CN112435164B - Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network - Google Patents

Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network Download PDF

Info

Publication number
CN112435164B
CN112435164B CN202011323293.9A CN202011323293A CN112435164B CN 112435164 B CN112435164 B CN 112435164B CN 202011323293 A CN202011323293 A CN 202011323293A CN 112435164 B CN112435164 B CN 112435164B
Authority
CN
China
Prior art keywords
dose
image
lung
lung image
low
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.)
Active
Application number
CN202011323293.9A
Other languages
Chinese (zh)
Other versions
CN112435164A (en
Inventor
金燕
姜智伟
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.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202011323293.9A priority Critical patent/CN112435164B/en
Publication of CN112435164A publication Critical patent/CN112435164A/en
Application granted granted Critical
Publication of CN112435164B publication Critical patent/CN112435164B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • 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]
    • 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
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30061—Lung

Landscapes

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

Abstract

The invention discloses a simultaneous super-resolution and denoising method for generating a low-dose CT lung image of an antagonism network based on multiple scales, which comprises the following steps: (1) Collecting a high-dose CT lung image and degrading the high-dose CT lung image into a low-dose CT lung image to form a sample; (2) Constructing a generating type countermeasure network comprising a generator and a discriminator, wherein the generator comprises a multi-scale feature extraction module, a convolution layer, a plurality of residual channel attention modules, a filling module, a residual module and an up-sampling module which are connected in sequence, and the discriminator comprises a series of convolution layer and BN modules; (3) Training the generated type countermeasure network optimization network parameters by using samples, wherein the determined network parameters and a generator form a super-resolution and denoising model; (4) When the method is applied, the low-dose CT lung image is acquired and input into the super-resolution and denoising model, and the high-dose CT lung image is calculated and output. The method improves the super-resolution imaging quality and the denoising effect.

Description

Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network
Technical Field
The invention relates to the fields of digital images, designed image denoising, image super-resolution, computer vision and deep learning, in particular to a simultaneous super-resolution and denoising method for generating a low-dose CT lung image of an countermeasure network based on multiple scales.
Background
Computed tomography (Computed Tomography, CT) scanning, also known as computed tomography, utilizes a computer to process a number of combined x-rays to scan an object over a particular area that measures cross-sections produced from different angles, allowing the user to see the interior of the object without curtailing. As the CT imaging technology is adopted for cross-section imaging, tissues or organs can be displayed in any direction through image reconstruction, so that lesions can be more comprehensively displayed, and omission is prevented; the method has high density resolution, can display fine lesions with density change, and can determine the nature of the lesions; in addition, CT has the advantages of noninvasive, fast imaging, etc. has become a widely used and highly safe medical diagnostic technique.
With the continuous development of CT, CT diagnosis plays an increasingly important role in early disease screening, but higher radiation is generated during CT scanning, which can cause harm to the human body. Medical studies now show that 1.5% -2% of tumors may be caused by the high radiation dose of CT. In particular, for patients with high-risk diseases and infants, CT diagnosis with high dosage is not applicable.
At the beginning of the 90 s, naidich et al first proposed the concept of Low Dose CT (LDCT) to reduce the radiation dose of radiation CT by reducing the tube current while ensuring that the other scan parameters are unchanged. Because of the large difference in structure between the lung and other tissue and organs, the air content is high and the density is low, so that a satisfactory image can be formed by CT scanning with 75% -90% lower dose than the conventional dose, and although the damage caused by radiation can be reduced by low-dose CT, the projected data can be polluted, and the low-dose CT lung image has obvious noise and streak artifacts. The interference of the noise can reduce the probability of screening lung cancer, and can interfere the judgment of a doctor on the disease to a certain extent, thereby being unfavorable for clinical diagnosis, and therefore, the method has great significance in denoising and restoring the low-dose CT lung image.
Although the traditional noise removal algorithm achieves good effect on restoration of low-dose CT lung images, the noise and the artifacts are restrained, and meanwhile, the loss of edge details is caused.
Disclosure of Invention
Based on the above, the invention aims to provide a simultaneous super-resolution and denoising method for generating a low-dose CT lung image of an antagonism network based on multiple scales, and combines an attention channel mechanism with the multiple scales to improve super-resolution imaging quality and denoising effect.
In order to achieve the above object, the present invention provides the following technical solutions:
A simultaneous super-resolution and denoising method for generating a low-dose CT lung image of an countermeasure network based on multiple scales, comprising the steps of:
(1) Collecting high-dose CT lung images, degrading the high-dose CT lung images into low-dose CT lung images by an image preprocessing method, and forming a sample by each high-dose CT lung image and the corresponding low-dose CT lung image;
(2) Constructing a generating type countermeasure network comprising a generator and a discriminator, wherein the generator comprises a multi-scale feature extraction module, a convolution layer, a plurality of residual channel attention modules, a filling module, a residual module and an up-sampling module which are connected in sequence, and the generator is used for generating a predicted high-dose CT lung image according to an input low-dose CT lung image, and the discriminator comprises a series of convolution layer and BN module and is used for judging the authenticity of the input high-dose CT lung image;
(3) Training the generated type countermeasure network optimization network parameters by using samples, wherein the determined network parameters and a generator form a super-resolution and denoising model;
(4) When the method is applied, the low-dose CT lung image is acquired and input into the super-resolution and denoising model, and the high-dose CT lung image is calculated and output.
Preferably, in the step (1), poisson noise is added to the high-dose CT lung image through an FBP algorithm, and the CT lung image after the poisson noise is added is downsampled at least 4 times by using a double interpolation method, so that degradation of the quality CT lung image is realized, and a low-dose CT lung image is obtained.
Preferably, the downsampled size is 100×100, and the sampling interval is 50.
Preferably, the multi-scale feature extraction module comprises at least three feature extraction paths, each feature extraction path adopts convolution check of different sizes to carry out convolution operation on receptive fields of different scales and different sizes of the low-dose CT lung images so as to extract different image features, and the extracted different image features are subjected to fusion operation to output fusion image features.
Preferably, the residual channel attention module includes an average pooling layer, at least 2 convolution layers, and a fusion operation, and an image feature obtained by fusing the input image feature and the output image feature of the input image feature after the average pooling layer and the at least 2 convolution layers through the fusion operation is used as an output of the residual channel attention module.
Preferably, the filling module is connected with the output of the residual channel attention module, and comprises at least 2 convolution layers, and is used for performing feature filling on the image features output by the residual channel attention module.
Preferably, the residual module fuses the image features output by the multi-scale feature extraction module and the image features output by the filling module to serve as input features of the up-sampling module.
Preferably, the up-sampling module comprises a convolution layer, PRelu nonlinear activation functions, an up-sampling layer and the convolution layer, and the super-resolved noiseless CT lung image distribution is extracted through up-sampling operation and is output as a high-dose CT lung image.
Preferably, when training the generating type countermeasure network, the generator and the discriminator are both optimized by adopting an Adam optimizer, the initial learning rate is set to be 0.001, then the attenuation of each 50 epochs is 0.1 times, and the loss function of the discriminator is set as follows:
D real =σ (D (y) -E [ D (G (x)) ] -
D fake =σ (D (G (x)) -E [ D (y) ]) -0 if G (x) is more authentic than y
Wherein x represents a low-dose CT lung image, y represents a real high-dose CT lung image, E (-) represents the expectation of the discriminator, sigma (-) represents a Sigmoid function, G (-) represents generator generated data, and D (-) represents discriminator generated data;
the loss function L GAN of the countermeasure network is generated according to the loss function definition of the discriminator:
The loss function L Generator of the generator is defined from the loss function L GAN and the loss function L MSE as:
LGenerator＝αLMSE+βLGAN
wherein alpha and beta are weight coefficients and the loss function L MSE.
Preferably, after the training of the generated countermeasure network is finished, the test sample is also required to be used for fine-tuning the network parameters of the generated countermeasure network, and the network parameters are optimized again.
Compared with the prior art, the invention has the beneficial effects that at least the following steps are included:
According to the simultaneous super-resolution and denoising method for generating the low-dose CT lung image based on the countermeasure network in the multi-scale mode, the characteristics of different scales are extracted through the convolution kernels of different receptive fields in the multi-scale characteristic extraction module, so that super-resolution and denoising performance is improved, image characteristic information is extruded into the same channel through the average pooling layer in the residual channel attention module, information is distributed according to different weights, and finally the distributed information and priori information are fused to obtain better performance, meanwhile, difficulty of network learning is reduced, and the performance of a restored image is improved through the generated countermeasure network of the generator and the discriminator.
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 required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only 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 schematic diagram of a generating type countermeasure network according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a multi-scale feature extraction module according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a residual channel attention module according to an embodiment of the present invention;
FIG. 4 is a graph of the recovery result of a low-dose CT lung image using a super-resolution and denoising model according to an embodiment of the present invention, where a-d is a high-dose CT lung image, e-h is a low-resolution (128×128) noisy low-dose CT lung image, and i-l is a CT lung image of size×4 after recovery by the super-resolution and denoising model.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the scope of the invention.
The embodiment provides a simultaneous super-resolution and denoising method for generating a low-dose CT lung image of an countermeasure network based on multiple scales, which is called CT-SDGAN for short and comprises the following steps of:
Step 1: a high dose CT lung dataset is collected and a degenerated low dose CT lung image is obtained by image preprocessing.
The specific process is as follows:
Step 1.1: collecting a high-dose CT lung data set, and converting the high-dose CT lung data set into a common image format, such as jpg, png and the like, so as to obtain an original high-dose CT lung image;
Step 1.2: adding poisson noise into the high-dose CT image through an FBP algorithm, downsampling the picture added with poisson noise by 4 times by using a double interpolation (BICUBIC) method so as to simulate a corresponding low-dose CT lung image, and dividing the processed paired high-dose CT lung image and low-dose lung image into two parts of training data and test data;
Step 1.3: and (3) carrying out data enhancement on the high-dose CT image and the low-dose CT image in the training data, simultaneously using a rotation and scaling technology, taking the enhanced data as a training set, wherein the downsampling size of a final image is 100 multiplied by 100, and the sampling interval is 50.
Step 2: a simultaneous super-resolution and denoising generative countermeasure network for a low dose CT image is constructed.
As shown in fig. 1, the generating type countermeasure network includes a generator and a discriminator, wherein the generator is configured to generate a predicted high dose CT lung image according to an input low dose CT lung image, and specifically includes a multi-scale feature extraction module, a convolution layer, a plurality of residual channel attention modules, a filling module, a residual module, and an upsampling module, and the discriminator is configured to determine authenticity of the input high dose CT lung image, and includes a series of convolution layer and BN modules.
The specific construction process of the generated countermeasure network comprises the following steps:
step 2.1: and constructing a multi-scale feature extraction module. As shown in fig. 2, the simultaneous super-resolution and denoising method based on multi-scale generation of low-dose lung CT images of an countermeasure network uses a multi-scale feature extraction module at layer 1, and performs convolution operation by using convolution check images of different sizes and receptive fields of different sizes to extract different image features. The convolution kernels with different sizes are respectively 1 multiplied by 1,3 multiplied by 3 and 5 multiplied by 5, and finally the features of different receptive fields are fused together through a feature fusion module so as to furthest reserve the features extracted in different scales.
Step 2.2: a residual channel attention module using a channel attention mechanism is constructed. As shown in fig. 3, the simultaneous super-resolution and denoising method for generating low-dose CT images of an countermeasure network based on multiple scales uses a residual channel attention module at layers 2-12; the cross correlation between features is obtained by squeezing features of an input network model into a channel through average global pooling, then feeding the features into 2 convolution layers, and finally fusing the squeezed features and prior features together through a feature fusion layer, wherein the convolution operation is to use 2 convolution kernels (64 of output channels) with the size of 3×3, and one convolution kernel (128 of output channels) with the fused features.
Step 2.3: a residual learning method is used. Directly forming a large residual error unit from input to output; the gradient may disappear or the prior knowledge constraint is not existed before the residual error unit is not existed, and the network convergence can be quickened through the jump structure, so that the training difficulty is reduced.
Step 2.4: an up-sampling module for constructing a reconstruction output, wherein the up-sampling module is used by the last layer of a simultaneous super-resolution and denoising method for generating a low-dose CT image of an countermeasure network based on multiple scales, and the up-sampling module comprises the following components: 3 x 3 convolutions, PRelu nonlinear activation functions, pixelshuffle upsampling layers and 3 x 3 convolutions. And extracting the distribution of the super-resolved noiseless CT lung images through up-sampling operation, and taking the distribution as output.
Step 2.5: constructing a discriminator which consists of a series of convolution layers, PRelu nonlinear activation functions and Batch Normalization (BN), wherein the convolution layers are used for acquiring image information, PRelu are used for activating the functions, and the BN layers are used for normalizing data, wherein the discriminator consists of 9 convolution kernels, and output channels of the convolution kernels are 64, 64, 128, 128, 256, 256, 512 and 512,1 respectively; the step sizes are 1,2,1,2,1,2,1,2,1 respectively; the remaining convolution kernels, except the first and last convolution kernels, are followed by PRelu activation functions and BN layers, with the first convolution kernel being followed by PRelu activation functions.
Step 3: and training the generated type countermeasure network optimization network parameters by using samples, wherein the determined network parameters and a generator form a super-resolution and denoising model.
The specific process is as follows:
Step 3.1: constructing a pre-training model, generating a low-dose CT image of an countermeasure network based on multiple scales, and simultaneously performing super-resolution and denoising, wherein a generator network and a discriminator network are optimized by adopting an Adam optimizer, the initial learning rate is set to be 0.001, then attenuation of each 50 epochs is 0.1 times, and the loss function of the discriminator network is set as follows:
D real =σ (D (y) -E [ D (G (x)) ] →1 if y is more realistic than G (x)
D feak =σ (D (G (x)) -E [ D (y) ]) -0 if G (x) is more authentic than y
Wherein E (-) represents the expectation of the arbiter, σ represents the Sigmoid function, G (-) represents the generator generated data, and D (-) represents the arbiter generated data; the generation of the antagonism network (GAN) loss function can then be defined by the arbiter network function as:
The loss function of the final generator network may be defined as a combination of GAN loss function and MSE function:
LGenerator＝αLMSE+βLGAN
Wherein alpha and beta are super parameters, and are respectively set to 0.06 and 0.009;
Step 3.2: and fine-tuning the model, and adjusting the learning rate and the super parameters in the loss function by loading the model with the highest precision in the test set to obtain a super-resolution and denoising CT image result.
Step 3.3: testing the super-resolution and denoising performance of the trained model, and verifying the denoising effect of the trained model by using a test set (which does not appear in a training set); the model denoising effect is evaluated from subjective visual effect and objective evaluation index by comparing with the original high-dose CT lung image.
And 4, acquiring a low-dose CT lung image when the method is applied, inputting the low-dose CT lung image into a super-resolution and denoising model, and outputting a high-dose CT lung image through calculation.
Specific experimental example
(1) Selecting experimental data
The data sets selected in the experiment are all data sets from AAPM low dose CT challenge large-scale competition disclosure, are provided by the Mayo clinic, and comprise 5936 low dose CT pictures with the resolution of 512 multiplied by 512, the medical images are converted into a common picture format by using fan beam projection and classical FBP algorithm, then the common picture format is downsampled by BICUBIC method, and simultaneously poisson noise is added to simulate a low dose CT image;
(2) Experimental results
According to the method for simultaneously super-resolution and denoising the low-dose CT lung image based on the multi-scale generation countermeasure network, training the multi-scale generation countermeasure network for super-resolution and denoising of the low-dose CT lung image, after a model is constructed, carrying out fine adjustment by loading a generator with highest precision in a training model, and then verifying the performance of the model by using a picture in a verification set to obtain a super-resolution and denoising model CT-SDGAN of the low-dose CT lung image, wherein the super-resolution is 4.
Taking 4 pictures in the verification set as an example, a-d is a high dose CT lung image with a resolution of 512×512, as shown in FIG. 4; e-h is a low-dose CT lung image containing poisson noise, and the resolution is 128 multiplied by 128; i-l is a restored image effect diagram of the model CT-SDGAN, the resolution is 512 multiplied by 512, the super resolution and the denoising performance of the CT-SDGAN can be intuitively seen, and the performance is good.
The foregoing detailed description of the preferred embodiments and advantages of the invention will be appreciated that the foregoing description is merely illustrative of the presently preferred embodiments of the invention, and that no changes, additions, substitutions and equivalents of those embodiments are intended to be included within the scope of the invention.

Claims (5)

1. A simultaneous super-resolution and denoising method for generating a low-dose CT lung image of an countermeasure network based on multiple scales, comprising the steps of:
(1) Collecting high-dose CT lung images, degrading the high-dose CT lung images into low-dose CT lung images by an image preprocessing method, and forming a sample by each high-dose CT lung image and the corresponding low-dose CT lung image;
(2) Constructing a generating type countermeasure network comprising a generator and a discriminator, wherein the generator comprises a multi-scale feature extraction module, a convolution layer, a plurality of residual channel attention modules, a filling module, a residual module and an up-sampling module which are connected in sequence, and the generator is used for generating a predicted high-dose CT lung image according to an input low-dose CT lung image, and the discriminator comprises a series of convolution layer and BN module and is used for judging the authenticity of the input high-dose CT lung image;
The multi-scale feature extraction module comprises at least three feature extraction passages, wherein each feature extraction passage adopts convolution check of different sizes to carry out convolution operation on the receptive fields of different scales and different sizes of the low-dose CT lung images so as to extract different image features, and the extracted different image features are subjected to fusion operation to output fusion image features;
The residual channel attention module comprises an average pooling layer, at least 2 convolution layers and fusion operation, wherein the image characteristics obtained by fusion operation of the input image characteristics and the output image characteristics of the average pooling layer and the at least 2 convolution layers are used as the output of the residual channel attention module;
The filling module is connected with the output of the residual channel attention module and comprises at least 2 convolution layers for carrying out feature filling on the image features output by the residual channel attention module;
the residual error module fuses the image features output by the multi-scale feature extraction module and the image features output by the filling module to serve as input features of the up-sampling module;
The up-sampling module comprises a convolution layer, PRelu nonlinear activation functions, an up-sampling layer and the convolution layer, and the super-resolved noiseless CT lung image distribution is extracted through up-sampling operation and is output as a high-dose CT lung image;
(3) Training the generated type countermeasure network optimization network parameters by using samples, wherein the determined network parameters and a generator form a super-resolution and denoising model;
(4) When the method is applied, the low-dose CT lung image is acquired and input into the super-resolution and denoising model, and the high-dose CT lung image is calculated and output.
2. The simultaneous super-resolution and denoising method for generating a low-dose CT lung image based on a countermeasure network according to claim 1, wherein in step (1), poisson noise is added to a high-dose CT lung image by an FBP algorithm, and a double interpolation method is used to downsample the CT lung image after the poisson noise is added by at least 4 times, so as to realize degradation of a quality CT lung image, and obtain a low-dose CT lung image.
3. The simultaneous super-resolution and denoising method for generating a low-dose CT lung image based on a countermeasure network according to claim 2, wherein the downsampling is 100×100 in size and the sampling interval is 50.
4. The simultaneous super-resolution and denoising method for generating low-dose CT lung images based on a countermeasure network according to claim 1, wherein, during training of the generated countermeasure network, both the generator and the arbiter are optimized by Adam optimizer, the initial learning rate is set to 0.001, then each 50 epochs decays by 0.1 times, and the loss function of the arbiter is set to:
D real =σ (D (y) -E [ D (G (x)) ] -
D fake =σ (D (G (x)) -E [ D (y) ]) -0 if G (x) is more authentic than y
Wherein x represents a low-dose CT lung image, y represents a real high-dose CT lung image, E (-) represents the expectation of the discriminator, sigma (-) represents a Sigmoid function, G (-) represents generator generated data, and D (-) represents discriminator generated data;
the loss function L GAN of the countermeasure network is generated according to the loss function definition of the discriminator:
The loss function L Generator of the generator is defined from the loss function L GAN and the loss function L MSE as:
LGenerator＝αLMSE+βLGAN
wherein alpha and beta are weight coefficients and the loss function L MSE.
5. The simultaneous super-resolution and denoising method for generating low-dose CT lung images of an countermeasure network based on multiple scales as claimed in claim 1 or 4, wherein after the training of the generated countermeasure network is finished, fine tuning of network parameters of the generated countermeasure network is further required by using test samples, and the network parameters are optimized again.
CN202011323293.9A 2020-11-23 2020-11-23 Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network Active CN112435164B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011323293.9A CN112435164B (en) 2020-11-23 2020-11-23 Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011323293.9A CN112435164B (en) 2020-11-23 2020-11-23 Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network

Publications (2)

Publication Number Publication Date
CN112435164A CN112435164A (en) 2021-03-02
CN112435164B true CN112435164B (en) 2024-04-30

Family

ID=74694632

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011323293.9A Active CN112435164B (en) 2020-11-23 2020-11-23 Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network

Country Status (1)

Country Link
CN (1) CN112435164B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113034361B (en) * 2021-03-03 2022-10-14 深圳市数字城市工程研究中心 Remote sensing image super-resolution reconstruction method based on improved ESRGAN
CN113158997B (en) * 2021-05-22 2023-04-18 河南工业大学 Grain depot monitoring image denoising method, device and medium based on deep learning
CN114331921A (en) * 2022-03-09 2022-04-12 南昌睿度医疗科技有限公司 Low-dose CT image noise reduction method and device
CN114693831B (en) * 2022-05-31 2022-09-02 深圳市海清视讯科技有限公司 Image processing method, device, equipment and medium
CN115953494B (en) * 2023-03-09 2023-05-30 南京航空航天大学 Multi-task high-quality CT image reconstruction method based on low dose and super resolution

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110415170A (en) * 2019-06-24 2019-11-05 武汉大学 A kind of image super-resolution method based on multiple dimensioned attention convolutional neural networks
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
CN111709895A (en) * 2020-06-17 2020-09-25 中国科学院微小卫星创新研究院 Image blind deblurring method and system based on attention mechanism
CN111932460A (en) * 2020-08-10 2020-11-13 北京大学深圳医院 MR image super-resolution reconstruction method and device, computer equipment and storage medium
CN111968195A (en) * 2020-08-20 2020-11-20 太原科技大学 Dual-attention generation countermeasure network for low-dose CT image denoising and artifact removal

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11232541B2 (en) * 2018-10-08 2022-01-25 Rensselaer Polytechnic Institute CT super-resolution GAN constrained by the identical, residual and cycle learning ensemble (GAN-circle)

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110415170A (en) * 2019-06-24 2019-11-05 武汉大学 A kind of image super-resolution method based on multiple dimensioned attention convolutional neural networks
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
CN111709895A (en) * 2020-06-17 2020-09-25 中国科学院微小卫星创新研究院 Image blind deblurring method and system based on attention mechanism
CN111932460A (en) * 2020-08-10 2020-11-13 北京大学深圳医院 MR image super-resolution reconstruction method and device, computer equipment and storage medium
CN111968195A (en) * 2020-08-20 2020-11-20 太原科技大学 Dual-attention generation countermeasure network for low-dose CT image denoising and artifact removal

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
图像超分辨率重建研究综述;唐艳秋;潘泓;朱亚平;李新德;;电子学报;20200715(07);全文 *
基于WGAN单帧人脸图像超分辨率算法;周传华;吴幸运;李鸣;;计算机技术与发展;20200910(09);全文 *
彭晏飞 ; 高艺 ; 杜婷婷 ; 桑雨 ; 訾玲玲 ; .生成对抗网络的单图像超分辨率重建方法.计算机科学与探索.(09),全文. *
金炜 ; 陈莹 ; .多尺度残差通道注意机制下的人脸超分辨率网络.计算机辅助设计与图形学学报.(06),全文. *

Also Published As

Publication number Publication date
CN112435164A (en) 2021-03-02

Similar Documents

Publication Publication Date Title
CN112435164B (en) Simultaneous super-resolution and denoising method for generating low-dose CT lung image based on multiscale countermeasure network
CN111325686B (en) Low-dose PET three-dimensional reconstruction method based on deep learning
CN110443867B (en) CT image super-resolution reconstruction method based on generation countermeasure network
US11120582B2 (en) Unified dual-domain network for medical image formation, recovery, and analysis
US20200294288A1 (en) Systems and methods of computed tomography image reconstruction
CN112258415B (en) Chest X-ray film super-resolution and denoising method based on generation countermeasure network
CN115953494B (en) Multi-task high-quality CT image reconstruction method based on low dose and super resolution
CN113516586A (en) Low-dose CT image super-resolution denoising method and device
Gajera et al. CT-scan denoising using a charbonnier loss generative adversarial network
JP2022550688A (en) Systems and methods for improving low-dose volume-enhanced MRI
Zhou et al. DuDoUFNet: dual-domain under-to-fully-complete progressive restoration network for simultaneous metal artifact reduction and low-dose CT reconstruction
CN112419173A (en) Deep learning framework and method for generating CT image from PET image
WO2022257959A1 (en) Multi-modality and multi-scale feature aggregation for synthesizing spect image from fast spect scan and ct image
CN112819914A (en) PET image processing method
Xia et al. Physics-/model-based and data-driven methods for low-dose computed tomography: A survey
Chen et al. LIT-Former: Linking in-plane and through-plane transformers for simultaneous CT image denoising and deblurring
CN115601268A (en) LDCT image denoising method based on multi-scale self-attention generation countermeasure network
JP2024507767A (en) Training data synthesizer for contrast-corrected machine learning systems
Xia et al. Deep residual neural network based image enhancement algorithm for low dose CT images
Liang et al. A self-supervised deep learning network for low-dose CT reconstruction
CN116402954A (en) Spine three-dimensional structure reconstruction method based on deep learning
CN116645283A (en) Low-dose CT image denoising method based on self-supervision perceptual loss multi-scale convolutional neural network
Xia et al. Dynamic controllable residual generative adversarial network for low-dose computed tomography imaging
CN115496659A (en) Three-dimensional CT image reconstruction method and device based on single projection data
JP2024507766A (en) Contrast enhancement using machine learning

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