CN117765110A - Equal-transformation near-end operator method and device suitable for CT deep learning reconstruction - Google Patents
Equal-transformation near-end operator method and device suitable for CT deep learning reconstruction Download PDFInfo
- Publication number
- CN117765110A CN117765110A CN202311717912.6A CN202311717912A CN117765110A CN 117765110 A CN117765110 A CN 117765110A CN 202311717912 A CN202311717912 A CN 202311717912A CN 117765110 A CN117765110 A CN 117765110A
- Authority
- CN
- China
- Prior art keywords
- reconstruction
- image
- metal artifact
- metal
- result
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 121
- 238000013135 deep learning Methods 0.000 title claims abstract description 30
- 239000002184 metal Substances 0.000 claims abstract description 248
- 238000005457 optimization Methods 0.000 claims abstract description 112
- 230000008569 process Effects 0.000 claims abstract description 71
- 238000013528 artificial neural network Methods 0.000 claims abstract description 51
- 230000006870 function Effects 0.000 claims abstract description 36
- 238000007781 pre-processing Methods 0.000 claims abstract description 35
- 238000012937 correction Methods 0.000 claims abstract description 29
- 238000011478 gradient descent method Methods 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 229910052755 nonmetal Inorganic materials 0.000 claims description 13
- 238000010276 construction Methods 0.000 claims description 8
- 238000003384 imaging method Methods 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 6
- 101100518501 Mus musculus Spp1 gene Proteins 0.000 claims 1
- 238000013527 convolutional neural network Methods 0.000 abstract description 9
- 238000002591 computed tomography Methods 0.000 description 102
- 238000010586 diagram Methods 0.000 description 13
- 238000012549 training Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000012733 comparative method Methods 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 230000000052 comparative effect Effects 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000013170 computed tomography imaging Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000000844 transformation Methods 0.000 description 2
- 101100365548 Caenorhabditis elegans set-14 gene Proteins 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003189 isokinetic effect Effects 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 229920001296 polysiloxane Polymers 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The application discloses an isomorphism near-end operator method and device suitable for CT deep learning reconstruction, comprising the following steps: preprocessing the CT image before the constructed total network model; acquiring processed CT image data through a preprocessing step; establishing a metal artifact correction model corresponding to the image damaged by the metal artifact, and constructing a reconstruction model to obtain a total reconstruction optimization process; performing an iterative solution step on the total weight optimization process by using a near-end gradient descent method; constructing an initial CT reconstruction neural network, inputting the metal artifact free image and the metal artifact damaged image obtained in the preprocessing step into the initial CT reconstruction neural network, and calculating a loss function; and in the testing stage, the image damaged by the metal artifact to be tested is transmitted into a final CT reconstruction neural network to carry out a forward inference step, the reconstructed image without the metal artifact is output, and the problem that the existing near-end operator network based on the convolutional neural network cannot effectively describe the prior of the rotational symmetry is solved.
Description
Technical Field
The application relates to the technical field of medical image processing and deep learning, in particular to an isomorphism near-end operator method and device suitable for CT deep learning reconstruction.
Background
CT imaging of medical X-rays has been widely used for disease diagnosis and treatment, and is more likely to be found in some abnormalities in the human body. Then during the actual acquisition of the CT image, some metal objects in the body of the person being CT scanned may cause metal artifacts in the radial form of the last imaged CT image. These radial metal artifacts not only reduce the imaging quality of the CT image, but also affect the subsequent physician's image-based diagnosis. Therefore, in recent years, in CT imaging, removal of metal artifacts from a CT image damaged by the metal artifacts has become a significant research problem, and many researchers have attracted attention.
Currently, there are many techniques for reconstructing CT images damaged by metal artifacts, which are mainly classified into a conventional method and a deep learning method, but the existing near-end operator network based on a convolutional neural network cannot effectively characterize the rotation symmetry prior.
Disclosure of Invention
In the embodiment of the application, the isomorphism near-end operator method suitable for CT deep learning reconstruction is provided, and the problem that the existing near-end operator network based on the convolutional neural network cannot effectively describe the rotational symmetry priori is solved by effectively embedding the rotational symmetry priori of a CT image into CT deep learning reconstruction.
In a first aspect, embodiments of the present application provide an isomorphic near-end operator method suitable for CT deep learning reconstruction, the method comprising: constructing a rotating and other variable near-end operator total network model with a residual error structure, and preprocessing a CT image before the constructed total network model; wherein the total network model comprises a first sub-network model and a second sub-network model; acquiring processed CT image data through the preprocessing step; the processed CT image data comprise a metal artifact free image, a metal artifact damaged image and a nonmetal area image; establishing a metal artifact correction model corresponding to an image damaged by metal artifacts, and constructing a reconstruction model of the metal artifact correction model to obtain a total reconstruction optimization process; the metal artifact correction model is established according to a physical imaging mechanism of an image damaged by metal artifacts; performing an iterative solution step on the total weight optimization process by using a near-end gradient descent method; the total reconstruction optimization process comprises a reconstruction optimization process of a metal artifact layer to be estimated and a reconstruction optimization process of a metal artifact-free image to be estimated; acquiring a first reconstruction optimization result of a metal artifact layer to be estimated, and acquiring a second reconstruction optimization result of a metal artifact-free image to be estimated; the first sub-network model is brought into a first reconstruction optimization result, a first updating result is obtained, the second sub-network model is brought into a second reconstruction optimization result, and a second updating result is obtained; constructing an initial CT reconstruction neural network according to the first updating result and the second updating result, inputting the metal artifact free image and the metal artifact damaged image obtained in the preprocessing step into the initial CT reconstruction neural network, and calculating a loss function; wherein the calculating the loss function comprises: according to the loss obtained by calculating the loss function, performing iterative updating steps on initial parameters of the initial CT reconstruction neural network by adopting a random gradient descent method to obtain a final CT reconstruction neural network; and in the testing stage, transmitting the image to be tested, damaged by the metal artifact, into a final CT reconstruction neural network to perform a forward inference step, and outputting the reconstructed image without the metal artifact.
With reference to the first aspect, in one possible implementation manner, the establishing a metal artifact correction model corresponding to the image damaged by the metal artifact, performing reconstruction model construction on the metal artifact correction model, and obtaining a total reconstruction optimization process includes: the relation of the metal artifact correction model is as follows: i +.y=i +.x+i +.m; wherein Y εR H×W To destroy images by metal artifacts, X, M ε R H×W X is a metal artifact free image, M is a metal artifact layer to be estimated, I E R H×W For the non-metallic artifact region represented by the binary mask, +.; the total image reconstruction optimization process comprises the following steps:
wherein the method comprises the steps ofAnd->Respectively a metal artifact free image to be estimated and a metal artifact layer to be estimated, lambda 1 And lambda (lambda) 2 As the weight coefficient, f 1 (. Cndot.) and f 2 (. Cndot.) represents for->And->Regular term of->Representing the square norm.
With reference to the first aspect, in one possible implementation manner, the reconstruction optimization process of the metal artifact layer to be estimated and the reconstruction optimization process of the metal artifact free image to be estimated include: metal artifact layer to be estimatedThe reconstruction optimization process of (1) is as follows: / >Wherein M is (k) Solving the obtained updated result for the kth iteration, eta 1 For the step size of the first update, +.>For the purpose of the gradient operator,metal artifact free image to be estimated +.>The reconstruction optimization process of (1) is as follows:wherein X is (k) Solving the obtained updated result for the kth iteration, eta 2 For the step size of the second update, +.>For gradient operator->
With reference to the first aspect, in one possible implementation manner, the obtaining a first reconstruction optimization result of the metal artifact layer to be estimated obtains a second reconstruction optimization result of the metal artifact free image to be estimated; metal artifact layer to be estimatedThe first reconstruction optimization result of (a) is: /> Wherein proxλ1η1 is the near-end operator for f1; metal artifact free image to be estimated +.>The second reconstruction optimization result of (2) is:
wherein,is about f 2 The near-end operator of (-).
With reference to the first aspect, in one possible implementation manner, the bringing the first sub-network model into the first reconstruction optimization result, obtaining a first update result, bringing the second sub-network model into the second reconstruction optimization result, obtaining a second update result includes: the first update result is: wherein,for the first subnetwork model, θ 1 Is a first learnable parameter; the second update results are:
Wherein,for the second sub-network model, θ 2 Is a second learnable parameter.
With reference to the first aspect, in one possible implementation manner, a calculation formula of the loss function is as follows: wherein XGT is a metal artifact free image, μk, α1 and α2 are weight parameters, and when k=0, X (0) For initialized metal artifact free images, M (0) Is an initialized image corrupted by metal artifacts.
With reference to the first aspect, in one possible implementation manner, the performing an iterative update step on initial parameters of the initial CT reconstruction neural network by using a random gradient descent method according to the loss obtained by calculating the loss function to obtain a final CT reconstruction neural network includes: stopping the iterative updating step when the iterative step number reaches a first preset threshold value, and storing the final parameters at the moment, so as to obtain a final CT reconstruction neural network; the iterative updating step comprises the following steps: scaling the pixel values of the image without metal artifact and the image destroyed by metal artifact after the preprocessing step to between 0 and 1; setting the pixel value of the non-metal area image to 1 when the pixel value is higher than a second preset threshold value, and setting the pixel value of the non-metal area image to 0 when the pixel value is lower than the second preset threshold value; inputting the image destroyed by the metal artifact into a reconstruction network, and estimating a first metal artifact layer by using the initialized metal artifact layer and the initialized metal artifact-free image; estimating a first metal artifact free image using the obtained first metal artifact layer, the initialized metal artifact free image and the metal artifact corrupted image, thereby completing an iterative updating step; repeating the above process until the iterative updating step is completed.
In a second aspect, embodiments of the present application provide an isomorphic near-end operator apparatus suitable for deep learning reconstruction, the apparatus comprising: the preprocessing module is used for constructing a rotating and other near-end operator total network model with a residual error structure, and preprocessing CT images before the constructed total network model; wherein the total network model comprises a first sub-network model and a second sub-network model; the data preprocessing module is used for acquiring processed CT image data through the preprocessing step; the processed CT image data comprise a metal artifact free image, a metal artifact damaged image and a nonmetal area image; the method comprises the steps of obtaining a total reconstruction optimization process module, wherein the total reconstruction optimization process module is used for establishing a metal artifact correction model corresponding to an image damaged by metal artifacts, and carrying out reconstruction model construction on the metal artifact correction model to obtain a total reconstruction optimization process; the metal artifact correction model is established according to a physical imaging mechanism of an image damaged by metal artifacts; the iteration solution module is used for carrying out an iteration solution step on the total weight optimization process by using a near-end gradient descent method; the total reconstruction optimization process comprises a reconstruction optimization process of a metal artifact layer to be estimated and a reconstruction optimization process of a metal artifact-free image to be estimated; the reconstruction optimization result acquisition module is used for acquiring a first reconstruction optimization result of the metal artifact layer to be estimated and acquiring a second reconstruction optimization result of the metal artifact-free image to be estimated; the updating result acquisition module is used for bringing the first sub-network model into a first reconstruction optimization result, acquiring a first updating result, bringing the second sub-network model into a second reconstruction optimization result, and acquiring a second updating result; the computing module is used for constructing an initial CT reconstruction neural network according to the first updating result and the second updating result, inputting the metal artifact free image and the metal artifact damaged image obtained in the preprocessing step into the initial CT reconstruction neural network, and computing a loss function; wherein the calculating the loss function comprises: according to the loss obtained by calculating the loss function, performing iterative updating steps on initial parameters of the initial CT reconstruction neural network by adopting a random gradient descent method to obtain a final CT reconstruction neural network; and the output module is used for transmitting the image to be tested, damaged by the metal artifact, into the final CT reconstruction neural network to perform a forward inference step in the test stage, and outputting the reconstructed image without the metal artifact.
With reference to the second aspect, in one possible implementation manner, the establishing a metal artifact correction model corresponding to the image damaged by the metal artifact, performing reconstruction model construction on the metal artifact correction model, and obtaining a total reconstruction optimization process includes: the relation of the metal artifact correction model is as follows: i +.y=i +.x+i +.m; wherein Y εR H×W To destroy images by metal artifacts, X, M ε R H×W X is a metal artifact free image, M is a metal artifact layer to be estimated, I E R H×W For the non-metallic artifact region represented by the binary mask, +.; the total image reconstruction optimization process comprises the following steps:
wherein the method comprises the steps ofAnd->Respectively a metal artifact free image to be estimated and a metal artifact layer to be estimated, lambda 1 And lambda (lambda) 2 As the weight coefficient, f 1 (. Cndot.) and f 2 (. Cndot.) represents for->And->Regular term of->Representing the square norm.
