CN117541481A - Low-dose CT image restoration method, system and storage medium - Google Patents

Low-dose CT image restoration method, system and storage medium Download PDF

Info

Publication number
CN117541481A
CN117541481A CN202410028983.3A CN202410028983A CN117541481A CN 117541481 A CN117541481 A CN 117541481A CN 202410028983 A CN202410028983 A CN 202410028983A CN 117541481 A CN117541481 A CN 117541481A
Authority
CN
China
Prior art keywords
image
dose
low
model
neural network
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.)
Granted
Application number
CN202410028983.3A
Other languages
Chinese (zh)
Other versions
CN117541481B (en
Inventor
邹利兰
叶爱萍
林聪�
李超传
毛鑫
蔡东耀
陈悦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Ocean University
Original Assignee
Guangdong Ocean University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangdong Ocean University filed Critical Guangdong Ocean University
Priority to CN202410028983.3A priority Critical patent/CN117541481B/en
Publication of CN117541481A publication Critical patent/CN117541481A/en
Application granted granted Critical
Publication of CN117541481B publication Critical patent/CN117541481B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Processing (AREA)

Abstract

The invention relates to the technical field of image processing, in particular to a low-dose CT image restoration method, a system and a storage medium, wherein the method comprises the following steps: constructing a deep neural network, inputting a data set into the deep neural network for training to obtain key information characteristics of a low-dose CT image; denoising key information features of the low-dose CT image through an improved denoising device to obtain a reconstructed image of the low-dose CT image, and further determining loss functions of the low-dose CT image and the normal-dose CT image; constructing an image restoration model and establishing a solving model of the image restoration model; performing iterative training on the deep neural network based on the solving model to obtain a low-dose CT image restoration model; acquiring a low-dose CT image to be repaired, inputting the low-dose CT image to be repaired into a low-dose CT image repair model, and obtaining a repaired CT image; the invention can improve the CT image restoration quality with low radiation dose.

