artifact removal method based on deep learning, characterized in that: include: Acquire a CT image, and perform sine transform and wavelet transform on the CT image respectively; Construct an image enhancement model, and optimize the processed image by passing the image enhancement model and random back-projection layer connected in sequence; The optimized image and the original image are coupled and input into the image enhancement model for further processing; The reprocessed image and the optimized image are element-wise added to obtain an artifact-removed CT image.

2. The method for removing CT artifacts based on deep learning according to claim 1, characterized in that: The acquisition of CT images includes: collecting medical CT data of patients, including oral, neurological and orthopedic data, establishing a CT image data set with artifacts, and selecting images with artifacts as a verification set.

3. The method for removing CT artifacts based on deep learning according to claim 1, characterized in that: The wavelet transform process includes: continuous wavelet transform and discrete wavelet transform, and the image signal is convolved with the wavelet basis function.

4. The method for removing CT artifacts based on deep learning according to claim 1, characterized in that: The image enhancement model is constructed by using the mamba module as a submodule to build a U-net network structure to obtain a UMamba-Net image enhancement model. The UMamba-Net image enhancement model includes three components: an encoder, a decoder and a skip connection.

5. The method for removing CT artifacts based on deep learning according to claim 4, characterized in that: The mamba module includes: Construct a continuous system by which a one-dimensional input function or sequence is mapped to an input via an intermediate hidden state; This continuous system is discretized, a time scale parameter is introduced and the state matrix and projection parameters are converted into discrete parameters using a fixed discretization rule; After discretization, the computation is performed via linear recursion or global convolution.

6. The method for removing CT artifacts based on deep learning according to claim 4, characterized in that: The structure of the UMamba-Net image enhancement model is: the VSS module is coupled to the Patch merging layer to obtain a first substructure, and the VSS module is coupled to the Patch expansion layer to obtain a second substructure; Placing the plurality of first substructures and the plurality of second substructures symmetrically, connecting the plurality of first substructures in sequence from top to bottom, with the topmost first substructure as input, and connecting the plurality of second substructures in sequence from bottom to top, with the topmost second substructure as output; The bottom-level first substructure is connected to the bottom-level second substructure through two VSS modules in turn, and the output value of the first substructure and the output value of the second substructure corresponding to the lower level of the second substructure are element-wise added as the input value of the corresponding second substructure.

7. The method for removing CT artifacts based on deep learning according to claim 6, characterized in that: The VSS module includes: after layer normalization, the input is divided into two branches, in the first branch, the input passes through a linear layer and then an activation function; in the second branch, the input is processed by a linear layer, a depth-separable convolution and an activation function, and then sent to the SS2D module for feature extraction; the extracted features are standardized by layer normalization and element-wise multiplied with the output of the first branch to merge the two paths, and finally the linear layer is used to mix the features, and this result is combined with the residual connection to form the output of the VSS block.

8. The method for removing CT artifacts based on deep learning according to claim 7, characterized in that: The activation function is the SiLU activation function.

9. The method for removing CT artifacts based on deep learning according to claim 7, characterized in that: The SS2D module is a 2D-selective scanning module, which consists of three parts: a scan extension operation, an S6 module and a scan merging operation. The scan extension operation expands the input image along four different directions to form a sequence. The sequence is processed by the S6 module to extract features and scan information from various directions. The scan merging operation sums and merges sequences from different directions.

10. A CT artifact removal system based on deep learning, characterized in that: include: Processing module: used to obtain CT images and perform sine transform and wavelet transform on the CT images; Model building and optimization module: used to build an image enhancement model, and optimize the processed image through the image enhancement model and random back-projection layer connected in sequence; Reprocessing module: used to couple the optimized image and the original image and input them into the image enhancement model for reprocessing; Artifact removal module: used to add the reprocessed image and the optimized image element by element to obtain an artifact-removed CT image.

Images