With reference to the second aspect, in one possible implementation manner, the reconstruction optimization process of the metal artifact layer to be estimated and the reconstruction optimization process of the metal artifact free image to be estimated include: metal artifact layer to be estimatedThe reconstruction optimization process of (1) is as follows: / >Wherein M is (k) Solving the obtained updated result for the kth iteration, eta 1 For the step size of the first update, +.>For gradient operator->Metal artifact free image to be estimated +.>The reconstruction optimization process of (1) is as follows: />Wherein X is (k) Solving the obtained updated result for the kth iteration, eta 2 For the step size of the second update, +.>For the purpose of the gradient operator,
with reference to the second aspect, in one possible implementation manner, the obtaining a first reconstruction optimization result of the metal artifact layer to be estimated obtains a second reconstruction optimization result of the metal artifact free image to be estimated; metal artifact layer to be estimatedThe first reconstruction optimization result of (a) is: /> Wherein proxλ1η1 is the near-end operator for f1; metal artifact free image to be estimated +.>The second reconstruction optimization result of (2) is:
wherein,is about f 2 The near-end operator of (-).
With reference to the second aspect, in oneIn a possible implementation manner, the bringing the first sub-network model into the first reconstruction optimization result, obtaining a first update result, and bringing the second sub-network model into the second reconstruction optimization result, obtaining a second update result includes: the first update result is: wherein,for the first subnetwork model, θ 1 Is a first learnable parameter; the second update results are:
Wherein,for the second sub-network model, θ 2 Is a second learnable parameter.
With reference to the second aspect, in one possible implementation manner, a calculation formula of the loss function is as follows:
wherein X is GT Mu, as a metal artifact free image k 、α 1 And alpha 2 As a weight parameter, when k=0, X (0) For initialized metal artifact free images, M (0) Is an initialized image corrupted by metal artifacts.
With reference to the second aspect, in one possible implementation manner, the performing an iterative update step on the initial parameters of the initial CT reconstruction neural network by using a random gradient descent method according to the loss obtained by calculating the loss function to obtain a final CT reconstruction neural network includes: stopping the iterative updating step when the iterative step number reaches a first preset threshold value, and storing the final parameters at the moment, so as to obtain a final CT reconstruction neural network; the iterative updating step comprises the following steps: scaling the pixel values of the image without metal artifact and the image destroyed by metal artifact after the preprocessing step to between 0 and 1; setting the pixel value of the non-metal area image to 1 when the pixel value is higher than a second preset threshold value, and setting the pixel value of the non-metal area image to 0 when the pixel value is lower than the second preset threshold value; inputting the image destroyed by the metal artifact into a reconstruction network, and estimating a first metal artifact layer by using the initialized metal artifact layer and the initialized metal artifact-free image; estimating a first metal artifact free image using the obtained first metal artifact layer, the initialized metal artifact free image and the metal artifact corrupted image, thereby completing an iterative updating step; repeating the above process until the iterative updating step is completed.
In a third aspect, embodiments of the present application provide an isomorphic near-end operator server suitable for deep learning reconstruction, including a memory and a processor; the memory is used for storing computer executable instructions; the processor is configured to execute the computer-executable instructions to implement the method of the first aspect or any one of the possible implementation manners of the first aspect.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium storing executable instructions that when executed by a computer enable the method according to the first aspect or any one of the possible implementation manners of the first aspect.
One or more technical solutions provided in the embodiments of the present application at least have the following technical effects:
the embodiment of the application provides an equal-change near-end operator method suitable for CT deep learning reconstruction, which is characterized by constructing a total network model of a rotational equal-change near-end operator with a residual structure, preparing training data and training and testing a network. Firstly, according to rich symmetry information (including symmetry of human tissue outline and symmetry of radial metal artifact in CT image) contained in CT image, a new rotary and other variable near-end operator capable of describing rotation symmetry priori is established, an iterative solution algorithm for CT deep learning reconstruction model is designed by utilizing the rotary and other variable near-end operator, namely, the iterative solution step of the total reconstruction optimization process is carried out by using a near-end gradient descent method, and then the iterative solution step of the algorithm is correspondingly developed into a network module one by one. In depth-expanded CT reconstruction deep learning algorithms, it is often necessary to design a near-end operator network through which a priori information contained in the CT image is learned. The current common near-end operator network is mainly constructed by a common convolutional neural network (Convolution Neural Network, CNN), and the near-end operator constructed based on the common CNN can only describe translational symmetry priori in the CT image, but the CT image still has rich rotational symmetry priori which cannot be described by the common CNN. Therefore, the convolutional neural network of the group equalization (Group Equivariant) is utilized to remove the rotation of the residual error structure and the like to become a near-end operator total network model, and the total network model constructed in this way can explicitly describe the rotation symmetry priori in the CT image. Specifically, for the problem of metal artifact removal of CT images in CT reconstruction, one typically breaks down the CT image with metal artifacts into two parts, namely a artifact-free human tissue CT image and a metal artifact image. Aiming at the estimation of a human tissue CT image without metal artifact and a metal artifact layer, the application provides a constant-change near-end operator network, which can accurately reconstruct the human structure outline in a CT image and simultaneously eliminate the metal artifact affecting the diagnosis of a doctor. The rotation and other variable near-end operators can be applied to CT image reconstruction, can be popularized to more general application problems, such as natural image super-resolution and natural image rain removal, and have wide application value for various image restoration problems.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments of the present invention or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of an isomorphism near-end operator method suitable for CT deep learning reconstruction provided in an embodiment of the present application;
FIG. 2 is a flowchart illustrating an iterative update procedure according to an embodiment of the present application;
FIG. 3 is a graph comparing the results of a conventional network model structure provided in an embodiment of the present application with the results of a rotating isomorphic near-end operator total network model with a residual structure constructed in the present application;
FIG. 4 is a flowchart of an initial CT reconstructed neural network model corrupted by metal artifacts provided by an embodiment of the present application;
fig. 5 is a schematic diagram of a network structure of the 1 st iterative update step provided in the embodiment of the present application;
FIG. 6 is a schematic flow chart of the kth iteration step provided in the embodiment of the present application;
FIG. 7 is a schematic flow chart of the Kth iteration step according to the embodiment of the present application;
FIG. 8 is a graph comparing results on a DeepLesion dataset when the types of metal artifacts included in training and testing CT images provided in embodiments of the present application are the same;
FIG. 9 is a graph showing a comparison of generalized results on CLINIC-metal data sets when the types of metal artifacts included in training and testing CT images provided in embodiments of the present application are different;
fig. 10 is a schematic diagram of a restoration effect of an image super-resolution model according to an embodiment of the present application;
fig. 11 is a schematic diagram of a rain removal effect of the rain removal model provided in the embodiment of the present application;
FIG. 12 is a schematic diagram of a near-end operator device for CT deep learning processing and the like according to an embodiment of the present application;
fig. 13 is a schematic diagram of an isomorphic near-end operator server suitable for CT deep learning reconstruction according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. It will be apparent that the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
Some of the techniques involved in the embodiments of the present application are described below to aid understanding, and they should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, for the sake of clarity and conciseness, descriptions of well-known functions and constructions are omitted in the following description.