Description

Low-dose CT image restoration method, system and storage medium
Technical Field
The invention relates to the technical field of image processing, in particular to a low-dose CT image restoration method, a system and a storage medium.
Background
With the increasing popularity of medical CT, the influence of X-rays on the human body is of great concern. High-dose CT examination does cause a certain injury to the human body. The most common method is to reduce the X-ray dose. However, reducing the X-ray dose reduces the tube current, and low dose CT (low dose CT) techniques can receive fewer photons by the detector. This means that the lower the X-ray flux, the more severely the quality of the CT image is affected, which reduces the quality of the CT image and may affect the diagnostic performance. Therefore, research to reduce radiation dose and maintain CT image quality has important research value and clinical significance, and this work has attracted great attention in the field of medical imaging. While most existing deep learning neural networks solve the problem by directly mapping low quality images to ideal normal quality images, the observation model that characterizes the image degradation process is largely ignored.
Therefore, how to reduce the radiation dose while improving the CT image quality is a problem to be solved.
Disclosure of Invention
The invention aims to provide a low-dose CT image restoration method, a system and a storage medium, which can improve the quality of CT image restoration with low radiation dose.
In order to achieve the above object, the present invention provides the following technical solutions:
in a first aspect, an embodiment of the present invention provides a low dose CT image restoration method, including the steps of:
s100, acquiring a data set, wherein the data set comprises a normal dose CT image and a corresponding low dose CT image;
s200, constructing a deep neural network for repairing a low-dose CT image, inputting the data set into the deep neural network for training, and obtaining key information features of the low-dose CT image;
s300, denoising key information features of the low-dose CT image through an improved denoising device to obtain a reconstructed image of the low-dose CT image, and determining loss functions of the low-dose CT image and a normal-dose CT image based on the reconstructed image;
s400, constructing an image restoration model and establishing a solving model of the image restoration model;
s500, performing iterative training on the deep neural network based on the solving model until a loss function is lower than a set loss threshold value, so as to obtain a low-dose CT image restoration model;
s600, acquiring a low-dose CT image to be repaired, and inputting the low-dose CT image to be repaired into the low-dose CT image repair model to obtain a repaired CT image.
Optionally, in S200, the constructing a deep neural network for repairing a low dose CT image, inputting the data set into the deep neural network for training, to obtain key information features of the low dose CT image, includes:
s210, constructing a deep neural network, and establishing a mapping relation model from a normal dose CT image to a low dose image based on the deep neural network, wherein the mapping relation model is as follows:
;
wherein,for low dose CT images, < >>For normal dose CT image, < >>For the repaired image, ++>Representing deconvolution in the deep neural network, namely mapping relation from a normal dose CT image to a low dose CT image;
s220, inputting the data set into the deep neural network for training by adopting a batch training method, and obtaining key information characteristics of the low-dose CT image through the constructed spatial attention module, the channel attention module and the spatial channel mixed attention module.
Optionally, in S300, the denoising, by using an improved denoising device, the key information feature of the low dose CT image, to obtain a reconstructed image of the low dose CT image, includes:
s310, constructing a denoising device based on an improved depth neural network model, wherein a series of fully-connected convolution layers are used as a stacked encoder, and a stacked decoder is formed by using the series of fully-connected deconvolution layers;
s320, denoising the low-dose CT image through the denoising device, and extracting key information from key information features of the low-dose CT image to obtain a reconstructed image of the low-dose CT image.
Optionally, the loss function of the low dose CT image and the normal dose CT image is calculated by the following formula:
;
wherein,representing normal dose CT image, < >>) Representing reconstructed image +.>Representing the i-th low dose CT image,represents the ith normal dose CT image in the dataset, and N represents the total number of low dose CT images in the dataset.
Optionally, in S400, the constructing an image restoration model for image restoration, converting the image restoration model into a solution model, includes:
s410, constructing an image restoration model, wherein the expression of the image restoration model is as follows:
;
wherein,is an auxiliary variable, +.>Representing the observation image +.>Representing the original image +.>Representing a linear operator +.>Representing a regularization factor associated with the prior, λ being a parameter weighted to the regularization term;
s420, converting the solved image recovery model into a constraint optimization problem, wherein the expression of the constraint optimization problem is as follows:
;
wherein,is a penalty parameter;
s430, converting the constraint optimization problem into two sub-problems using alternative ADMM technology:
;
where v denotes the augmented lagrangian multiplier, k denotes the time interval parameter,represents the augmented lagrangian multiplier at time k,/>Representing a low dose CT image at the next instant of k,/and (b)>Low dose CT image representing time k, +.>Represents +.1 at time k->Sub-problems (S)>Representing +.>Sub-problems; />The sub-problem is also called +.>A calculation problem of the proximity operator of (a) at point u;
s440, establishing by using Newton-Raphson iterative algorithm with integral termSolution model of->The solution model of (2) is:
;
wherein,representing the identity matrix;
will beThe solution model of (c) is rewritten as:
;
wherein, gamma is a scale factor, gamma >0, E is an integral term, and W represents a characteristic weight matrix value of the image;
;
wherein,is a steady state error, +.>、/>And->Respectively representing different system parameters;
the solution model of (2) is:
wherein->Representing the denoising means.
Optionally, in S500, the performing iterative training on the deep neural network based on the solution model until a loss function is lower than a set loss threshold value, to obtain a low-dose CT image restoration model, including:
building the data set into a training setAnd test set->During training, the training set is treated with batch training method>According to the batch size->The input data divided into a plurality of batches are sequentially input into a deep neural network for training, and normal dose CT images are obtained after the input data of each batch pass through the deep neural network;
integrating the test setAccording to the batch size->The test data divided into a plurality of batches are sequentially input into a deep neural network for training, and the test data of each batch is +.>Calculating model loss of the deep neural network by using the loss function;
and updating the back propagation parameters of the deep neural network based on the model loss, and performing iterative training on the deep neural network by using a solving model until the loss function is determined to be lower than a set loss threshold value, so as to obtain a low-dose CT image restoration model.
In a second aspect, embodiments of the present invention provide a low dose CT image restoration system, the system comprising:
at least one processor;
at least one memory for storing at least one program;
the at least one program, when executed by the at least one processor, causes the at least one processor to implement a low dose CT image restoration method as set forth in any one of the preceding claims.
In a third aspect, an embodiment of the present invention provides a computer readable storage medium, in which a processor executable program is stored, wherein the processor executable program when executed by a processor is configured to perform a low dose CT image restoration method as set forth in any one of the preceding claims.
The beneficial effects of the invention are as follows: the invention discloses a low-dose CT image restoration method, a system and a storage medium.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a low dose CT image restoration method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a denoising structure according to an embodiment of the present invention;
FIG. 3a is a schematic diagram of an attention module according to an embodiment of the present invention;
FIG. 3b is a schematic diagram of a feature encoder architecture according to an embodiment of the present invention;
FIG. 3c is a schematic diagram of a feature decoder according to an embodiment of the present invention;
FIG. 4a is a normal dose CT image in an embodiment of the present invention;
FIG. 4b is a low dose CT image in an embodiment of the present invention;
FIG. 4c is the reconstructed CT image of FIG. 4 b;
fig. 5 is a schematic structural diagram of a low dose CT image restoration system according to an embodiment of the present invention.
Detailed Description
The conception, specific structure, and technical effects produced by the present invention will be clearly and completely described below with reference to the embodiments and the drawings to fully understand the objects, aspects, and effects of the present invention. It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other.
Referring to fig. 1, fig. 1 is a low dose CT image restoration method according to the present invention, the method includes the following steps:
s100, acquiring a data set, wherein the data set comprises a Normal dose CT (Normal-dose computed tomography) image and a corresponding low dose CT (low-dose computed tomography) image;
specifically, standard data of CT images are acquired and used for training the neural network related to the invention, normal dose CT images are set to be the same size, after the training set is enhanced, the data set is divided into a training set and a testing set, and corresponding low dose CT images are generated through fan beam projection transformation, so that noise meets Poisson distribution.
In one embodiment, the data in the disclosed cancer medical image database (TCIA) is used to divide its data set into a training set and a test set for network parameter training and final result testing, respectively. And 8:2, dividing the training set and the test set proportionally, and setting all images as
512×512, and performing rotation, left-right overturn, up-down overturn and normalization image preprocessing operation on the remote sensing images in the training set before training.
S200, constructing a deep neural network for repairing a low-dose CT image, inputting the data set into the deep neural network for training, and obtaining key information features of the low-dose CT image;
s300, denoising key information features of the low-dose CT image through an improved denoising device to obtain a reconstructed image of the low-dose CT image, and determining loss functions of the low-dose CT image and a normal-dose CT image based on the reconstructed image;
s400, constructing an image restoration model and establishing a solving model of the image restoration model;
s500, performing iterative training on the deep neural network based on the solving model until a loss function is lower than a set loss threshold value, so as to obtain a low-dose CT image restoration model;
in the embodiment, the solution model effectively helps the deep neural network to quickly iterate and converge, and improves the robustness of the deep neural network.
S600, acquiring a low-dose CT image to be repaired, and inputting the low-dose CT image to be repaired into the low-dose CT image repair model to obtain a repaired CT image.
Specifically, after a normal dose CT image is obtained, preprocessing the CT image to obtain a standard image; the CT image is set to a set size, for example, 512 x 512 size, so as to adapt the subsequent scene recognition model.
Compared with the existing scene recognition method, the method provided by the invention has the advantages that the integral term exists in the model, the robustness of the model under a large amount of noise interference can be ensured, and the influence of external interference in the image recovery process can be overcome. The deployment application in the actual scene is more convenient.
In a preferred embodiment, in S200, the constructing a deep neural network for repairing a low dose CT image, inputting the data set into the deep neural network for training, and obtaining key information features of the low dose CT image includes:
s210, constructing a deep neural network, and establishing a mapping relation model from a normal dose CT image to a low dose image based on the deep neural network, wherein the mapping relation model is as follows:
；
wherein,for low dose CT images, < >>For normal dose CT image, < >>For the repaired image, ++>Representing deconvolution in the deep neural network, namely mapping relation from a normal dose CT image to a low dose CT image;
in some embodiments, the improved deep convolutional neural network model can meet the requirement of image restoration due to the characteristic of the convolutional mapping image of the deep convolutional neural network, and the deep convolutional neural network is excellent in image restoration. In addition, the depth convolution neural network does not depend on the statistical distribution of image noise, so that the problem of low-illumination image recovery can be simplified, and the mathematical model of the problem of low-illumination image recovery is as follows:wherein->As a noise function, i.e.)>Is an image contaminated with noise. It should be noted that, for networks with more than two layers, any function can be fitted by different parameters, in this embodiment +.>Is->Is a good approximation of the best approximation of (a).
S220, inputting the data set into the deep neural network for training by adopting a batch training method, and obtaining key information characteristics of the low-dose CT image through the constructed spatial attention module, the channel attention module and the spatial and channel mixed attention module.
In a preferred embodiment, in S300, denoising key information features of the low dose CT image by using an improved denoising device to obtain a reconstructed image of the low dose CT image, including:
s310, constructing a denoising device based on a modified deep neural network (Modified deep neural network), wherein a series of fully-connected convolutional layers are used as a stacked encoder, and a stacked decoder is formed using the series of fully-connected deconvolution layers;
s320, denoising the low-dose CT image through the denoising device, and extracting key information from key information features of the low-dose CT image to obtain a reconstructed image of the low-dose CT image.
In some embodiments, a denoising device is constructed by adopting an improved depth neural network model, and an encoder is formed by a convolution layer and a ReLU unit; reconstructing an image by a decoder; and denoising the low-dose CT image by using a denoising device, further retaining key information in patches gradually extracted from low level to high level, and gradually restraining image noise and artifacts.
In a preferred embodiment, the loss functions of the low dose CT image and the normal dose CT image are calculated by the following formula
；
Wherein,representing normal dose CT image, < >>) Representing reconstructed image +.>Representing the i-th low dose CT image,representing the ith sheet in the datasetThe normal dose CT images, i, represent the numbers of the normal dose CT images and the corresponding low dose CT images in the data set, and N represents the total number of the normal dose CT images in the data set and also the total number of the low dose CT images in the data set.
In this embodiment, the loss between the CT image and the reference normal dose CT image is determined by a minimum estimation method. The Mean Square Error (MSE) is used as a loss function to amplify the difference and improve the optimized denoiser.
In a preferred embodiment, in S400, the constructing an image restoration model for image restoration, converting the image restoration model into a solution model, includes:
s410, constructing an image restoration model, wherein the expression of the image restoration model is as follows:；
wherein,is an auxiliary variable, +.>Representing the observation image +.>Representing the original image +.>Representing a linear operator +.>Representing a regularization factor associated with the prior, λ being a parameter weighted to the regularization term;
in some embodiments, the image restoration problem is represented as a mathematical model by decoupling the accuracy term and regularization term using variable separation techniques:
wherein (1)>Representing an image to be restored->，/>Representing a linear operator +.>U represents the restored image, u ∈ ->，/>Representing noise->。/>Can be, for example, a convolution operator in image deblurring or a fourier operator in CT image reconstruction, a being a pathological matrix, it is difficult to recover from the image to be restored->And the restored image u is obtained. The present embodiment is implemented by introducing the auxiliary variable +.>Mathematical model of the problem of restoration of images +.>Conversion to an image restoration model by the formula +.>An estimated original image can be obtained, again by the formula +.>Regular optimization to suppress image noise. S420, converting the solved image recovery model into a constraint optimization problem, wherein the expression of the constraint optimization problem is as follows:
；
wherein,is a penalty parameter;
s430, converting the constraint optimization problem into two sub-problems using alternative ADMM technology:
；
where v denotes the augmented lagrangian multiplier, k denotes the time interval parameter,represents the augmented lagrangian multiplier at time k,/>Representing a low dose CT image at the next instant of k,/and (b)>Low dose CT image representing time k, +.>Represents +.1 at time k->Sub-problems (S)>Representing +.>Sub-problems; />Sub-problemsAlso called +.>A calculation problem of the proximity operator of (a) at point u;
s440, using a modified Newton-Raphson iterative algorithm with integral term (Newton-Raphson iterative with integral) to buildSolution model of->The solution model of (2) is:
；
wherein,representing the identity matrix;
it can be seen that the light source is,the sub-problem is a quadratic optimization problem. In some embodiments, the +.f is calculated using Newton-Raphson iterative algorithm with integral term>；
In one embodiment, the method willThe solution model of (c) is rewritten as:
；
wherein, gamma is a scale factor, gamma >0, E is an integral term, and W represents a characteristic weight matrix value of the image;
since the prototype of the PID algorithm is a continuous function, it needs to be discretized to accommodate the computer operating system.
；
Wherein,is a steady state error, +.>、/>And->Representing different system parameters, respectively.
The integral term is used to eliminate steady state errors and adjusts the performance of the system by combining proportional and derivative controls. Before the node position is reached,always positive, with an integral always greater than 0. When there is a steady state error in the system, the error remains the same value but the integral term changes. This means that the previous steady state error value will counteract the proportional control algorithm and the integral term will continue to have an effect, resulting in +.>The output of (c) continues to increase, thereby eliminating steady state errors. In the invention, the integral term plays an important role in the iterative algorithm, and the Newton-Raphson iterative algorithm with the integral term is adopted, so that the algorithm convergence can be helped and the robustness of the algorithm can be improved. />The sub-problem is->The solution of the adjacent operator of (2) can be obtained through a denoising device; in one embodiment, the->The solution model of (2) is:
，
wherein,representing the denoising means.
In some embodiments, based onUpdating the reconstructed image of the solution model of (2)>So that->Converging to the local optimal solution, updating once +.>Is sufficient to enable +.>Converging to a locally optimal solution. The v sub-problem is at the point +.>Calculated->Is obtained by denoising, i.e. +.>。
In a preferred embodiment, in S500, the performing iterative training on the deep neural network based on the solution model until a loss function is lower than a set loss threshold value, to obtain a low dose CT image restoration model includes:
building the data set into a training setAnd test set->During training, the training set is treated with batch training method>According to the batch size->Input data divided into a plurality of batches are sequentially input into a deep neural network for training, and the input data of each batch is +.>After passing through the deep neural network, obtaining a normal dose CT image, namely a normal CT image;
integrating the test setAccording to the batch size->The test data divided into a plurality of batches are sequentially input into a deep neural network for training, and the test data of each batch is +.>Calculating model loss of the deep neural network by using the loss function;
and updating the back propagation parameters of the deep neural network based on the model loss, and performing iterative training on the deep neural network by using a solving model until the loss function is determined to be lower than a set loss threshold value, so as to obtain a low-dose CT image restoration model.
In some embodiments, a deep neural network of low dose CT image restoration is constructed, using six denoising devices to help the deep neural network converge, forcing all parameters to be shared to avoid overfitting. Inputting the data set into a deep neural network for training by adopting a batch training method; and calculating model loss according to the repair result and the real result, and then back-propagating and updating model parameters. In some embodiments, model loss may be achieved by minimizing the estimated CT image and the reference normal dose CT imageLoss betweenTo realize the method.
The invention combines an optimization algorithm with an integral term and a modified deep neural network, and provides a method for solving the problem of low-dose CT image recovery. Simulation and test results show great potential of our approach in noise suppression. In practical application scenarios, noise disturbances are unavoidable. Because the integral term exists in the model, the robustness of the model under a large amount of noise interference can be ensured, and the influence of external interference in the image recovery process can be overcome.
In order to verify the beneficial effects of the invention, the embodiment of the invention also provides the following test experiments:
as shown in fig. 2, the architecture of the denoising consists of two distinct parts, namely an encoder and a decoder, for a total of 13 layers.
As shown in fig. 3a, 3b and 3c, the architecture of the attention module of fig. 3 a; FIG. 3b architecture of a feature encoder; fig. 3c architecture of the feature decoder.
As shown in fig. 4a, 4b and 4c, the comparison of CT images in the AAPM dataset recovers. Fig. 4a normal dose CT image. Fig. 4b low dose CT image. Fig. 4c reconstructed CT image.
Table 1 lists the various metrics of the inventive method and other prior methods in the AAPM dataset, and the results of some metrics were also changed due to the relatively low noise content in the AAPM dataset, as shown in table 2. The BM3D algorithm improves sparse representation of the image in the transform domain, and better retains details in the image than the K-SVD algorithm. It can thus be seen from Table 2 that the PSNR value of BM3D is higher than that of K-SVD. However, in some cases BM3D may produce image artifacts or be overly smooth, and the present invention may alleviate this problem to some extent, as can be seen from the results of table 2. The PSNR obtained by the method of the invention is 4.835dB higher than the original low-dose CT image, thereby further proving the usability of the invention.
TABLE 1 quantitative results (simulation dataset) for different algorithms
Table 2: quantitative results (test dataset) for different algorithms
Corresponding to the method of fig. 1, referring to fig. 5, an embodiment of the present invention provides a low dose CT image restoration system, comprising:
at least one processor;
at least one memory for storing at least one program;
the at least one program, when executed by the at least one processor, causes the at least one processor to implement the method described above.
It can be seen that the content in the above method embodiment is applicable to the system embodiment, and the functions specifically implemented by the system embodiment are the same as those of the method embodiment, and the beneficial effects achieved by the method embodiment are the same as those achieved by the method embodiment.
Furthermore, the embodiment of the invention also discloses a computer program product or a computer program, and the computer program product or the computer program is stored in a computer readable storage medium. The computer program may be read from a computer readable storage medium by a processor of a computer device, the processor executing the computer program causing the computer device to perform the method as described above. Similarly, the content in the above method embodiment is applicable to the present storage medium embodiment, and the specific functions of the present storage medium embodiment are the same as those of the above method embodiment, and the achieved beneficial effects are the same as those of the above method embodiment.
Those of ordinary skill in the art will appreciate that all or some of the methods, systems, and methods disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as known to those skilled in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Furthermore, as is well known to those of ordinary skill in the art, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.
While the preferred embodiments of the present disclosure have been illustrated and described, the present disclosure is not limited to the above-described embodiments, and various equivalent modifications and substitutions can be made by those skilled in the art without departing from the spirit of the present disclosure, and these equivalent modifications and substitutions are intended to be included in the scope of the present disclosure as defined in the appended claims.

Claims (7)

1. A method of low dose CT image restoration, the method comprising the steps of:
s100, acquiring a data set, wherein the data set comprises a normal dose CT image and a corresponding low dose CT image;
s200, constructing a deep neural network for repairing a low-dose CT image, inputting the data set into the deep neural network for training, and obtaining key information features of the low-dose CT image;
s300, denoising key information features of the low-dose CT image through an improved denoising device to obtain a reconstructed image of the low-dose CT image, and determining loss functions of the low-dose CT image and a normal-dose CT image based on the reconstructed image;
s400, constructing an image restoration model and establishing a solving model of the image restoration model;
s500, performing iterative training on the deep neural network based on the solving model until a loss function is lower than a set loss threshold value, so as to obtain a low-dose CT image restoration model;
s600, acquiring a low-dose CT image to be repaired, and inputting the low-dose CT image to be repaired into the low-dose CT image repair model to obtain a repaired CT image;
in S300, the denoising processing is performed on the key information feature of the low dose CT image by using an improved denoising device, to obtain a reconstructed image of the low dose CT image, including:
s310, constructing a denoising device based on an improved depth neural network model, wherein a series of fully-connected convolution layers are used as a stacked encoder, and a stacked decoder is formed by using the series of fully-connected deconvolution layers;
s320, denoising the low-dose CT image through the denoising device, and extracting key information from key information features of the low-dose CT image to obtain a reconstructed image of the low-dose CT image.
2. The method for repairing a low dose CT image according to claim 1, wherein in S200, the constructing a deep neural network for repairing a low dose CT image, inputting the data set into the deep neural network for training, obtaining key information features of the low dose CT image, includes:
s210, constructing a deep neural network, and establishing a mapping relation model from a normal dose CT image to a low dose image based on the deep neural network, wherein the mapping relation model is as follows:；
wherein,for low dose CT images, < >>For normal dose CT image, < >>For the repaired image, ++>Representing deconvolution in the deep neural network, namely mapping relation from a normal dose CT image to a low dose CT image;
s220, inputting the data set into the deep neural network for training by adopting a batch training method, and obtaining key information characteristics of the low-dose CT image through the constructed spatial attention module, the channel attention module and the spatial channel mixed attention module.
3. A low dose CT image restoration method according to claim 2, wherein the loss functions of the low dose CT image and the normal dose CT image are calculated by the following formula:
；
wherein,representing normal dose CT image, < >>) Representing reconstructed image +.>Representing the i-th low dose CT image, < >>Represents the ith normal dose CT image in the dataset, and N represents the total number of low dose CT images in the dataset.
4. A low dose CT image restoration method as claimed in claim 3 wherein in S400, said constructing an image restoration model for image restoration, converting said image restoration model into a solution model, comprises:
s410, constructing an image restoration model, wherein the expression of the image restoration model is as follows:
；
wherein,is an auxiliary variable, +.>Representing the observation image +.>Representing the original image +.>Representing a linear operator +.>Representing a regularization factor associated with the prior, λ being a parameter weighted to the regularization term;
s420, converting the solved image recovery model into a constraint optimization problem, wherein the expression of the constraint optimization problem is as follows:
；
wherein,is a penalty parameter;
s430, converting the constraint optimization problem into two sub-problems using alternative ADMM technology:
；
where v denotes the augmented lagrangian multiplier, k denotes the time interval parameter,represents the augmented lagrangian multiplier at time k,/>Representing a low dose CT image at the next instant of k,/and (b)>Low dose CT image representing time k, +.>Represents +.1 at time k->Sub-problems (S)>Representing +.>Sub-problems; />The sub-problem is also called +.>A calculation problem of the proximity operator of (a) at point u;
s440, establishing by using Newton-Raphson iterative algorithm with integral termSolution model of->The solution model of (2) is:
；
wherein,representing the identity matrix;
will beThe solution model of (c) is rewritten as:
；
wherein, gamma is a scale factor, gamma >0, E is an integral term, and W represents a characteristic weight matrix value of the image;
；
wherein,is a steady state error, +.>、/>And->Respectively representing different system parameters;
the solution model of (2) is:
wherein->Representing the denoising means.
5. The method according to claim 4, wherein in S500, the performing iterative training on the deep neural network based on the solution model until a loss function is lower than a set loss threshold value, to obtain a low-dose CT image restoration model, includes:
building the data set into a training setAnd test set->During training, the training set is treated with batch training method>According to the batch size->The input data divided into a plurality of batches are sequentially input into a deep neural network for training, and normal dose CT images are obtained after the input data of each batch pass through the deep neural network;
integrating the test setAccording to the batch size->The test data divided into a plurality of batches are sequentially input into a deep neural network for training, and the test data of each batch is +.>Calculating model loss of the deep neural network by using the loss function;
and updating the back propagation parameters of the deep neural network based on the model loss, and performing iterative training on the deep neural network by using a solving model until the loss function is determined to be lower than a set loss threshold value, so as to obtain a low-dose CT image restoration model.
6. A low dose CT image restoration system, the system comprising:
at least one processor;
at least one memory for storing at least one program;
the at least one program, when executed by the at least one processor, causes the at least one processor to implement the low dose CT image restoration method as set forth in any one of claims 1 to 5.
7. A computer readable storage medium, in which a processor executable program is stored, characterized in that the processor executable program is for performing the method according to any one of claims 1 to 5 when being executed by a processor.
CN202410028983.3A 2024-01-09 2024-01-09 Low-dose CT image restoration method, system and storage medium Active CN117541481B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410028983.3A CN117541481B (en) 2024-01-09 2024-01-09 Low-dose CT image restoration method, system and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410028983.3A CN117541481B (en) 2024-01-09 2024-01-09 Low-dose CT image restoration method, system and storage medium

Publications (2)

Publication Number Publication Date
CN117541481A true CN117541481A (en) 2024-02-09
CN117541481B CN117541481B (en) 2024-04-05

Family

ID=89794209

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410028983.3A Active CN117541481B (en) 2024-01-09 2024-01-09 Low-dose CT image restoration method, system and storage medium

Country Status (1)

Country Link
CN (1) CN117541481B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111047524A (en) * 2019-11-13 2020-04-21 浙江工业大学 Low-dose CT lung image denoising method based on deep convolutional neural network
WO2022120883A1 (en) * 2020-12-07 2022-06-16 深圳先进技术研究院 Training method for low-dose image denoising network and denoising method for low-dose image
CN114708345A (en) * 2022-03-17 2022-07-05 上海长征医院 CT image reconstruction method, device, equipment and storage medium

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111047524A (en) * 2019-11-13 2020-04-21 浙江工业大学 Low-dose CT lung image denoising method based on deep convolutional neural network
WO2022120883A1 (en) * 2020-12-07 2022-06-16 深圳先进技术研究院 Training method for low-dose image denoising network and denoising method for low-dose image
CN114708345A (en) * 2022-03-17 2022-07-05 上海长征医院 CT image reconstruction method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN117541481B (en) 2024-04-05

Similar Documents

Publication Publication Date Title
US11430162B2 (en) System and method for image conversion
CN107527359B (en) PET image reconstruction method and PET imaging equipment
CN107481297B (en) CT image reconstruction method based on convolutional neural network
US7831097B2 (en) System and method for image reconstruction
Zhou et al. Limited view tomographic reconstruction using a cascaded residual dense spatial-channel attention network with projection data fidelity layer
CN107958471B (en) CT imaging method and device based on undersampled data, CT equipment and storage medium
JP2020534929A (en) Deep learning-based scattering correction
EP2992504B1 (en) De-noised reconstructed image data edge improvement
US11645736B2 (en) Image processing methods, apparatuses and systems
CN110651302B (en) Method and apparatus for image reconstruction
US20180089864A1 (en) Cbct image processing method
CN112102213A (en) Low-dose CT image processing method, scanning system and computer storage medium
CN111161182B (en) MR structure information constrained non-local mean guided PET image partial volume correction method
CN115272511A (en) System, method, terminal and medium for removing metal artifacts in CBCT image based on dual decoders
CN117541481B (en) Low-dose CT image restoration method, system and storage medium
CN111862255A (en) Regularization image reconstruction method, system, readable storage medium and device
Wang et al. Hybrid pre-log and post-log image reconstruction for computed tomography
CN111899315A (en) Method for reconstructing low-dose image by using multi-scale feature perception depth network
CN112116677B (en) Low-dose CT reconstruction method based on low-dimensional manifold priori
US20210330274A1 (en) Computer-implemented method, computer program, systems and x-ray facility for correction of x-ray image data with regard to noise effects
KR20140133384A (en) Method and apparatus for dynamic image reconstruction using patch based low rank regularization in positron emission tomography
CN112215910A (en) X-ray system, image processing device, and program
CN116312983B (en) Hidden regularization low-dose CT image reconstruction method and system based on noise generation mechanism
CN117115046B (en) Method, system and device for enhancing sparse sampling image of radiotherapy CBCT
CN110269638B (en) Image reconstruction method, system, readable storage medium and apparatus

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant