CN108961237B - Low-dose CT image decomposition method based on convolutional neural network - Google Patents
Low-dose CT image decomposition method based on convolutional neural network Download PDFInfo
- Publication number
- CN108961237B CN108961237B CN201810706749.6A CN201810706749A CN108961237B CN 108961237 B CN108961237 B CN 108961237B CN 201810706749 A CN201810706749 A CN 201810706749A CN 108961237 B CN108961237 B CN 108961237B
- Authority
- CN
- China
- Prior art keywords
- image
- dose
- low
- neural network
- convolutional neural
- 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
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 66
- 210000003484 anatomy Anatomy 0.000 claims abstract description 38
- 238000012545 processing Methods 0.000 claims abstract description 24
- 238000013507 mapping Methods 0.000 claims abstract description 9
- 230000004913 activation Effects 0.000 claims description 18
- 230000009466 transformation Effects 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 238000013170 computed tomography imaging Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 6
- 238000000844 transformation Methods 0.000 claims description 6
- 238000002591 computed tomography Methods 0.000 abstract description 183
- 230000006870 function Effects 0.000 description 18
- 230000000694 effects Effects 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 8
- 230000006872 improvement Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 5
- 210000001519 tissue Anatomy 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000003672 processing method Methods 0.000 description 3
- 230000005855 radiation Effects 0.000 description 3
- 238000000926 separation method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 210000001015 abdomen Anatomy 0.000 description 2
- 230000003187 abdominal effect Effects 0.000 description 2
- 239000000470 constituent Substances 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 206010011732 Cyst Diseases 0.000 description 1
- 206010028980 Neoplasm Diseases 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 208000031513 cyst Diseases 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 231100000405 induce cancer Toxicity 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 208000032839 leukemia Diseases 0.000 description 1
- 210000004185 liver Anatomy 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 210000000952 spleen Anatomy 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001225 therapeutic effect Effects 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 210000003462 vein Anatomy 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
-
- 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
- G06T2211/00—Image generation
- G06T2211/40—Computed tomography
- G06T2211/416—Exact reconstruction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a low-dose CT image decomposition method based on a convolutional neural network, and belongs to the technical field of X-ray computed tomography. The invention comprises the following steps: step 1, respectively reconstructing a training image: low dose CT imageAnd conventional dose CT imagesLow dose CT imageAnd conventional dose CT imagesSubtracting to obtain a noise artifact imageStep 2, constructing a low-dose CT imageAnd noise artifact imagesNsA mapping convolutional neural network therebetween; step 3, using a certain amount of low-dose CT imagesWith corresponding noise-artifact images NsTraining the constructed convolutional neural network; step 4, processing the selected low-dose CT image by using the trained convolutional neural networkEnabling selected low dose CT imagesThe decomposition of the medium anatomical structure components with the noise artifact structure components. The invention provides a method capable of effectively separating star-strip artifact noise and structural features in a low-dose CT image.
Description
Technical Field
The invention relates to a method for decomposing a low-dose CT image, in particular to a method for decomposing a low-dose CT image based on a convolutional neural network, and belongs to the technical field of X-ray computed tomography.
Background
As a clinical imaging technique, X-ray Computed Tomography (CT) is widely used in disease screening, diagnosis, emergency treatment, interventional therapy and therapeutic effect supervision due to its advantages of high spatial resolution and low cost, and is one of the currently available clinical medical diagnostic tools. However, excessive X-ray exposure may induce cancer, leukemia or increase other physiological risks, and thus the problem of radiation in CT is also becoming increasingly important. However, reducing the radiation dose can drastically degrade the quality of the reconstructed image, and speckle noise and streak artifacts in the image affect clinical analysis and diagnosis. How to obtain the best CT diagnostic image with the lowest radiation dose on the basis of ensuring the image quality has become a common consensus in the industry.
Current methods for improving the quality of low-dose CT images are mainly divided into two main categories: projection space data based processing and image space data based processing. The method based on projection space data processing mainly provides more accurate projection data with less noise for reconstruction by correcting low-dose CT projection data, restoring and denoising, so as to improve the reconstruction quality, such as adaptive filtering, bilateral filtering and the like. The quality of a reconstructed low-dose image is improved directly through an image space processing technology, the method has the advantages of independence on original projection data and high processing speed, and is generally carried out by using a nonlinear processing method, such as Total Variation (Total Variation) and a Wavelet (Wavelet) transformation method, wherein artifact and noise are removed by keeping image edge information, however, important non-local properties in the image are ignored in the method, and a satisfactory effect is difficult to achieve; for example, in a sparse representation image processing algorithm based on dictionary learning, the method obtains a set of over-complete dictionaries (bases) through training, so that artifacts and noise cannot be well represented, the purpose of removing the artifacts and the noise is achieved, but the processing time is too long. With the prevalence of large data sets and large samples, deep learning has attracted extensive attention in both the industrial and academic fields, and is also gradually applied to the field of CT images. For example, Chen et al uses a residual coding network, which greatly reduces the noise artifact in the CT image and improves the identification rate of tumor lesion tissues.
In the prior art, a sparse representation method based on dictionary learning has been proved to have a good effect in low-dose abdominal CT image processing, and the dose is less than one fifth of the conventional dose, so that high image quality can be obtained. However, this method requires training high-frequency detail images in different directions, and is computationally expensive and time-consuming, and is difficult to be widely applied in an actual three-dimensional medical image processing system. In order to effectively suppress three-dimensional artifacts and noises, chinese patent application 2015105909015 proposes a low-dose CT image decomposition method based on three-dimensional distinctive feature representation, which uses a three-dimensional distinctive dictionary composed of a feature dictionary and an artifact and noise dictionary to represent a clinical low-dose CT image, and obtains a feature image represented by the feature dictionary and an artifact and noise image represented by the artifact and noise dictionary, thereby realizing the decomposition of the low-dose CT image, and being capable of effectively filtering the artifacts and noises in the low-dose three-dimensional CT image.
In summary, how to overcome the problem that the star streak artifact and the noise cannot be effectively separated in the existing low-dose CT image processing method is an urgent problem to be solved in the prior art.
The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.
Disclosure of Invention
1. Technical problem to be solved by the invention
The invention aims to solve the problem that star-stripe artifact and Noise cannot be effectively separated in the existing low-dose CT image processing method, and provides a Neural Network capable of effectively separating the star-stripe artifact Noise and structural features in a low-dose CT image by using a Convolutional Neural Network method, which is called a Noise-artifact Separation Convolutional Neural Network (NaSCNN).
2. Technical scheme
In order to achieve the purpose, the technical scheme provided by the invention is as follows:
the invention discloses a low-dose CT image decomposition method based on a convolutional neural network, which comprises the following steps of:
step 1, respectively obtaining a plurality of groups of matched low-dose CT projection numbersAccording to the CT projection data of the conventional dose, respectively reconstructing a corresponding training image: low dose CT imageAnd conventional dose CT imagesLow dose CT imageAnd conventional dose CT imagesSubtracting to obtain a noise artifact image
Step 2, constructing a low-dose CT imageAnd noise artifact image NsA mapping convolutional neural network therebetween;
step 3, using a certain amount of low-dose CT imagesWith corresponding noise-artifact images NsTraining the constructed convolutional neural network;
step 4, processing the selected low-dose CT image by using the trained convolutional neural networkEnabling selected low dose CT imagesThe decomposition of the medium anatomical structure components with the noise artifact structure components.
As a further improvement of the present invention, in step 1, the low dose CT imageIs obtained by an analytic FBP reconstruction algorithm and is a conventional dosage CT imageObtained by an iterative tv (total variation) reconstruction algorithm.
As a further improvement of the present invention, in step 2, the convolutional neural network comprises the following three convolutional modules: the device comprises a CBR module, a branch module and a residual module.
As a further improvement of the present invention, the CBR module is a combination of convolution, scaling and ReLU activation function operations; the branch module is the sum of two parallel branches, the first branch is two convolution operations, and the second branch is the combination of convolution, scale transformation and ReLU activation function operation; the residual module is the sum of two series-connected scale transformations, a ReLU activation function and a convolution operation and one convolution operation.
As a further improvement of the present invention, in step 3, a low dose CT image is takenOutputting a noise artifact prediction image in an input convolutional neural networkParameters in the convolutional neural network are updated through training so as to reduce noise artifact predicted image output by the convolutional neural networkNoise artifact image N corresponding to realitysWhen the change of the mean square error before and after the training period is less than 0.1%, the training is finished. As a further improvement of the invention, in step 4, a selected low-dose CT image V is selectedt ldInputting into the trained convolutional neural network to obtain a low-dose CT image V based on the selectiont ldNoise artifact prediction image ofAnd anatomical structure component image Vt p。
As a further development of the invention, in step 4 the anatomical structure constituent images Vt pExpressed as:
wherein, Vt ldFor the selected low-dose CT image,the image is predicted for noise artifacts based on the above selected low dose CT image.
As a further improvement of the invention, in step 3, the training sample for training the convolutional neural network is a low-dose CT imageImage block set ofLabeled as the actual corresponding noise artifact image NsImage block set of
3. Advantageous effects
Compared with the prior art, the technical scheme provided by the invention has the following remarkable effects:
(1) the invention discloses a low-dose CT image decomposition method based on a convolutional neural network, which comprises the steps of firstly obtaining a plurality of CT projection data of matched low dose and conventional dose, reconstructing a training image, and subtracting the CT images of the low dose and the conventional dose to obtain a noise artifact image; secondly, constructing a mapping convolution neural network between the low-dose CT image and the noise artifact image, wherein the network comprises three different convolution modules so as to extract artifact and noise characteristic information in the low-dose CT image; then, training the constructed convolutional neural network by using a certain amount of low-dose CT images and noise artifact images; finally, the trained convolutional neural network is used for processing the low-dose CT image to realize the decomposition of anatomical structure components and noise artifact structure components in the low-dose CT image; the low-dose CT image decomposition method can effectively distinguish star-strip artifacts and noise in low-dose CT data from human anatomy structures, has a processing effect superior to that of a traditional dictionary learning method, can meet the requirements of clinical analysis, diagnosis and the like on image quality, and improves the image effect of low-dose CT imaging.
(2) The low-dose CT image decomposition method based on the convolutional neural network is characterized in that an image space is modeled based on a deep learning method, artifacts and noise are distinguished from human anatomical structures through the powerful feature representation capability of the convolutional neural network, the test time is short, the processing effect is good, the powerful representation capability of the convolutional network is fully utilized, and the low-dose CT image noise artifacts and the decomposition between the low-dose CT image noise artifacts and the anatomical structures are realized.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flowchart of a method for low-dose CT image decomposition based on a convolutional neural network according to an embodiment of the present invention;
FIG. 2 shows five typical axial training charts (a 1-a 5: conventional dose chart; b 1-b 5: low dose chart) in the example of the present invention;
FIG. 3 is an axial conventional dose CT image in an embodiment of the present invention;
FIG. 4 is an axial low dose CT image in accordance with an embodiment of the present invention;
FIG. 5 shows the axial clinical low dose CT images separated by DFRDL (dictionary learning method) in an embodiment of the present invention (a: anatomical structure components; b: noise artifact components);
FIG. 6 shows the results of an axial clinical low dose CT image isolated using the method of the present invention (a: anatomical components; b: noise artifact components) in an embodiment of the present invention;
FIG. 7 is a sagittal conventional dose CT image of an embodiment of the present invention;
FIG. 8 is a sagittal low dose CT image of an embodiment of the present invention;
FIG. 9 is a graph of the results of a sagittal clinical low dose CT image separation using the dictionary learning method DFRDL (a: anatomical components; b: noise artifact components) in an embodiment of the present invention;
FIG. 10 shows the results of a sagittal clinical low dose CT image isolated using the method of the present invention NaSCNN (a: anatomical components; b: noise artifact components) in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
For a further understanding of the invention, reference should be made to the following detailed description taken in conjunction with the accompanying drawings and examples.
Example 1
Referring to fig. 1 to 10, the method for decomposing a low-dose CT image based on a convolutional neural network of the present embodiment includes the following steps:
step 1, respectively obtaining a plurality of groups of matched low-dose CT projection data and conventional-dose CT projection data (in specific operation, the scanning current in low dose is five times of the scanning current in conventional doseOne-third to one-fourth) and respectively reconstruct corresponding training images: low dose CT imageAnd conventional dose CT imagesLow dose CT imageAnd conventional dose CT imagesSubtracting to obtain a noise artifact image
Step 1, low dose CT imagingIs obtained by an analytic FBP reconstruction algorithm and is a conventional dosage CT imageThe method is obtained through an iterative TV (total variation) reconstruction algorithm, and particularly, a specific training data set is used, for example, low-dose CT imaging is carried out on the abdomen, a large number of matched abdomen projections can be utilized, and other parameters are the same except that the scanning current is different (such as scanning tube voltage, scanning angle and voxel size). Firstly, CT projection data under conventional dose scanning is subjected to TV iterative reconstruction algorithm to obtain high-quality CT image dataWherein the TV reconstruction is defined as:
wherein G is a projection matrix, u is a reconstructed CT image, and W is a statistical weight of projection dataCalculated as the variance of the projection data p; i | · | purple windWIs weighted by L2Norm, lambda is a regularization parameter, TV (u) is a TV regularization constraint term, and the formula (1) obtains a reconstructed image in an alternating solution modeThen, the CT projection data under low dose scanning is processed by the traditional analysis reconstruction algorithm FBP to obtain the low dose CT image dataFinally, the CT images of low dose and normal dose are subtracted to obtain a noise artifact image
Step 2, constructing a low-dose CT imageAnd noise artifact image NsA convolutional neural network, wherein the convolutional neural network comprises the following three different convolutional modules: the system comprises a CBR module, a branch module and a residual error module, so as to extract artifact and noise characteristic information in the low-dose CT image.
In particular, with low dose CT imagesAs sample data, a noise artifact image NsThe mapping transformation network from the low-dose image to the noise artifact image in three-dimensional end-to-end is designed as a training set of label data to estimate the low-dose CT imageThe noise artifact component of (1). This Network we call Noise-artifact Separation Convolutional Neural Network (NaSCNN), as shown in FIG. 1. The NaSCNN network comprises three different convolution modules which are respectively a CBR module, a branch module and a residual error module, wherein the CBR moduleThe block is a combination of convolution, scale transformation and ReLU activation function operation, and mainly has the function of extracting low-layer characteristic information of the low-dose CT image; the branch module is the sum of two parallel branches, the first branch is two convolution operations, the second branch is the combination of convolution, scale transformation and ReLU activation function operation, and the branch module is mainly used for mixing the characteristics extracted by different convolution kernels by increasing the network width so as to improve the representation capability of the network; the residual error module is the sum of two series-connected scale transformations, a ReLU activation function and convolution operation and one convolution operation, and mainly has the effects of shortening training time, reducing the redundancy of a characteristic convolution kernel under the same representation capability and avoiding gradient dispersion in training.
Step 3, using a certain amount of low-dose CT imagesWith corresponding noise-artifact images NsTraining the constructed convolutional neural network; wherein the training sample for training the convolutional neural network is a low-dose CT imageImage block set ofLabeled as the actual corresponding noise artifact image NsImage block set ofLow dose CT imageOutputting a noise artifact prediction image in an input convolutional neural networkParameters in the convolutional neural network are updated through training, so that noise artifact predicted images output by the convolutional neural network are reducedNoise artifact image N corresponding to realitysWhen the change of the mean square error before and after the training period is less than 0.1 percent, the training is finished; and finally obtaining the neural network with strong generalization capability through inputting mass projection data.
In particular, low dose CT images in a training setSum noise artifact image NsAccording to certain size n × n × t and pixel interval l1×l2×l3Performing tile extraction (e.g., tile size of 65 × 65 × 32 pixels, and tile interval of 12 × 12 × 12 pixels) to obtain tile setAndimage block setAndput into the network, by reducing the neural network Loss function Loss, that is, the set of predicted noise artifact image blocksWith actual noise artifact patch setThe mean square error of the neural network is used for training and learning parameters in NaSCNN, and finally the neural network with strong generalization capability is obtained. The Loss function Loss is defined as:
step 4, processing by using the trained convolutional neural networkSelected low dose CT imagesEnabling selected low dose CT imagesDecomposition of mid-anatomical structure components and noise artifact structure components:
selecting a low-dose CT image Vt ldInputting into the trained convolutional neural network to obtain a low-dose CT image V based on the selectiont ldNoise artifact prediction image ofAnd valid anatomical structure component images Vt pIn particular, the anatomical structure constituent partial images Vt pExpressed as:
wherein, Vt ldFor the selected low-dose CT image,the image is predicted for noise artifacts based on the above selected low dose CT image.
Specifically, first, a low-dose CT image V to be processed is sett ldAccording to size n × n × t and pixel spacing l1×l2×l3Partitioning to obtain an image block set Pt ld(ii) a Then, P is addedt ldInputting the NaSCNN after training to obtain a predicted noise artifact image block setNext, according to the pixel interval l1×l2×l3Image block setCombined into noise-artifact imagesFinally, the low-dose CT image V is processedt ldAnd noise artifact imagesSubtracting to obtain an effective anatomical structure component image Vt pThe relation can be expressed as:
criteria for evaluation of effects
Firstly, a plurality of groups of abdominal data are obtained, data published by a Low dose Challenge game used in the experiment are obtained from a Somatom Definition AS + CT device, the specific scanning parameters are that the tube voltage is 100KVp, the tube current is 360mAs (conventional dose)/85 mAs (Low dose), the detector size is 736 × 64, and the size of each detector unit is 1.2856 × 1.0947mm2The distances from the source to the center of the object and the center of the detector are 59.5cm and 108.56cm respectively, 1152 projection data are acquired in each circle in the full-angle mode, the screw pitch is 0.6, other parameters adopt default values of a machine, a reconstructed image is obtained after FDK (Feldkamp, Davis, Kress Algorithm) and TV reconstruction respectively, the size of the reconstructed image is 512 × 512, and the size of a pixel is 0.8 × 0.8.8 mm2The layer thickness is 1mm, and the three-dimensional continuity is better.
Selecting nine groups of scanning data as training data, wherein five typical axial training images are shown in FIG. 2; a set of scan data was used as test data, with conventional dose CT images selected as shown in fig. 3 (axial) and 7 (sagittal) and low dose CT images selected as shown in fig. 4 (axial) and 8 (sagittal). The window widths of the low-dose CT image, the normal-dose CT image and the decomposed anatomical structure components are 300HU (Housfield Units, HU), and the window level is 50 HU; the window width of the noise artifact component is 200HU and the window level is-1000 HU.
Visual assessment
By observing the conventional dose and low dose CT images shown in FIGS. 3-10, and the images decomposed by the conventional DFRDL method and the method of the present invention, it can be seen that although the conventional DFRDL method can completely decompose noise and streak artifacts, in the processing process, the anatomical structure components lose part of tissue details, and part of the regions have certain blurring, such as liver, spleen vein vessels and vessel cyst regions; the quality of the characteristic image decomposed by the method is obviously improved, the anatomical structure components and the noise artifact components are effectively decomposed, the decomposed anatomical structure components contain less noise and artifacts, and meanwhile, the method has better tissue distinguishing capability, can well keep the edges and fine structures of the anatomical structure, and the visual texture of the image is closer to that of a CT image under the conventional dose.
Quantitative evaluation
In order to quantitatively verify the effectiveness of the method, the peak signal-to-noise ratio and the structural similarity of a plurality of images (a low-dose CT image, an anatomical structure component after DFRDL decomposition and an anatomical structure component after NaSCNN decomposition of the invention) and a conventional dose CT image are compared through calculation, wherein the peak signal-to-noise ratio PSNR is defined as:
wherein I represents a normal dose CT image, K represents an image to be calculated including K, and LIThe maximum image pixel value that the representative image I can represent, I, j are the pixel indices of the image, respectively, and m, n are the length and width of the image, respectively.
The structural similarity SSIM is defined as:
wherein muI、μKAre the mean values, σ, of the images I, K, respectivelyI、σKAre respectivelyStandard deviation of I, K, σIKIs the covariance of images I and K, C1And C2Is two constants, wherein C1=(0.01×L)2,C2=(0.03×L)2. From the following table 1, it can be seen that the decomposition method of the present invention can greatly reduce the noise in the decomposed anatomical structure components, improve the signal-to-noise ratio, and obtain a CT image closer to the normal dose.
TABLE 1
As can be seen from the above experiments, the method of the invention can effectively decompose the anatomical structure components and the noise artifact structure components in the low-dose CT image, obtain the human anatomy structure image of the CT information close to the normal dose level under the low-dose condition, and reduce the interference of the noise artifact on the analysis and diagnosis of the clinician; in addition, the low-dose CT image decomposition method based on the convolutional neural network has the advantages that the neural network does not need to be trained repeatedly once being built, the actual processing time is short, the speed is high, and the method has a wide application range.
Compared with the prior art, the invention discloses a low-dose CT image decomposition method based on a convolutional neural network, which comprises the steps of firstly obtaining a plurality of matched CT projection data with low dose and conventional dose, reconstructing a corresponding training image, and subtracting the CT images with low dose and conventional dose to obtain a noise artifact image; secondly, constructing a mapping convolution neural network between the low-dose CT image and the noise artifact image, wherein the network comprises three different convolution modules so as to extract artifact and noise characteristic information in the low-dose CT image; then, training the constructed convolutional neural network by using a certain amount of low-dose CT images and noise artifact images; finally, the trained convolutional neural network is used for processing the low-dose CT image, so that the effective decomposition of anatomical structure components and noise artifact structure components in the low-dose CT image is realized; the low-dose CT image decomposition method can effectively distinguish star-strip artifacts and noise in low-dose CT data from human anatomy structures, has a processing effect superior to that of a traditional dictionary learning method, can meet the requirements of clinical analysis, diagnosis and the like on image quality, and improves the image effect of low-dose CT imaging.
The new convolutional neural network constructed by the invention inhibits noise artifacts from the decomposition angle, and the convolutional neural network is added with branch modules on the basis of a residual error network, thereby improving the width of the network and increasing the generalization capability of the network; in addition, the invention adopts the iterative reconstruction image as the label, can reduce the influence of noise artifact components in the sample data on the network training to a certain extent, so as to improve the capability of the network for extracting the noise artifact characteristics, thereby achieving the effect of separating the characteristic structure components and the noise artifact components in the low-dose CT image.
The low-dose CT image decomposition method based on the convolutional neural network is characterized in that an image space is modeled based on a deep learning method, artifacts and noise are distinguished from human anatomical structures through the powerful feature representation capability of the convolutional neural network, the test time is short, the processing effect is good, the powerful representation capability of the convolutional network is fully utilized, and the low-dose CT image noise artifacts and the decomposition between the low-dose CT image noise artifacts and the anatomical structures are realized.
The low-dose CT image decomposition apparatus based on the convolutional neural network of the present embodiment includes:
the image acquisition module is used for respectively acquiring a plurality of groups of matched low-dose CT projection data and conventional-dose CT projection data and respectively reconstructing corresponding training images: low dose CT image Vs ldAnd conventional dose CT imagesLow dose CT imageAnd conventional dose CT imagesSubtracting to obtain a noise artifact image
Convolutional neural network construction module for constructing low-dose CT imageAnd noise artifact image NsA mapping convolutional neural network therebetween;
training module for using a volume of low dose CT imagesWith corresponding noise-artifact images NsTraining the constructed convolutional neural network;
low-dose CT image decomposition module for processing selected low-dose CT images using trained convolutional neural networkEnabling selected low dose CT imagesThe decomposition of the medium anatomical structure components with the noise artifact structure components.
Low dose CT image in image acquisition moduleIs obtained by an analytic FBP reconstruction algorithm and is a conventional dosage CT imageIs obtained by an iterative TV reconstruction algorithm.
In the convolutional neural network construction module, the convolutional neural network comprises the following three convolutional modules: the device comprises a CBR module, a branch module and a residual module.
The CBR module is a combination of convolution, scale transformation and ReLU activation function operation; the branch module is the sum of two parallel branches, the first branch is two convolution operations, and the second branch is the combination of convolution, scale transformation and ReLU activation function operation; the residual module is the sum of two series-connected scale transformations, a ReLU activation function and a convolution operation and one convolution operation.
In the training module, low-dose CT images are acquiredOutputting a noise artifact prediction image in an input convolutional neural networkUpdating parameters in the convolutional neural network through training to reduce noise artifact predicted image output by the convolutional neural networkNoise artifact image N corresponding to realitysWhen the change of the mean square error before and after the training period is less than 0.1%, the training is finished.
In the low-dose CT image decomposition module, the selected low-dose CT image V ist ldInputting into the trained convolutional neural network to obtain a low-dose CT image V based on the selectiont ldNoise artifact prediction image ofAnd anatomical structure component image Vt p。
In the low-dose CT image decomposition module, the anatomical structure forms a partial image Vt pExpressed as:
wherein, Vt ldFor the selected low-dose CT image,the image is predicted for noise artifacts based on the above selected low dose CT image.
In the training module, the training sample for training the convolutional neural network is a low-dose CT imageImage block set ofLabeled as the actual corresponding noise artifact image NsImage block set of
The low-dose CT image decomposition device of the present embodiment includes: the data acquisition equipment is used for acquiring and storing CT projection data; a computer for receiving CT projection data; the computer is programmed to perform the steps of:
step 1, respectively obtaining a plurality of groups of matched low-dose CT projection data and conventional-dose CT projection data, and respectively reconstructing corresponding training images: low dose CT imageAnd conventional dose CT imagesLow dose CT imageAnd conventional dose CT imagesSubtracting to obtain a noise artifact image
Step 2, constructing a low-dose CT imageAnd noise artifact image NsMapping convolutional neural network between;
Step 3, using a certain amount of low-dose CT imagesWith corresponding noise-artifact images NsTraining the constructed convolutional neural network;
step 4, processing the selected low-dose CT image by using the trained convolutional neural networkEnabling selected low dose CT imagesThe decomposition of the medium anatomical structure components with the noise artifact structure components.
Step 1, low dose CT imagingIs obtained by an analytic FBP reconstruction algorithm and is a conventional dosage CT imageIs obtained by an iterative TV reconstruction algorithm.
In step 2, the convolutional neural network comprises the following three convolutional modules: the device comprises a CBR module, a branch module and a residual module.
The CBR module is a combination of convolution, scale transformation and ReLU activation function operation; the branch module is the sum of two parallel branches, the first branch is two convolution operations, and the second branch is the combination of convolution, scale transformation and ReLU activation function operation; the residual module is the sum of two series-connected scale transformations, a ReLU activation function and a convolution operation and one convolution operation.
In step 3, the low dose CT image is takenOutputting a noise artifact prediction image in an input convolutional neural networkUpdating parameters in the convolutional neural network through training to reduce noise artifact predicted image output by the convolutional neural networkNoise artifact image N corresponding to realitysWhen the change of the mean square error before and after the training period is less than 0.1%, the training is finished.
In step 4, the selected low-dose CT image V is processedt ldInputting into the trained convolutional neural network to obtain a low-dose CT image V based on the selectiont ldNoise artifact prediction image ofAnd anatomical structure component image Vt p。
In step 4, the anatomical structure constitutes a partial image Vt pExpressed as:
wherein, Vt ldFor the selected low-dose CT image,the image is predicted for noise artifacts based on the above selected low dose CT image.
In step 3, the training sample for training the convolutional neural network is a low-dose CT imageImage block set ofLabeled as the actual corresponding noise artifact image NsImage block set of
In this embodiment, the data acquisition device is a CT machine, which obtains different CT projection data by shooting different tissues to provide a training image to a computer. In an alternative embodiment, the data acquisition device is a data storage medium having different CT projection data stored thereon for providing training images to the computer.
The computer-readable storage medium of the present embodiment stores a computer program that executes, using a computer, the steps of:
step 1, respectively obtaining a plurality of groups of matched low-dose CT projection data and conventional-dose CT projection data, and respectively reconstructing corresponding training images: low dose CT imageAnd conventional dose CT imagesLow dose CT imageAnd conventional dose CT imagesSubtracting to obtain a noise artifact image
Step 2, constructing a low-dose CT imageAnd noise artifact image NsA mapping convolutional neural network therebetween;
step 3, using a certain amount of low-dose CT imagesWith corresponding noise-artifact images NsTraining the constructed convolutional neural network;
step 4, processing the selected low-dose CT image by using the trained convolutional neural networkEnabling selected low dose CT imagesThe decomposition of the medium anatomical structure components with the noise artifact structure components.
Step 1, low dose CT imagingIs obtained by an analytic FBP reconstruction algorithm and is a conventional dosage CT imageIs obtained by an iterative TV reconstruction algorithm.
In step 2, the convolutional neural network comprises the following three convolutional modules: the device comprises a CBR module, a branch module and a residual module.
The CBR module is a combination of convolution, scale transformation and ReLU activation function operation; the branch module is the sum of two parallel branches, the first branch is two convolution operations, and the second branch is the combination of convolution, scale transformation and ReLU activation function operation; the residual module is the sum of two series-connected scale transformations, a ReLU activation function and a convolution operation and one convolution operation.
In step 3, the low dose CT image is takenOutputting a noise artifact prediction image in an input convolutional neural networkUpdating parameters in the convolutional neural network through training to reduce noise artifact predicted image output by the convolutional neural networkIn accordance with the realityNoise artifact image NsWhen the change of the mean square error before and after the training period is less than 0.1%, the training is finished.
In step 4, the selected low-dose CT image V is processedt ldInputting into the trained convolutional neural network to obtain a low-dose CT image V based on the selectiont ldNoise artifact prediction image ofAnd anatomical structure component image Vt p。
In step 4, the anatomical structure constitutes a partial image Vt pExpressed as:
wherein, Vt ldFor the selected low-dose CT image,the image is predicted for noise artifacts based on the above selected low dose CT image.
In step 3, the training sample for training the convolutional neural network is a low-dose CT imageImage block set ofLabeled as the actual corresponding noise artifact image NsImage block set of
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (3)
1. A low-dose CT image decomposition method based on a convolutional neural network is characterized by comprising the following steps:
step 1, respectively obtaining a plurality of groups of matched low-dose CT projection data and conventional-dose CT projection data, and respectively reconstructing corresponding training images: low dose CT imageAnd conventional dose CT imagesLow dose CT imageAnd conventional dose CT imagesSubtracting to obtain a noise artifact image
Step 1, low dose CT imagingIs obtained by an analytic FBP reconstruction algorithm and is a conventional dosage CT imageIs obtained by an iterative TV reconstruction algorithm;
step 2, constructing a low-dose CT imageAnd noise artifact imagesNsA mapping convolutional neural network therebetween;
in step 2, the convolutional neural network comprises the following three convolutional modules: the system comprises a CBR module, a branch module and a residual error module;
the CBR module is a combination of convolution, scale transformation and ReLU activation function operation; the branch module is the sum of two parallel branches, the first branch is two convolution operations, and the second branch is the combination of convolution, scale transformation and ReLU activation function operation; the residual error module is the sum of two series-connected scale transformations, a ReLU activation function and convolution operation and one convolution operation;
step 3, using a certain amount of low-dose CT imagesWith corresponding noise-artifact images NsTraining the constructed convolutional neural network;
in step 3, the low dose CT image is takenOutputting a noise artifact prediction image in an input convolutional neural networkUpdating parameters in the convolutional neural network through training to reduce noise artifact predicted image output by the convolutional neural networkNoise artifact image N corresponding to realitysWhen the change of the mean square error before and after the training period is less than 0.1 percent, the training is finished;
step 4, processing the selected low-dose CT image by using the trained convolutional neural networkEnabling selected low dose CT imagesDecomposition of medium anatomical structure components and noise artifact structure components;
2. The convolutional neural network based low dose CT image decomposition method of claim 1, wherein in step 4, the anatomical structure component image Vt pExpressed as:
3. The convolutional neural network based low-dose CT image decomposition method as claimed in claim 1 or 2, wherein in step 3, the training sample for training the convolutional neural network is the low-dose CT imageImage block set ofLabeled as the actual corresponding noise artifact image NsImage block set of
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810706749.6A CN108961237B (en) | 2018-06-28 | 2018-06-28 | Low-dose CT image decomposition method based on convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810706749.6A CN108961237B (en) | 2018-06-28 | 2018-06-28 | Low-dose CT image decomposition method based on convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108961237A CN108961237A (en) | 2018-12-07 |
CN108961237B true CN108961237B (en) | 2020-08-21 |
Family
ID=64484742
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810706749.6A Active CN108961237B (en) | 2018-06-28 | 2018-06-28 | Low-dose CT image decomposition method based on convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108961237B (en) |
Families Citing this family (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110009613A (en) * | 2019-03-28 | 2019-07-12 | 东南大学 | Low-dose CT imaging method, apparatus and system based on the dense network of depth |
CN109949215B (en) * | 2019-03-29 | 2023-03-31 | 浙江明峰智能医疗科技有限公司 | Low-dose CT image simulation method |
CN110097611B (en) * | 2019-04-28 | 2023-09-22 | 上海联影智能医疗科技有限公司 | Image reconstruction method, device, equipment and storage medium |
CN110176045A (en) * | 2019-05-05 | 2019-08-27 | 东南大学 | A method of dual-energy CT image is generated by single energy CT image |
CN110517194A (en) * | 2019-07-19 | 2019-11-29 | 深圳安科高技术股份有限公司 | A kind of training method of substance decomposition model, substance decomposition method and relevant device |
CN111091516B (en) * | 2019-12-24 | 2021-06-04 | 广州柏视医疗科技有限公司 | Anti-scattering grating method and device based on artificial intelligence |
CN111192224B (en) * | 2020-01-13 | 2024-03-19 | 北京联合大学 | Image enhancement method, image enhancement device, electronic equipment and computer readable storage medium |
CN111325695B (en) * | 2020-02-29 | 2023-04-07 | 深圳先进技术研究院 | Low-dose image enhancement method and system based on multi-dose grade and storage medium |
CN111951346B (en) * | 2020-08-18 | 2023-06-23 | 安徽工程大学 | 4D-CBCT reconstruction method combining motion estimation and space-time tensor enhancement representation |
CN112348936B (en) * | 2020-11-30 | 2023-04-18 | 华中科技大学 | Low-dose cone-beam CT image reconstruction method based on deep learning |
CN112258642B (en) * | 2020-12-21 | 2021-04-09 | 之江实验室 | Low-dose PET data three-dimensional iterative updating reconstruction method based on deep learning |
US11890124B2 (en) | 2021-02-01 | 2024-02-06 | Medtronic Navigation, Inc. | Systems and methods for low-dose AI-based imaging |
CN113053496B (en) * | 2021-03-19 | 2023-08-29 | 深圳高性能医疗器械国家研究院有限公司 | Deep learning method for low-dose estimation of medical image |
CN113012252A (en) * | 2021-03-24 | 2021-06-22 | 苏州深透智能科技有限公司 | SPECT imaging prediction model creation method, device, equipment and storage medium |
CN113256752B (en) * | 2021-06-07 | 2022-07-26 | 太原理工大学 | Low-dose CT reconstruction method based on double-domain interleaving network |
CN113379868A (en) * | 2021-07-08 | 2021-09-10 | 安徽工程大学 | Low-dose CT image noise artifact decomposition method based on convolution sparse coding network |
CN113436118B (en) * | 2021-08-10 | 2022-09-27 | 安徽工程大学 | Low-dose CT image restoration method based on multi-scale convolutional coding network |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101412575B1 (en) * | 2012-07-02 | 2014-06-27 | 충남대학교산학협력단 | Cone beam CT apparatus using low dose x-ray |
CN105118066A (en) * | 2015-09-16 | 2015-12-02 | 东南大学 | Low-dosage CT image decomposition method based on three-dimensional distinctive feature representation |
CN106600568A (en) * | 2017-01-19 | 2017-04-26 | 沈阳东软医疗系统有限公司 | Low-dose CT image denoising method and device |
CN106778918A (en) * | 2017-01-22 | 2017-05-31 | 北京飞搜科技有限公司 | A kind of deep learning image identification system and implementation method for being applied to mobile phone terminal |
CN107844743A (en) * | 2017-09-28 | 2018-03-27 | 浙江工商大学 | A kind of image multi-subtitle automatic generation method based on multiple dimensioned layering residual error network |
CN107945132A (en) * | 2017-11-29 | 2018-04-20 | 深圳安科高技术股份有限公司 | A kind of artifact correction method and device of the CT images based on neutral net |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
BR112013009965A8 (en) * | 2010-10-27 | 2016-09-13 | Koninklijke Philips Electronics Nv | "SYSTEM AND METHOD" |
KR101636207B1 (en) * | 2015-05-07 | 2016-07-05 | 한국과학기술원 | Development of iterative reconstruction framework using analytic principle for low dose x-ray ct |
-
2018
- 2018-06-28 CN CN201810706749.6A patent/CN108961237B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101412575B1 (en) * | 2012-07-02 | 2014-06-27 | 충남대학교산학협력단 | Cone beam CT apparatus using low dose x-ray |
CN105118066A (en) * | 2015-09-16 | 2015-12-02 | 东南大学 | Low-dosage CT image decomposition method based on three-dimensional distinctive feature representation |
CN106600568A (en) * | 2017-01-19 | 2017-04-26 | 沈阳东软医疗系统有限公司 | Low-dose CT image denoising method and device |
CN106778918A (en) * | 2017-01-22 | 2017-05-31 | 北京飞搜科技有限公司 | A kind of deep learning image identification system and implementation method for being applied to mobile phone terminal |
CN107844743A (en) * | 2017-09-28 | 2018-03-27 | 浙江工商大学 | A kind of image multi-subtitle automatic generation method based on multiple dimensioned layering residual error network |
CN107945132A (en) * | 2017-11-29 | 2018-04-20 | 深圳安科高技术股份有限公司 | A kind of artifact correction method and device of the CT images based on neutral net |
Non-Patent Citations (5)
Title |
---|
CT artifact reduction via U-net CNN;C. Zhang等;《Proc.of SPIE》;20180505;第10574卷;第1-6页 * |
Discriminative Feature Representation to Improve Projection Data Inconsistency for Low Dose CT Imaging;Jin Liu等;《IEEE Transactions on Medical Imaging》;20170815;第36卷(第12期);第2499-2509页 * |
Low-dose CBCT reconstruction via 3D dictionary learning;Jin Liu 等;《2016 IEEE 13th International Symposium on Biomedical Imaging》;20160616;第735-738页 * |
Low-dose CT denoising with convolutional neural network;Hu Chen等;《2017 IEEE 14th International Symposium on Biomedical Imaging》;20170619;第143-146页 * |
基于卷积神经网络的低剂量CT图像去噪方法;章云港等;《光学学报》;20180430;第38卷(第4期);第0410003-1—0410003-7页 * |
Also Published As
Publication number | Publication date |
---|---|
CN108961237A (en) | 2018-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108961237B (en) | Low-dose CT image decomposition method based on convolutional neural network | |
CN109166161B (en) | Low-dose CT image processing system based on noise artifact suppression convolutional neural network | |
CN108898642B (en) | Sparse angle CT imaging method based on convolutional neural network | |
Hu et al. | Artifact correction in low‐dose dental CT imaging using Wasserstein generative adversarial networks | |
Yin et al. | Domain progressive 3D residual convolution network to improve low-dose CT imaging | |
Yang et al. | Low-dose CT image denoising using a generative adversarial network with Wasserstein distance and perceptual loss | |
Li et al. | Investigation of low-dose CT image denoising using unpaired deep learning methods | |
Huang et al. | CaGAN: A cycle-consistent generative adversarial network with attention for low-dose CT imaging | |
Kuanar et al. | Low dose abdominal CT image reconstruction: an unsupervised learning based approach | |
CN109102550B (en) | Full-network low-dose CT imaging method and device based on convolution residual error network | |
Wang et al. | Multiscale penalized weighted least-squares sinogram restoration for low-dose X-ray computed tomography | |
CN105118066B (en) | A kind of low-dose CT picture breakdown method based on three-dimensional distinctiveness character representation | |
CN106373163B (en) | A kind of low-dose CT imaging method indicated based on three-dimensional projection's distinctive feature | |
Gajera et al. | CT-scan denoising using a charbonnier loss generative adversarial network | |
Tao et al. | VVBP-tensor in the FBP algorithm: its properties and application in low-dose CT reconstruction | |
CN110599530B (en) | MVCT image texture enhancement method based on double regular constraints | |
Chen et al. | Low-dose CT image denoising model based on sparse representation by stationarily classified sub-dictionaries | |
Feng et al. | Dual residual convolutional neural network (DRCNN) for low-dose CT imaging | |
Li et al. | Unpaired low‐dose computed tomography image denoising using a progressive cyclical convolutional neural network | |
Rahim et al. | Total variant based average sparsity model with reweighted analysis for compressive sensing of computed tomography | |
CN117876261A (en) | CBCT scattering correction imaging method based on deep learning | |
CN111080736B (en) | Low-dose CT image reconstruction method based on sparse transformation | |
Ding et al. | A dataset-free deep learning method for low-dose CT image reconstruction | |
Liang et al. | A model-based unsupervised deep learning method for low-dose CT reconstruction | |
Liu et al. | Cascade resunet with noise power spectrum loss for low dose ct imaging |
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 |