The embodiment of the application provides a constant-variation near-end operator method suitable for CT deep learning reconstruction, as shown in fig. 1, comprising steps S101 to S108. Wherein fig. 1 is only one execution order shown in the embodiments of the present application, and does not represent a unique execution order of an isokinetic near-end operator method suitable for CT deep learning reconstruction, and the steps shown in fig. 1 may be executed in parallel or in reverse, where the final result is achievable.
S101: and constructing a rotating and other variable near-end operator total network model with a residual error structure, and preprocessing a CT image before the constructed total network model. Wherein the total network model comprises a first sub-network model and a second sub-network model.
Specifically, the method for constructing the rotating and other variable near-end operator total network model with the residual error structure comprises the following steps: 1. convolution kernel parameterization: the convolution kernel is parametrized using a fourier function as a basis function. This can be achieved by applying a fourier transform to the convolution kernel, in such a way that it can be converted into a parameterized form. 2. Rotating group constant network construction: and constructing a rotating group isomorphism network, namely a rotating isomorphism near-end operator by utilizing the parameterized convolution kernel. Such networks have invariance to transformations such as rotation, translation, etc., and can better process image data with complex geometric transformations. 3. Introducing a residual structure: in constructing the network, a residual structure is introduced to approximate the solution of the near-end operator. The residual structure is that jump connection is introduced in the network, so that the network can learn complex features in the image better and the expression capability of the network is improved.
Specifically, fig. 3 is a comparison diagram of the conventional network model structure provided in the embodiment of the present application and the result of the rotating and other near-end operator total network model with residual error structure constructed in the present application, as shown in fig. 3, the left side is a structural diagram of the near-end operator network model constructed by the conventional convolutional neural network, and it can be seen from the structural diagram that the rotational symmetry priori of the CT image cannot be depicted. The right is a structure diagram of a rotating and other variable near-end operator total network model with a residual error structure constructed by the method, and the structure diagram can describe the rotation symmetry priori of a CT image.
S102: and acquiring the processed CT image data through a preprocessing step. The processed CT image data comprises a metal artifact free image, a metal artifact damaged image and a nonmetal area image.
Specifically, the preprocessing step may be to perform operations of cleaning, denoising, normalizing, etc. on the image, so as to enhance the quality and characteristics of the image. The process of acquiring processed CT image data through the preprocessing step is mainly for reconstructing an input image. A metal artifact free image is generally an image that is not disturbed or affected by a metal object and that provides relatively realistic tissue structure information. The image destroyed by the metal artifact refers to an image interfered or affected by the metal object, and the metal artifact may appear in the image, so that the accurate judgment of the tissue structure is affected. Nonmetallic area images refer to areas of the image that do not contain metallic objects, and these areas typically contain other types of tissue structure information.
S103: and establishing a metal artifact correction model corresponding to the image destroyed by the metal artifact, and constructing a reconstruction model of the metal artifact correction model to obtain a reconstruction optimization process. Wherein the metal artifact correction model is established according to a physical imaging mechanism of an image destroyed by metal artifacts.
The relation of the metal artifact correction model is as follows: i +.Y=i +.X+i +.M. Wherein Y εR H×W To destroy images by metal artifacts, X, M ε R H×W X is a metal artifact free image, M is a metal artifact layer to be estimated, I E R H×W For the non-metallic artifact region represented by the binary mask, +..
The total image reconstruction optimization process comprises the following steps:
wherein the method comprises the steps ofAnd->Respectively a metal artifact free image to be estimated and a metal artifact layer to be estimated, lambda 1 And lambda (lambda) 2 As the weight coefficient, f 1 (. Cndot.) and f 2 (. Cndot.) represents for->And->Regular term of->Representing the square norm. In particular, the potential metal artifact free image and metal artifact layer can be estimated from existing metal artifact corrupted images by solving the overall image reconstruction optimization process.
S104: and performing an iterative solving step on the total weight optimization process by using a near-end gradient descent method. The total reconstruction optimization process comprises a reconstruction optimization process of a metal artifact layer to be estimated and a reconstruction optimization process of a metal artifact-free image to be estimated.
Metal artifact layer to be estimatedThe reconstruction optimization process of (1) is as follows: /> Wherein M is (k) Solving the obtained updated result for the kth iteration, eta 1 For the step size of the first update, +.>For gradient operator->
Metal artifact free image to be estimatedThe reconstruction optimization process of (1) is as follows: /> Wherein Xk is an update result obtained by solving the kth iteration, eta 2 is a step length of the second update, and +.>For gradient operator->
S105: and acquiring a first reconstruction optimization result of the metal artifact layer to be estimated, and acquiring a second reconstruction optimization result of the metal artifact-free image to be estimated.
Metal artifact layer to be estimatedThe first reconstruction optimization result of (a) is: /> Wherein proxλ1η1 is a near-end operator with respect to f1.
Metal artifact free image to be estimatedThe second reconstruction optimization result of (2) is:
wherein,is about f 2 The near-end operator of (-).
Specifically, the overall image reconstruction optimization process consists of K iterative steps, M in the first iterative step (0) Initializing to be 0, firstly adopting a traditional CT reconstruction algorithm to pre-estimate the image Y damaged by metal artifact, and then adopting a simple ResNet to adjust the image Y as X (0) . In the kth (i.e. last) iteration step, the result X is output (K) By calculating X (k) K=1, …, K and true metal artifact free image X GT The back propagation of updated images gross parameters of the reconstruction optimization process.
S106: and the first sub-network model is brought into a first reconstruction optimization result, a first updating result is obtained, and the second sub-network model is brought into a second reconstruction optimization result, and a second updating result is obtained.
The first update result is:wherein (1)>For the first subnetwork model, θ 1 Is the first learnable parameter.
The second update results are:
wherein,for the second sub-network model, θ 2 Is a second learnable parameter.
S107: and constructing an initial CT reconstruction neural network according to the first updating result and the second updating result, inputting the metal artifact free image and the metal artifact damaged image obtained in the preprocessing step into the initial CT reconstruction neural network, and calculating a loss function. Specifically, S107 is a training step.
Further, calculating the loss function includes: and (3) carrying out iterative updating steps on initial parameters of the initial CT reconstruction neural network by adopting a random gradient descent method according to the loss obtained by calculating the loss function, so as to obtain the final CT reconstruction neural network.
The process disassembles the total image reconstruction optimization process into a series of simple and executable network modules, and establishes an initial CT reconstruction neural network. The whole initial CT reconstruction neural network consists of K iterative updating steps, wherein M is in the first iterative updating step 0 ) Initializing to 0, and performing a conventional CT reconstruction algorithm to obtain a metal artifact damaged image YA simple ResNet was adjusted as X (0) . In the Kth (i.e. last) iterative updating step, the result X belonging to the output (K) By calculating X (k) K=1, …, K, and true metal artifact free CT image X GT The back propagation updates the model parameters of the initial CT reconstructed neural network.
Specifically, fig. 4 is a flowchart of an initial CT reconstructed neural network model corrupted by metal artifacts provided in an embodiment of the present application, and is composed of K iterative update steps, where each iterative step includes an estimation network of metal artifactsAnd an estimation network X-net of clean CT images.
Specifically, fig. 5 is a schematic diagram of a network structure of the 1 st iterative update step provided in the embodiment of the present application. As shown in fig. 4, M initialized by the corrupted image estimate by metal artifacts (0) And X (0) For input, a first sub-network model comprising rotation proposed in the present applicationAnd a second subnetwork model->Wherein->And->Representing a learnable parameter of the network.
Specifically, fig. 6 is a schematic flow chart of a kth iteration step provided in an embodiment of the present application, where the flowchart includes a rotating constant near-end operator network proposed in the present application And->
Specifically, fig. 7 is a schematic flow chart of the kth iteration step provided in the embodiment of the present application, as shown in fig. 7, where the flowchart includes a first sub-network model proposed in the present applicationAnd a second subnetwork model->And a loss function section.
Specifically, for the near-end operator part included in the initial CT reconstruction neural network, a first sub-network model and a second sub-network model are designed and constructed according to a convolution kernel parameterization method. Specifically, the total network model includes a first sub-network model and a second sub-network model.
The calculation formula of the metal artifact free image loss function to be estimated is as follows: wherein XGT is a metal artifact free image, μk, α1 and α2 are weight parameters, and when k=0, X (0) is an initialized metal artifact free image, M (0) Is an initialized image corrupted by metal artifacts.
Specifically, when the number of iterative steps reaches a first preset threshold, stopping the iterative updating step, and storing the final parameters at the moment, thereby obtaining the final CT reconstruction neural network. The first preset threshold is the set maximum iteration step number.
Fig. 2 is a specific flowchart of the iterative updating step provided in the embodiment of the present application, as shown in fig. 2, including steps S201 to S205.
S201: and scaling the pixel values of the image without metal artifact and the image destroyed by metal artifact to between 0 and 1 after the preprocessing step.
S202: the pixel value of the non-metal area image is set to 1 when the pixel value is higher than a second preset threshold value, and is set to 0 when the pixel value is lower than the second preset threshold value.
S203: the image corrupted by the metal artifact is input into a reconstruction network and the first metal artifact layer is estimated using the initialized metal artifact layer and the initialized metal artifact free image.
S204: and estimating the first metal artifact free image by using the obtained first metal artifact layer, the initialized metal artifact free image and the metal artifact damaged image, thereby completing an iterative updating step.
Further, an X is estimated from the corrupted image by metal artifacts using conventional CT reconstruction algorithms (0) Let M again (0) All 0 s. Then X is taken up (0) And M (0) To iteration 1And X-net to obtain X (1) And M (1) . The following K-th iteration step k=2, …, K-1 is continued in sequence until the K-th, and then the estimation result X is obtained (K) . Then iteratively updating the intermediate result X of the step (k) ，X (K) And X GT To calculate the loss function.
S205: repeating the above process until the iterative updating step is completed.
S108: and in the testing stage, transmitting the image to be tested, damaged by the metal artifact, into a final CT reconstruction neural network to perform a forward inference step, and outputting the reconstructed image without the metal artifact.
Specifically, S108 is a test step, specifically: normalizing the updated model parameters of the initial CT reconstruction neural network, namely the model parameters of the final CT reconstruction neural network to be between 0 and 1, and estimating an X by using a traditional CT reconstruction algorithm according to the model parameters of the final CT reconstruction neural network (0) Let M (0) Estimating an I according to the value of the model parameter, and performing forward calculation to obtain an output result X (K) Namely, the image without metal artifact.
TABLE 1
Specifically, table 1 is a comparative experiment of different comparative methods on the deepleion dataset. The rotating and other variable near-end operator total network model with the residual structure constructed by the method can be applied to a plurality of latest CT (computed tomography) deep learning reconstruction models ACDNet, DICDNet, OSCNet based on depth expansion, and after the near-end operator network in the methods is replaced by the rotating and other variable near-end operator network proposed by the method, ACDNet-E can be obviously observed from the table 1 X 、DICDNet-E X 、OSCNet-E X 、/>The numerical accuracy (PSNR, SSIM) of the deep version test dataset is significantly improved, and it should be noted that, the bolded representation in table 1 brings improved results after using the rotation with residual structure and the like constructed in the present application to become the total network model of the near-end operator.
TABLE 2
Specifically, table 2 is a comparative experiment of different comparative methods on the Urban100, BSD100, set14, set5 datasets. The rotation and other near-end operator total network model with the residual structure constructed by the method is a general tool, so that the model can be applied to other practical problems. For example, for the problem of image super resolution, in recent years, an image super resolution model KXNet based on depth expansion has also been developed, so the total network model proposed in the application is embedded into the model and then recorded as KXNet-E X FIG. 10 is a diagram illustrating super resolution of an image according to an embodiment of the present applicationAs shown in fig. 10, the restoration effect of the super-resolution image model can be further improved. Meanwhile, as can be seen from table 2, the numerical accuracy (PSNR, SSIM) is further improved by embedding the image super-resolution model of the rotating and other variable near-end operator networks. The bolded results in table 2 represent the results of the improvement obtained by using the rotation with residual structure and the like constructed in the present application to become the total network model of the near-end operator.
TABLE 3 Table 3
Specifically, table 3 shows comparative experiments on the data sets disclosed in Rain100L, rain100H, rain, rain12 for different comparative methods. In the field of natural image rain removal which has attracted much attention in recent years, an image rain removal model RCDNet based on depth expansion has also been developed, so that a rotational and other variable near-end operator total network model with a residual structure constructed by the method is embedded into the model and then recorded asFig. 11 is a schematic diagram of a rain removal effect of a rain removal model according to an embodiment of the present application, as shown in fig. 11, the rain removal effect of the original model can be further improved, and meanwhile, a rain strip can be estimated more accurately. Meanwhile, as can be seen from table 3, the numerical accuracy (PSNR, SSIM) is further improved by embedding the image rain removal model of the rotating and other variable near-end operator networks. Incidentally, the results of the first name in the comparative method are shown by bold in Table 3.
Fig. 8 is a graph comparing results on the deep version dataset when the types of metal artifacts included in the training and testing CT images provided in the embodiments of the present application are the same. Where Input represents a corrupted image by metal artifacts and group Truth represents its corresponding artifact free image. CNNMAR, DSCMAR, duDoNet, inDuDoNet are respectively the newly proposed deep learning methods in recent years. ACDNet, DICDNet, OSCNet is a recently proposed depth-expansion-based CT deep learning reconstruction model, which is applied to After the near-end operator network in the methods is replaced by the rotating and other near-end operator total network model with the residual error structure constructed by the method, ACDNet-E can be obviously observed X 、DICDNet-E X 、OSCNet-E X 、The restored visual effect of (a) is significantly improved.
Fig. 9 is a generalized result comparison chart on a CLINIC-metal dataset when the types of metal artifacts included in training and testing CT images provided in the embodiments of the present application are different. ACDNet, DICDNet, OSCNet is a recently proposed CT deep learning reconstruction model based on deep expansion, and ACDNet-E can be obviously observed after the near-end operator network in the methods is replaced by a rotating and other near-end operator total network model with a residual structure constructed by the application X 、DICDNet-E X 、OSCNet-E X 、/>The restored visual effect of (2) is also significantly improved.
For the reconstruction model based on depth expansion in a plurality of application problems, after the reconstruction model with residual error structure built by the application is embedded into the total network model of the rotating and other variable near-end operators, the performance of the application results are improved, including numerical precision and visual effect. From the above, it is obvious that the present application brings about improvement in a plurality of data sets in a plurality of application problems, and the superior performance and the good generalization ability of the present application are embodied.
The application makes fair quantitative comparison with the current representative technology under the condition that the training is consistent/inconsistent with the test setting on different application problems and different data sets, and performs visual display on the experimental result of the application, so that the advantages and reasonable feasibility of the application are fully verified.
The embodiment of the present application further provides an isomorphism near-end operator device 1200 suitable for CT deep learning reconstruction, as shown in fig. 12, the device includes: the system comprises a preprocessing module 1201, a data preprocessing module 1202, an obtained total weight optimization process module 1203, an iterative solution module 1204, an obtained reconstructed optimization result module 1205, an obtained updated result module 1206, a calculation module 1207 and an output module 1208.
The preprocessing module 1201 is configured to construct a total network model of a rotating and other variable near-end operators with a residual structure, and perform a preprocessing step on a CT image before the constructed total network model. Wherein the total network model comprises a first sub-network model and a second sub-network model.
The data preprocessing module 1202 is configured to acquire processed CT image data through a preprocessing step. The processed CT image data comprises a metal artifact free image, a metal artifact damaged image and a nonmetal area image.
The total reconstruction optimization process module 1203 is configured to establish a metal artifact correction model corresponding to the image damaged by the metal artifact, and perform reconstruction model construction on the metal artifact correction model to obtain a total reconstruction optimization process. Wherein the metal artifact correction model is established according to a physical imaging mechanism of an image destroyed by metal artifacts.
The iteration solution module 1204 is configured to perform an iteration solution step on the total weight optimization process using a near-end gradient descent method. The total reconstruction optimization process comprises a reconstruction optimization process of a metal artifact layer to be estimated and a reconstruction optimization process of a metal artifact-free image to be estimated.
The acquire reconstruction optimization result module 1205 is configured to acquire a first reconstruction optimization result of the metal artifact layer to be estimated, and acquire a second reconstruction optimization result of the metal artifact free image to be estimated.
The update result obtaining module 1206 is configured to bring the first sub-network model into the first reconstruction optimization result, obtain the first update result, and bring the second sub-network model into the second reconstruction optimization result, obtain the second update result.
The computing module 1207 is configured to construct an initial CT reconstruction neural network according to the first update result and the second update result, input the metal artifact free image and the metal artifact damaged image obtained in the preprocessing step into the initial CT reconstruction neural network, and calculate the loss function. Wherein calculating the loss function comprises: and carrying out iterative updating steps on initial parameters of the initial CT reconstruction neural network by adopting a random gradient descent method according to the loss obtained by calculating the loss function, so as to obtain the final CT reconstruction neural network.
The output module 1208 is configured to transmit, in a testing stage, an image to be tested, which is damaged by metal artifacts, to a final CT reconstructed neural network for performing a forward inference step, and output a reconstructed image without metal artifacts.
Some of the modules of the apparatus described herein may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The apparatus or module set forth in the embodiments of the application may be implemented in particular by a computer chip or entity, or by a product having a certain function. For convenience of description, the above devices are described as being functionally divided into various modules, respectively. The functions of the modules may be implemented in the same piece or pieces of software and/or hardware when implementing the embodiments of the present application. Of course, a module that implements a certain function may be implemented by a plurality of sub-modules or a combination of sub-units.
The methods, apparatus or modules described herein may be implemented in computer readable program code means and in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, application specific integrated circuits (english: application Specific Integrated Circuit; ASIC for short), programmable logic controllers and embedded microcontrollers, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in a pure computer readable program code, it is well possible to implement the same functionality by logically programming the method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Such a controller can be regarded as a hardware component, and means for implementing various functions included therein can also be regarded as a structure within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.
As shown in fig. 13, the embodiment of the present application further provides an isomorphic near-end operator server suitable for CT deep learning reconstruction, including a memory 1301 and a processor 1302; memory 1301 is used to store computer executable instructions; the processor 1302 is configured to execute computer-executable instructions to implement an isomorphic near-end operator method suitable for CT deep learning reconstruction as described in the embodiments of the present application.
The embodiment of the application also provides a computer readable storage medium, wherein the computer readable storage medium stores executable instructions, and the computer can realize the isomorphous near-end operator method suitable for CT deep learning reconstruction when executing the executable instructions.
From the description of the embodiments above, it will be apparent to those skilled in the art that the present application may be implemented in software plus necessary hardware. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product, or may be embodied in the implementation of data migration. The computer software product may be stored on a storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and include several instructions for causing a computer device (which may be a personal computer, mobile terminal, server, or network device, etc.) to perform the methods described in the embodiments of the present application.
In this specification, each embodiment is described in a progressive manner, and the same or similar parts of each embodiment are referred to each other, and each embodiment is mainly described as a difference from other embodiments. All or portions of the present application can be used in a number of general purpose or special purpose computer system environments or configurations.
The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the present application; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced with equivalents; such modifications and substitutions do not depart from the spirit of the corresponding technical solutions.
Claims (10)
1. An isomorphism near-end operator method suitable for CT deep learning reconstruction, which is characterized by comprising the following steps:
constructing a rotating and other variable near-end operator total network model with a residual error structure, and preprocessing a CT image before the constructed total network model; wherein the total network model comprises a first sub-network model and a second sub-network model;
Acquiring processed CT image data through the preprocessing step; the processed CT image data comprise a metal artifact free image, a metal artifact damaged image and a nonmetal area image;
establishing a metal artifact correction model corresponding to an image damaged by metal artifacts, and constructing a reconstruction model of the metal artifact correction model to obtain a total reconstruction optimization process; the metal artifact correction model is established according to a physical imaging mechanism of an image damaged by metal artifacts;
performing an iterative solution step on the total weight optimization process by using a near-end gradient descent method; the total reconstruction optimization process comprises a reconstruction optimization process of a metal artifact layer to be estimated and a reconstruction optimization process of a metal artifact-free image to be estimated;
acquiring a first reconstruction optimization result of a metal artifact layer to be estimated, and acquiring a second reconstruction optimization result of a metal artifact-free image to be estimated;
the first sub-network model is brought into a first reconstruction optimization result, a first updating result is obtained, the second sub-network model is brought into a second reconstruction optimization result, and a second updating result is obtained;
constructing an initial CT reconstruction neural network according to the first updating result and the second updating result, inputting the metal artifact free image and the metal artifact damaged image obtained in the preprocessing step into the initial CT reconstruction neural network, and calculating a loss function;
Wherein the calculating the loss function comprises: according to the loss obtained by calculating the loss function, performing iterative updating steps on initial parameters of the initial CT reconstruction neural network by adopting a random gradient descent method to obtain a final CT reconstruction neural network;
and in the testing stage, transmitting the image to be tested, damaged by the metal artifact, into a final CT reconstruction neural network to perform a forward inference step, and outputting the reconstructed image without the metal artifact.
2. The method according to claim 1, wherein the creating a metal artifact correction model corresponding to the image corrupted by the metal artifact, performing reconstruction model construction on the metal artifact correction model, and obtaining a total reconstruction optimization process includes:
the relation of the metal artifact correction model is as follows: i +.y=i +.x+i +.m; wherein Y εR H×W To destroy images by metal artifacts, X, M ε R H×W X is an image without metal artifact, M is a metal to be estimatedArtifact layer, I.epsilon.R H×W For the non-metallic artifact region represented by the binary mask, +.;
the total image reconstruction optimization process comprises the following steps:
wherein the method comprises the steps ofAnd->Respectively a metal artifact free image to be estimated and a metal artifact layer to be estimated, lambda 1 And lambda (lambda) 2 As the weight coefficient, f 1 (. Cndot.) and f 2 (. Cndot.) represents for->And->Regular term of->Representing the square norm.
3. The method according to claim 1, wherein the reconstruction optimization process of the metal artifact layer to be estimated and the reconstruction optimization process of the metal artifact free image to be estimated comprise:
metal artifact layer to be estimatedThe reconstruction optimization process of (1) is as follows: /> Wherein MK is an update result obtained by solving the kth iteration, eta 1 is a step length of the first update, and +.>For gradient operator, g1Mk=IXk-1+I+Mk-I.YF2;
metal artifact free image to be estimatedThe reconstruction optimization process of (1) is as follows: /> Wherein Xk is an update result obtained by solving the kth iteration, eta 2 is a step length of the second update, and +.>For the purpose of the gradient operator,
4. a method according to claim 3, wherein the obtaining a first reconstruction optimization result of the metal artifact layer to be estimated and obtaining a second reconstruction optimization result of the metal artifact free image to be estimated;
metal artifact layer to be estimatedThe first reconstruction optimization result of (a) is: /> Wherein proxλ1η1 is the near-end operator for f1;
metal artifact free image to be estimatedThe second reconstruction optimization result of (2) is:
Wherein,is about f 2 The near-end operator of (-).
5. The method of claim 1, wherein the bringing the first sub-network model into the first reconstruction optimization result, obtaining the first updated result, and the bringing the second sub-network model into the second reconstruction optimization result, obtaining the second updated result, comprises:
the first update result is:wherein (1)>For the first subnetwork model, θ 1 Is a first learnable parameter;
the second update results are:
wherein,for the second sub-network model, θ 2 Is a second learnable parameter.
6. The method of claim 1, wherein the loss function is calculated as:
wherein X is GT Mu, as a metal artifact free image k 、α 1 And alpha 2 As a weight parameter, when k=0, X (0) For initialized metal artifact free images, M (0) Is an initialized image corrupted by metal artifacts.
7. The method according to claim 1, wherein the step of iteratively updating initial parameters of the initial CT reconstruction neural network by a random gradient descent method based on the loss obtained by calculating the loss function to obtain a final CT reconstruction neural network comprises:
stopping the iterative updating step when the iterative step number reaches a first preset threshold value, and storing the final parameters at the moment, so as to obtain a final CT reconstruction neural network;
The iterative updating step comprises the following steps:
scaling the pixel values of the image without metal artifact and the image destroyed by metal artifact after the preprocessing step to between 0 and 1;
setting the pixel value of the non-metal area image to 1 when the pixel value is higher than a second preset threshold value, and setting the pixel value of the non-metal area image to 0 when the pixel value is lower than the second preset threshold value;
inputting the image destroyed by the metal artifact into a reconstruction network, and estimating a first metal artifact layer by using the initialized metal artifact layer and the initialized metal artifact-free image;
estimating a first metal artifact free image using the obtained first metal artifact layer, the initialized metal artifact free image and the metal artifact corrupted image, thereby completing an iterative updating step;
repeating the above process until the iterative updating step is completed.
8. An isomorphic near-end operator device suitable for deep learning reconstruction, comprising:
the preprocessing module is used for constructing a rotating and other near-end operator total network model with a residual error structure, and preprocessing CT images before the constructed total network model; wherein the total network model comprises a first sub-network model and a second sub-network model;
The data preprocessing module is used for acquiring preprocessed CT image data through the preprocessing step; the preprocessed CT image data comprise a metal artifact free image, a metal artifact damaged image and a nonmetal area image;
the method comprises the steps of obtaining a total reconstruction optimization process module, wherein the total reconstruction optimization process module is used for establishing a metal artifact correction model corresponding to an image damaged by metal artifacts, and carrying out reconstruction model construction on the metal artifact correction model to obtain a total reconstruction optimization process; the metal artifact correction model is established according to a physical imaging mechanism of an image damaged by metal artifacts;
the iteration solution module is used for carrying out an iteration solution step on the total weight optimization process by using a near-end gradient descent method; the total reconstruction optimization process comprises a reconstruction optimization process of a metal artifact layer to be estimated and a reconstruction optimization process of a metal artifact-free image to be estimated;
the reconstruction optimization result acquisition module is used for acquiring a first reconstruction optimization result of the metal artifact layer to be estimated and acquiring a second reconstruction optimization result of the metal artifact-free image to be estimated;
the updating result acquisition module is used for bringing the first sub-network model into a first reconstruction optimization result, acquiring a first updating result, bringing the second sub-network model into a second reconstruction optimization result, and acquiring a second updating result;
The computing module is used for constructing an initial CT reconstruction neural network according to the first updating result and the second updating result, inputting the metal artifact free image and the metal artifact damaged image obtained in the preprocessing step into the initial CT reconstruction neural network, and computing a loss function; wherein the calculating the loss function comprises: according to the loss obtained by calculating the loss function, performing iterative updating steps on initial parameters of the initial CT reconstruction neural network by adopting a random gradient descent method to obtain a final CT reconstruction neural network;
and the output module is used for transmitting the image to be tested, damaged by the metal artifact, into the final CT reconstruction neural network to perform a forward inference step in the test stage, and outputting the reconstructed image without the metal artifact.
9. The constant-variation near-end operator server suitable for CT deep learning reconstruction is characterized by comprising a memory and a processor;
the memory is used for storing computer executable instructions;
the processor is configured to execute the computer-executable instructions to implement the method of any of claims 1-7.
10. A computer readable storage medium storing executable instructions which when executed by a computer enable the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311717912.6A CN117765110A (en) | 2023-12-14 | 2023-12-14 | Equal-transformation near-end operator method and device suitable for CT deep learning reconstruction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311717912.6A CN117765110A (en) | 2023-12-14 | 2023-12-14 | Equal-transformation near-end operator method and device suitable for CT deep learning reconstruction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117765110A true CN117765110A (en) | 2024-03-26 |
Family
ID=90317388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311717912.6A Pending CN117765110A (en) | 2023-12-14 | 2023-12-14 | Equal-transformation near-end operator method and device suitable for CT deep learning reconstruction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117765110A (en) |
-
2023
- 2023-12-14 CN CN202311717912.6A patent/CN117765110A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110728729B (en) | Attention mechanism-based unsupervised CT projection domain data recovery method | |
| CN111429379B (en) | Low-dose CT image denoising method and system based on self-supervision learning | |
| WO2023202265A1 (en) | Image processing method and apparatus for artifact removal, and device, product and medium | |
| Li et al. | MRI reconstruction with interpretable pixel-wise operations using reinforcement learning | |
| CN111127579A (en) | CT reconstruction method for removing cone beam artifacts based on neural network | |
| CN108038840B (en) | Image processing method and device, image processing equipment and storage medium | |
| CN111210431B (en) | Blood vessel segmentation method, device, equipment and storage medium | |
| Savanier et al. | Deep unfolding of the dbfb algorithm with application to roi ct imaging with limited angular density | |
| CN110599530A (en) | MVCT image texture enhancement method based on double regular constraints | |
| CN112991220A (en) | Method for correcting image artifacts by convolutional neural network based on multiple constraints | |
| JP2021065707A (en) | Medical image processing device, learned model and medical image processing method | |
| CN109685803B (en) | Left ventricle image segmentation method, device, equipment and storage medium | |
| US11308660B2 (en) | Motion compensated cardiac valve reconstruction | |
| CN117765110A (en) | Equal-transformation near-end operator method and device suitable for CT deep learning reconstruction | |
| Tran et al. | Deep learning-based inpainting for chest X-ray image | |
| Jin et al. | Low-dose CT image restoration based on noise prior regression network | |
| US20220292673A1 (en) | On-Site training of a machine-learning algorithm for generating synthetic imaging data | |
| Yan et al. | Calcium removal from cardiac CT images using deep convolutional neural network | |
| Trbalić et al. | CT metal artefacts reduction using convolutional neural networks | |
| Zhu et al. | CT metal artifact correction assisted by the deep learning-based metal segmentation on the projection domain | |
| CN116823660B (en) | Construction method, device and application of double-flow network model for CT image restoration | |
| CN117274080B (en) | Low-dose CT chord graph recovery method and related device | |
| KR102490586B1 (en) | Repetitive Self-supervised learning method of Noise reduction | |
| US20240062438A1 (en) | Systems and methods for processing medical images with invertible neural networks | |
| CN116977364A (en) | Image processing method, device, electronic equipment and computer storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |