CN108961237A - A kind of low-dose CT picture breakdown method based on convolutional neural networks - Google Patents
A kind of low-dose CT picture breakdown method based on convolutional neural networks Download PDFInfo
- Publication number
- CN108961237A CN108961237A CN201810706749.6A CN201810706749A CN108961237A CN 108961237 A CN108961237 A CN 108961237A CN 201810706749 A CN201810706749 A CN 201810706749A CN 108961237 A CN108961237 A CN 108961237A
- Authority
- CN
- China
- Prior art keywords
- image
- dose
- low
- convolutional neural
- neural networks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000015556 catabolic process Effects 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 57
- 239000004615 ingredient Substances 0.000 claims abstract description 35
- 210000003484 anatomy Anatomy 0.000 claims abstract description 28
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 20
- 239000000470 constituent Substances 0.000 claims abstract description 16
- 238000013507 mapping Methods 0.000 claims abstract description 9
- 230000004913 activation Effects 0.000 claims description 18
- 230000008859 change Effects 0.000 claims description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 15
- 238000002224 dissection Methods 0.000 claims description 12
- 230000001537 neural effect Effects 0.000 claims description 4
- 238000002591 computed tomography Methods 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000013528 artificial neural network Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 10
- 238000012545 processing Methods 0.000 description 8
- 230000006872 improvement Effects 0.000 description 7
- 238000000926 separation method Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 4
- 210000001015 abdomen Anatomy 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000013170 computed tomography imaging Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000003672 processing method Methods 0.000 description 3
- 230000005855 radiation Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 208000019838 Blood disease Diseases 0.000 description 1
- 206010011732 Cyst Diseases 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 230000003187 abdominal effect Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 208000031513 cyst Diseases 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000009415 formwork Methods 0.000 description 1
- 208000014951 hematologic disease Diseases 0.000 description 1
- 208000018706 hematopoietic system disease Diseases 0.000 description 1
- 231100000405 induce cancer Toxicity 0.000 description 1
- 210000004185 liver Anatomy 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001613 neoplastic effect Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000007115 recruitment Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 210000000955 splenic vein Anatomy 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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 low-dose CT picture breakdown method based on convolutional neural networks that the invention discloses a kind of, belongs to X ray computer tomography technology field.The present invention the following steps are included: step 1, reconstruct training image respectively: low-dose CT imageWith routine dose CT imageBy low-dose CT imageWith routine dose CT imageSubtract each other to obtain noise artifacts imageStep 2, building low-dose CT imageWith noise artifacts image NsBetween mapping convolutional neural networks；Step 3 uses a certain amount of low-dose CT imageWith corresponding noise artifacts image NsThe convolutional neural networks that oneself builds are trained；Step 4 handles selected low-dose CT image using trained convolutional neural networksRealize selected low-dose CT imageThe decomposition of middle anatomical structure ingredient and noise artifacts constituent.The present invention provides the methods that one kind can efficiently separate low-dose CT image culminant star strip artifact noise and structure feature.
Description
Technical field
The present invention relates to the decomposition method of low-dose CT image more particularly to a kind of low dosages based on convolutional neural networks
CT picture breakdown method belongs to X ray computer tomography technology field.
Background technique
As a kind of clinical imaging technique, X ray computer tomographic imaging (X-ray Computer Tomography,
CT) the high, advantage at low cost with its spatial resolution is wide in the supervision of disease screening, diagnosis, first aid, interventional therapy and curative effect
General use is one of current conventional effective clinical medicine diagnostic tool.However excess X-radiation irradiation may induce cancer, white
Blood disease increases other physiological risks, therefore the radiation problem in CT is also increasingly paid attention to by people.However, reducing radiation
Dosage can make the picture quality rebuild decrease sharply, and the speckle noise and strip artifact in image affect clinical analysis and examine
It is disconnected.How on the basis of guaranteeing picture quality, optimal CT diagnosis imaging is obtained with minimum dose of radiation and has become industry
Common recognition.
The current method for improving low-dose CT picture quality is broadly divided into two major classes: based on projection space data processing and base
In image space data processing.Method based on projection space data processing mainly passes through the school to low-dose CT data for projection
Just, restore to restore and denoising is to rebuild and providing more acurrate, the less data for projection of noise, to improve the quality of its reconstruction, such as
Adaptive-filtering and bilateral filtering etc..The low dosage picture quality rebuild directly is improved by image space processing technique,
Have the advantages that not depending on Raw projection data and processing speed is fast, is carried out usually using nonlinear processing method, it is such as complete to become
Poor (Total Variation), small echo (Wavelet) transform method remove artifact by keeping image edge information and make an uproar
Sound, however such method has ignored non local property important in image, it is also difficult to achieve the effect that satisfied；For another example it is based on dictionary
The rarefaction representation image processing algorithm of study, this method obtain one group of excessively complete dictionary (base) by training, make artifact and make an uproar
Sound cannot be indicated preferably, realize the purpose of removal artifact and noise, but handle overlong time.With large data sets, full-page proof
Originally universal, deep learning all has received widespread attention in industry and academia, is also gradually applied in CT image domains.
If Chen et al. uses a kind of residual coding network, the noise artifacts in CT image are considerably reduced, neoplastic disease is improved
Become the discrimination of tissue.
In the prior art, sparse representation method dictionary-based learning has been proved in the processing of low dosage abdominal CT images
In there is preferable effect, dosage is in routine dose 1/5th hereinafter, higher picture quality can be obtained.But it is this
Method needs to train the high frequency detail image of different directions, and calculation amount is excessive and time-consuming, it is difficult at practical three dimensional medical images
It is widely applied in reason system.In order to effectively inhibit three-dimensional artifact and noise, Chinese patent application 2015105909015 proposes one
Low-dose CT picture breakdown method of the kind based on three-dimensional distinctiveness character representation, utilizes characteristics dictionary and artifact and noise dictionary
The three-dimensional distinctiveness dictionary of composition indicates clinical low-dose CT image, obtain characteristics dictionary expression characteristic image and artifact and
The artifact and noise image that noise dictionary indicates can effectively filter out low dosage three to realize the decomposition of low-dose CT image
Artifact and noise in Victoria C T image, but this method be easy under the conditions of low dosage in CT image stronger star strip or
Blocky artifact is understood as the structural information of image, thus it can not effectively be inhibited, it is tight in low dosage star strip artifact
Under conditions of weight, it is easy to introduce blocky artifact in final CT image.
In conclusion star strip puppet cannot be efficiently separated by how overcoming present in existing low-dose CT image processing method
The problem of shadow and noise is the problem of urgent need to resolve in the prior art.
The information disclosed in the background technology section is intended only to increase the understanding to general background of the invention, without answering
When being considered as recognizing or imply that the information constitutes the prior art already known to those of ordinary skill in the art in any form.
Summary of the invention
1. technical problems to be solved by the inivention
Star item cannot be efficiently separated present in existing low-dose CT image processing method it is an object of the invention to overcome
The problem of shape artifact and noise, using the method for convolutional neural networks, low-dose CT image can be efficiently separated by proposing one kind
The neural network of culminant star strip artifact noise and structure feature, referred to as noise artifacts separate convolutional neural networks (Noise-
Artifact Separation Convolutional Neural Network, NaSCNN).
2. technical solution
In order to achieve the above objectives, technical solution provided by the invention are as follows:
Low-dose CT picture breakdown method based on convolutional neural networks of the invention, comprising the following steps:
Step 1, the CT data for projection for obtaining several groups matched low-dose CT data for projection and routine dose respectively, and point
Corresponding training image: low-dose CT image is not reconstructedWith routine dose CT imageBy low-dose CT imageWith
Routine dose CT imageSubtract each other to obtain noise artifacts image
Step 2, building low-dose CT imageWith noise artifacts image NsBetween mapping convolutional neural networks；
Step 3 uses a certain amount of low-dose CT imageWith corresponding noise artifacts image NsTo the convolution built
Neural network is trained；
Step 4 handles selected low-dose CT image using trained convolutional neural networksRealization is selected low
Dosage CT imageThe decomposition of middle anatomical structure ingredient and noise artifacts constituent.
As further improvement of the present invention, in step 1, low-dose CT imageIt is to rebuild to calculate by the FBP of parsing
Method obtains, routine dose CT imageIt is to be obtained by TV (Total Variation) algorithm for reconstructing of iteration.
As further improvement of the present invention, in step 2, convolutional neural networks include following three kinds of convolution modules: CBR
Module, branch module and residual error module.
As further improvement of the present invention, CBR module is the group of convolution, change of scale and the operation of ReLU activation primitive
It closes；Branch module is two the sum of branches in parallel, and first branches into two convolution operations, and second branches into convolution, ruler
The combination of degree transformation and the operation of ReLU activation primitive；Residual error module is two concatenated change of scale, ReLU activation primitive and volume
Product operation, the sum of with a convolution operation.
As further improvement of the present invention, in step 3, by low-dose CT imageIt inputs in convolutional neural networks,
Output noise artifact forecast imageThe parameter in convolutional neural networks is updated by training, to reduce convolutional Neural net
The noise artifacts forecast image of network outputWith practical corresponding noise artifacts image NsMean square error, before cycle of training
When mean square error afterwards is changed less than 0.1%, training terminates.As further improvement of the present invention, in step 4, will select
Low-dose CT image Vt ldIn the convolutional neural networks that input training is completed, obtain based on the selected low-dose CT image Vt ld
Noise artifacts forecast imageWith dissection constituent image Vt p。
As further improvement of the present invention, in step 4, anatomical structure ingredient image Vt pIt indicates are as follows:
Wherein, Vt ldFor selected low-dose CT image,It is pseudo- for the noise based on low-dose CT image selected above
Shadow forecast image.
It is low to the training sample that convolutional neural networks are trained in step 3 as further improvement of the present invention
Dosage CT imageImage block collectionLabel is practical corresponding noise artifacts image NsImage block collection
3. beneficial effect
Using technical solution provided by the invention, compared with prior art, there is following remarkable result:
(1) the low-dose CT picture breakdown method based on convolutional neural networks that the invention discloses a kind of, if obtaining first
The CT data for projection of matched low dosage and routine dose is done, and reconstructs training image, and by low dosage and routine dose
CT image subtraction is to obtain noise artifacts image；Secondly, the Map Volume between building low-dose CT image and noise artifacts image
Product neural network, which includes three kinds of different convolution modules, to extract the artifact and noise characteristic in low-dose CT image
Information；Then, the convolutional neural networks built are carried out using a certain amount of low-dose CT image and noise artifacts image
Training；Finally, trained convolutional neural networks to be handled to low-dose CT image, anatomical structure in low-dose CT image is realized
The decomposition of ingredient and noise artifacts constituent；The low-dose CT picture breakdown method can be by the star strip in low-dose CT data
Artifact and noise and human dissection institutional framework are effectively distinguished, and treatment effect is better than traditional dictionary learning method, picture quality
The requirement such as clinical analysis, diagnosis can be met, improve the image effect of low-dose CT imaging.
(2) the low-dose CT picture breakdown method of the invention based on convolutional neural networks is based on deep learning method
Image space is modeled, and artifact and noise and human dissection are distinguished by the powerful character representation ability of convolutional neural networks
Structure, the testing time is short, high treating effect, makes full use of the expression ability that convolutional network is powerful, realizes that low-dose CT image is made an uproar
Decomposition between sound artifact and anatomical structure.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is the flow chart of the low-dose CT picture breakdown method based on convolutional neural networks in the embodiment of the present invention；
Fig. 2 is five typical axially directed training figure (a1~a5: routine dose figure in the embodiment of the present invention；B1~b5: low dose
Spirogram)；
Fig. 3 is axial routine dose CT image in the embodiment of the present invention；
Fig. 4 is axial low-dose CT image in the embodiment of the present invention；
After Fig. 5 is separated for clinical low-dose CT image axial in the embodiment of the present invention using dictionary learning method DFRDL
As a result (a: anatomical structure ingredient；B: noise artifacts ingredient)；
Fig. 6 is that axial clinical low-dose CT image uses the knot after the method for the present invention NaSCNN separation in the embodiment of the present invention
Fruit (a: anatomical structure ingredient；B: noise artifacts ingredient)；
Fig. 7 is arrowhead routine dose CT image in the embodiment of the present invention；
Fig. 8 is arrowhead low-dose CT image in the embodiment of the present invention；
After Fig. 9 is separated for arrowhead clinic low-dose CT image in the embodiment of the present invention using dictionary learning method DFRDL
As a result (a: anatomical structure ingredient；B: noise artifacts ingredient)；
After Figure 10 is separated for arrowhead clinic low-dose CT image in the embodiment of the present invention using the method for the present invention NaSCNN
As a result (a: anatomical structure ingredient；B: noise artifacts ingredient).
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Therefore, below to the embodiment of the present invention provided in the accompanying drawings
Detailed description be not intended to limit the range of claimed invention, but be merely representative of selected embodiment of the invention.
Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts all
Other embodiments shall fall within the protection scope of the present invention.
To further appreciate that the contents of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments.
Embodiment 1
With reference to Fig. 1-10, the low-dose CT picture breakdown method based on convolutional neural networks of the present embodiment, including it is following
Step:
Step 1, the CT data for projection of the matched low-dose CT data for projection of acquisition several groups and routine dose is (specific respectively
When operation, 1/5th of sweep current arrive one third when sweep current is routine dose when low dosage), and reconstruct respectively
Corresponding training image: low-dose CT imageWith routine dose CT imageBy low-dose CT imageAnd routine dose
CT imageSubtract each other to obtain noise artifacts image
In step 1, low-dose CT imageIt is to be obtained by the FBP algorithm for reconstructing of parsing, routine dose CT imageIt is
It is obtained by TV (Total Variation) algorithm for reconstructing of iteration, specifically, using specific training dataset, such as to abdomen
Portion carries out low-dose CT imaging, projects using a large amount of matched abdomens, except sweep current is different, other parameters it is all the same (such as:
Scan tube voltage, scanning angle, voxel size).TV iterative approximation is passed through by the CT data for projection under routine dose scanning first
Algorithm obtains high quality CT image dataWherein TV is rebuild is defined as:
Wherein, G is projection matrix, and u is CT image reconstruction, and W is the statistical weight of data for projection, according to data for projection p's
Variance calculates；||·||WTo weight L2Norm, λ are regularization parameter, and TV (u) is TV regularization constraint item, and formula (1) passes through friendship
Reconstruction image is obtained for the mode of solutionThen, traditional analytic reconstruction is passed through by the CT data for projection under low-dose scanning
Algorithm FBP obtains low-dose CT image dataFinally, passing through the CT image subtraction of low dosage and routine dose, noise is obtained
Artifacts
Step 2, building low-dose CT imageWith noise artifacts image NsBetween mapping convolutional neural networks, wherein
Convolutional neural networks include following three kinds of different convolution modules: CBR module, branch module and residual error module, to extract low dose
Measure the artifact and noise characteristic information in CT image.
Specifically, with low-dose CT imageAs sample data, noise artifacts image NsIt is training as label data
Collection, to design three-dimensional end to end from low dosage image to the mapping transformation network noise artifacts image, to estimate low dose
Measure CT imageIn noise artifacts ingredient.We term it noise artifacts to separate convolutional neural networks (Noise- for the network
Artifact Separation Convolutional Neural Network, NaSCNN), as shown in Figure 1.NaSCNN network
In containing there are three types of different convolution modules, respectively CBR module, branch module and residual error module, wherein CBR module be convolution,
The combination of change of scale and the operation of ReLU activation primitive, main function are the low-level feature information for extracting low-dose CT image；Point
Formwork is two the sum of branches in parallel, and first branches into two convolution operations, and second branches into convolution, change of scale
With the combination of ReLU activation primitive operation, main function is by increasing network-wide, to mix different convolution kernels extractions
Feature improves the expression ability of network；Residual error module be two concatenated change of scale, ReLU activation primitive and convolution operation,
The sum of with a convolution operation, main function is to shorten the training time, reduces the redundancy of feature convolution kernel under same expression ability
Degree avoids gradient disperse in training.
Step 3 uses a certain amount of low-dose CT imageWith corresponding noise artifacts image NsTo the convolution built
Neural network is trained；It wherein, is low-dose CT image to the training sample that convolutional neural networks are trainedImage
Block collectionLabel is practical corresponding noise artifacts image NsImage block collectionBy low-dose CT imageInput convolution
In neural network, output noise artifact forecast imageThe parameter in convolutional neural networks is updated by training, to drop
The noise artifacts forecast image of low convolutional neural networks outputWith practical corresponding noise artifacts image NsMean square error,
When mean square error before and after cycle of training is changed less than 0.1%, training terminates；By the input of magnanimity data for projection, finally
Obtain the stronger neural network of generalization ability.
Specifically, by the low-dose CT image in training setWith noise artifacts image NsAccording to certain size n × n ×
T and pixel separation l1×l2×l3Carry out image block extraction (such as: segment having a size of 65 × 65 × 32 pixels, be divided into 12 between piecemeal ×
12 × 12 pixels), respectively obtain image block collectionWithBy image block collectionWithIt is put into network, by reducing nerve
Network losses function Loss, that is, the noise artifacts image block collection predictedWith actual noise artifacts block collectionIt is equal
Square error finally obtains the stronger neural network of generalization ability to train and learn the parameter in NaSCNN.Loss function Loss
Is defined as:
Step 4 handles selected low-dose CT image using trained convolutional neural networksRealization is selected low
Dosage CT imageThe decomposition of middle anatomical structure ingredient and noise artifacts constituent:
By selected low-dose CT image Vt ldIn the convolutional neural networks that input training is completed, obtain selected based on this
Low-dose CT image Vt ldNoise artifacts forecast imageWith effective anatomical structure ingredient image Vt p, specifically, dissection knot
Constitute partial image Vt pIt indicates are as follows:
Wherein, Vt ldFor selected low-dose CT image,It is pseudo- for the noise based on low-dose CT image selected above
Shadow forecast image.
Specifically, firstly, by low-dose CT image V to be treatedt ldIt is according to having a size of n × n × t and pixel separation
l1×l2×l3Piecemeal is carried out, image block collection P is obtainedt ld；Then, by Pt ldIn the NaSCNN that input training is completed, predicted
Noise artifacts image block collectionNext, according to pixel separation l1×l2×l3By image block collectionIt is combined into noise artifacts figure
PictureFinally, by low-dose CT image Vt ldWith noise artifacts imageSubtract each other, obtains effective anatomical structure ingredient image
Vt p, relational expression may be expressed as:
Recruitment evaluation criterion
Multiple groups abdomen data are obtained first, and the data that the match of Low dose Challenge used in experiment is announced are come
From Somatom Definition AS+CT equipment, specific sweep parameter are as follows: tube voltage 100KVp, tube current 360mAs
(routine dose)/85mAs (low dosage), detector size are 736 × 64, each detector cells having a size of 1.2856 ×
1.0947mm2, the distance at radiographic source to object center and detector center is respectively 59.5cm and 108.56cm, full angle mode
Under every circle acquire 1152 data for projection, screw pitch 0.6, other parameters use machine default value.Pass through FDK respectively
(Feldkamp, Davis, Kress Algorithm) and TV obtain reconstruction image after rebuilding, and reconstruction image size is 512 ×
512, pixel size is 0.8 × 0.8mm2, thickness 1mm, three-dimensional continuity is preferable.
Wherein nine groups of scan datas are as training data for selection, wherein five typical axially directed training figures are as shown in Figure 2；One
Group scan data as test data, wherein selected by routine dose CT image as shown in Fig. 3 (axial direction) and Fig. 7 (arrowhead),
Low-dose CT image is as shown in Fig. 4 (axial direction) and Fig. 8 (arrowhead).It is solved after low-dose CT image, normal dose CT image and decomposition
The window width for cuing open constituent is 300HU (Housfield Units, HU), and window position is 50HU；The window width of noise artifacts ingredient is
200HU, window position are -1000HU.
Visual assessment
By observing the routine dose of Fig. 3-10 and the CT image of low dosage, and tradition DFRDL method and side of the present invention
Image after method decomposition, it can be seen that although traditional DFRDL method can decomposite noise and strip artifact entirely, in treatment process
In, anatomical structure ingredient is lost portion of tissue details, partial region have it is certain fuzzy, as liver, splenic vein blood vessel and blood
Tubular cyst region；And the characteristic image quality after being decomposed using the method for the present invention is significantly improved, anatomical structure ingredient and noise are pseudo-
Shadow ingredient has obtained effective decomposition, and the anatomical structure ingredient after decomposition contains less noise and artifact, while having preferably
Tissue division ability, can be good at keeping anatomical tissue structure edge and tiny structure, the more adjunction of image vision texture
It is bordering on CT image under routine dose.
Quantitative evaluation
For the validity of verifying the method for the present invention of quantization, we compare multiple image (low-dose CTs by calculating
The anatomical structure ingredient after anatomical structure ingredient and NaSCNN of the present invention decomposition after figure, DFRDL decomposition) scheme with routine dose CT
Y-PSNR and structural similarity, Y-PSNR PSNR here is defined as:
Wherein, I represents normal dose CT image herein, and K represents to contain represents image to be calculated, L hereinIRepresentative image I
The maximum image pixel value that can be represented, i, j are respectively the pixel index of image, and m, n are respectively the length and width of image.
Structural similarity SSIM's is defined as:
Wherein μI、μKIt is image I, the mean value of K, σ respectivelyI、σKIt is I, the standard deviation of K, σ respectivelyIKIt is the association side of image I and K
Difference, C1And C2For two constants, wherein C1=(0.01 × L)2, C2=(0.03 × L)2.From the following table 1 it can be seen that point of the invention
The noise after decomposed in anatomical structure ingredient can be greatly lowered in solution method, improve signal-to-noise ratio, and acquisition is more nearly normal agent
The CT image of amount.
Table 1
It can be seen that, the dissection knot in low-dose CT image can be effectively decomposed using method of the invention from the above experiment
Composition point and noise artifacts constituent obtain the human dissection close to the CT information of normal dosage levels under the conditions of low dosage
Structural images reduce the interference that noise artifacts analyze and diagnose clinician；And it is of the invention based on convolutional neural networks
Low-dose CT picture breakdown method, neural network once built, be not necessarily to repetition training, the actual treatment time is short, speed
Fastly, there is biggish application range.
Compared with prior art, the invention discloses a kind of low-dose CT picture breakdown side based on convolutional neural networks
Method, obtains the CT data for projection of several matched low dosages and routine dose first, and reconstructs corresponding training image, will be low
The CT image subtraction of dosage and routine dose is to obtain noise artifacts image；Secondly, building low-dose CT image and noise artifacts
Mapping convolutional neural networks between image, which includes three kinds of different convolution modules, to extract in low-dose CT image
Artifact and noise characteristic information；Then, using a certain amount of low-dose CT image and noise artifacts image to having built
Convolutional neural networks are trained；Finally, trained convolutional neural networks to be handled to low-dose CT image, low dosage is realized
The decomposition of effective anatomical structure ingredient and noise artifacts constituent in CT image；The low-dose CT picture breakdown method can incite somebody to action
Star strip artifact and noise and human dissection institutional framework in low-dose CT data are effectively distinguished, and treatment effect is better than traditional
Dictionary learning method, picture quality can meet the requirement such as clinical analysis, diagnosis, improve the image effect of low-dose CT imaging.
The new convolutional neural networks that the present invention constructs, inhibit noise artifacts from the angle of decomposition, the convolutional Neural net
Network is to increase branch module on residual error network foundation, improves the width of network, increases its generalization ability；In addition, of the invention
Using iterative approximation image as label, noise artifacts ingredient can be reduced in sample data to a certain extent to network training
Influence, with improve network to the ability of noise artifacts feature extraction to reach to the feature structure in low-dose CT image at
Divide the effect separated with noise artifacts ingredient.
Low-dose CT picture breakdown method based on convolutional neural networks of the invention is based on deep learning method to figure
Image space modeling, and artifact and noise and human dissection knot are distinguished by the powerful character representation ability of convolutional neural networks
Structure, the testing time is short, high treating effect, makes full use of the expression ability that convolutional network is powerful, realizes low-dose CT picture noise
Decomposition between artifact and anatomical structure.
The low-dose CT picture breakdown device based on convolutional neural networks of the present embodiment, comprising:
Image collection module, the CT for obtaining the matched low-dose CT data for projection of several groups and routine dose respectively are thrown
Shadow data, and corresponding training image: low-dose CT image V is reconstructed respectivelys ldWith routine dose CT imageBy low dosage
CT imageWith routine dose CT imageSubtract each other to obtain noise artifacts image
Convolutional neural networks construct module, for constructing low-dose CT imageWith noise artifacts image NsBetween mapping
Convolutional neural networks；
Training module is used to use a certain amount of low-dose CT imageWith corresponding noise artifacts image NsTo having built
Convolutional neural networks be trained；
Low-dose CT picture breakdown module, the low-dose CT figure for being selected using the processing of trained convolutional neural networks
PictureRealize selected low-dose CT imageThe decomposition of middle anatomical structure ingredient and noise artifacts constituent.
In image collection module, low-dose CT imageIt is to be obtained by the FBP algorithm for reconstructing of parsing, routine dose CT
ImageIt is to be obtained by the TV algorithm for reconstructing of iteration.
Convolutional neural networks construct in module, and convolutional neural networks include following three kinds of convolution modules: CBR module, branch
Module and residual error module.
CBR module is the combination of convolution, change of scale and the operation of ReLU activation primitive；Branch module is two points in parallel
The sum of branch, first branches into two convolution operations, and second branches into convolution, change of scale and the operation of ReLU activation primitive
Combination；Residual error module is two concatenated change of scale, ReLU activation primitive and convolution operation, the sum of with a convolution operation.
In training module, by low-dose CT imageIt inputs in convolutional neural networks, output noise artifact forecast imageThe parameter in convolutional neural networks is updated by training, to reduce the noise artifacts prognostic chart of convolutional neural networks output
PictureWith practical corresponding noise artifacts image NsMean square error, before and after cycle of training mean square error variation is less than
When 0.1%, training terminates.
In low-dose CT picture breakdown module, by selected low-dose CT image Vt ldThe convolutional Neural that input training is completed
In network, obtain based on the selected low-dose CT image Vt ldNoise artifacts forecast imageWith dissection constituent image
Vt p。
In low-dose CT picture breakdown module, anatomical structure ingredient image Vt pIt indicates are as follows:
Wherein, Vt ldFor selected low-dose CT image,It is pseudo- for the noise based on low-dose CT image selected above
Shadow forecast image.
It is low-dose CT image to the training sample that convolutional neural networks are trained in training moduleImage block
CollectionLabel is practical corresponding noise artifacts image NsImage block collection
The low-dose CT picture breakdown device of the present embodiment, decomposer include: for obtaining, storing CT data for projection
Data acquisition equipment；For receiving the computer of CT data for projection；Computer is programmed to execute following steps:
Step 1, the CT data for projection for obtaining several groups matched low-dose CT data for projection and routine dose respectively, and point
Corresponding training image: low-dose CT image is not reconstructedWith routine dose CT imageBy low-dose CT imageWith
Routine dose CT imageSubtract each other to obtain noise artifacts image
Step 2, building low-dose CT imageWith noise artifacts image NsBetween mapping convolutional neural networks；
Step 3 uses a certain amount of low-dose CT imageWith corresponding noise artifacts image NsTo the convolution built
Neural network is trained；
Step 4 handles selected low-dose CT image using trained convolutional neural networksRealization is selected low
Dosage CT imageThe decomposition of middle anatomical structure ingredient and noise artifacts constituent.
In step 1, low-dose CT imageIt is to be obtained by the FBP algorithm for reconstructing of parsing, routine dose CT image
It is to be obtained by the TV algorithm for reconstructing of iteration.
In step 2, convolutional neural networks include following three kinds of convolution modules: CBR module, branch module and residual error module.
CBR module is the combination of convolution, change of scale and the operation of ReLU activation primitive；Branch module is two points in parallel
The sum of branch, first branches into two convolution operations, and second branches into convolution, change of scale and the operation of ReLU activation primitive
Combination；Residual error module is two concatenated change of scale, ReLU activation primitive and convolution operation, the sum of with a convolution operation.
In step 3, by low-dose CT imageIt inputs in convolutional neural networks, output noise artifact forecast image
The parameter in convolutional neural networks is updated by training, to reduce the noise artifacts forecast image of convolutional neural networks outputWith practical corresponding noise artifacts image NsMean square error, mean square error before and after cycle of training changes less than 0.1%
When, training terminates.
In step 4, by selected low-dose CT image Vt ldIn the convolutional neural networks that input training is completed, it is based on
The selected low-dose CT image Vt ldNoise artifacts forecast imageWith dissection constituent image Vt p。
In step 4, anatomical structure ingredient image Vt pIt indicates are as follows:
Wherein, Vt ldFor selected low-dose CT image,It is pseudo- for the noise based on low-dose CT image selected above
Shadow forecast image.
It is low-dose CT image to the training sample that convolutional neural networks are trained in step 3Image block collectionLabel is practical corresponding noise artifacts image NsImage block collection
In specific the present embodiment, data acquisition equipment is CT machine, different by carrying out shooting acquisition to different tissues
CT data for projection, to provide training image to computer.Or in embodiment, data acquisition equipment is data storage medium, should
Different CT data for projection is stored on data storage medium, to provide training image to computer.
The computer readable storage medium for being stored with computer program of the present embodiment, computer program are held using computer
Row following steps:
Step 1, the CT data for projection for obtaining several groups matched low-dose CT data for projection and routine dose respectively, and point
Corresponding training image: low-dose CT image is not reconstructedWith routine dose CT imageBy low-dose CT imageWith
Routine dose CT imageSubtract each other to obtain noise artifacts image
Step 2, building low-dose CT imageWith noise artifacts image NsBetween mapping convolutional neural networks；
Step 3 uses a certain amount of low-dose CT imageWith corresponding noise artifacts image NsTo the convolution built
Neural network is trained；
Step 4 handles selected low-dose CT image using trained convolutional neural networksRealization is selected low
Dosage CT imageThe decomposition of middle anatomical structure ingredient and noise artifacts constituent.
In step 1, low-dose CT imageIt is to be obtained by the FBP algorithm for reconstructing of parsing, routine dose CT image
It is to be obtained by the TV algorithm for reconstructing of iteration.
In step 2, convolutional neural networks include following three kinds of convolution modules: CBR module, branch module and residual error module.
CBR module is the combination of convolution, change of scale and the operation of ReLU activation primitive；Branch module is two points in parallel
The sum of branch, first branches into two convolution operations, and second branches into convolution, change of scale and the operation of ReLU activation primitive
Combination；Residual error module is two concatenated change of scale, ReLU activation primitive and convolution operation, the sum of with a convolution operation.
In step 3, by low-dose CT imageIt inputs in convolutional neural networks, output noise artifact forecast image
The parameter in convolutional neural networks is updated by training, to reduce the noise artifacts forecast image of convolutional neural networks outputWith practical corresponding noise artifacts image NsMean square error, mean square error before and after cycle of training changes less than 0.1%
When, training terminates.
In step 4, by selected low-dose CT image Vt ldIn the convolutional neural networks that input training is completed, it is based on
The selected low-dose CT image Vt ldNoise artifacts forecast imageWith dissection constituent image Vt p。
In step 4, anatomical structure ingredient image Vt pIt indicates are as follows:
Wherein, Vt ldFor selected low-dose CT image,For the noise artifacts based on low-dose CT image selected above
Forecast image.
It is low-dose CT image to the training sample that convolutional neural networks are trained in step 3Image block collectionLabel is practical corresponding noise artifacts image NsImage block collection
The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although with reference to the foregoing embodiments
Invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each implementation
Technical solution documented by example is modified or equivalent replacement of some of the technical features；And these modification or
Replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution.
Claims (8)
1. a kind of low-dose CT picture breakdown method based on convolutional neural networks, which comprises the following steps:
Step 1, the CT data for projection for obtaining several groups matched low-dose CT data for projection and routine dose respectively, and weigh respectively
Build out corresponding training image: low-dose CT image Vs ldWith routine dose CT image Vs rd, by low-dose CT image Vs ldAnd routine
Dosage CT image Vs rdSubtract each other to obtain noise artifacts image
Step 2, building low-dose CT image Vs ldWith noise artifacts image NsBetween mapping convolutional neural networks；
Step 3 uses a certain amount of low-dose CT image Vs ldWith corresponding noise artifacts image NsThe convolutional Neural that oneself is built
Network is trained；
Step 4 handles selected low-dose CT image V using trained convolutional neural networkss ld, realize selected low dosage
CT image Vs ldThe decomposition of middle anatomical structure ingredient and noise artifacts constituent.
2. the low-dose CT picture breakdown method according to claim 1 or 2 based on convolutional neural networks, feature exist
In: in step 1, low-dose CT image Vs ldIt is to be obtained by the FBP algorithm for reconstructing of parsing, routine dose CT image Vs rdIt is to pass through
The TV algorithm for reconstructing of iteration obtains.
3. the low-dose CT picture breakdown method according to claim 1 based on convolutional neural networks, it is characterised in that: step
In rapid 2, convolutional neural networks include following three kinds of convolution modules: CBR module, branch module and residual error module.
4. the low-dose CT picture breakdown method according to claim 3 based on convolutional neural networks, it is characterised in that:
CBR module is the combination of convolution, change of scale and the operation of ReLU activation primitive；Branch module is two the sum of branches in parallel,
First branches into two convolution operations, and second branches into the combination of convolution, change of scale and the operation of ReLU activation primitive；It is residual
Difference module is two concatenated change of scale, ReLU activation primitive and convolution operation, the sum of with a convolution operation.
5. the low-dose CT picture breakdown method according to claim 1 based on convolutional neural networks, it is characterised in that: step
In rapid 3, by low-dose CT imageIt inputs in convolutional neural networks, output noise artifact forecast imageBy training come more
Parameter in new convolutional neural networks, to reduce the noise artifacts forecast image of convolutional neural networks outputIt is corresponding to reality
Noise artifacts image NsMean square error, when mean square error before and after cycle of training is changed less than 0.1%, training terminates.
6. the low-dose CT picture breakdown method according to claim 5 based on convolutional neural networks, it is characterised in that: step
In rapid 4, by selected low-dose CT image Vt ldIn the convolutional neural networks that input training is completed, obtain selected low based on this
Dosage CT image Vt ldNoise artifacts forecast imageWith dissection constituent image Vt p。
7. the low-dose CT picture breakdown method according to claim 6 based on convolutional neural networks, which is characterized in that step
In rapid 4, anatomical structure ingredient image Vt pIt indicates are as follows:
Wherein, Vt ldFor selected low-dose CT image,For based on the prediction of the noise artifacts of low-dose CT image selected above
Image.
8. the low-dose CT picture breakdown method according to claim 5 or 6 or 7 based on convolutional neural networks, feature
It is, is low-dose CT image to the training sample that convolutional neural networks are trained in step 3Image block collectionMark
Label are practical corresponding noise artifacts image NsImage block collection Ps N。
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 true CN108961237A (en) | 2018-12-07 |
| CN108961237B 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) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109949215A (en) * | 2019-03-29 | 2019-06-28 | 浙江明峰智能医疗科技有限公司 | A kind of low-dose CT image simulation method |
| 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 |
| CN110097611A (en) * | 2019-04-28 | 2019-08-06 | 上海联影智能医疗科技有限公司 | Image rebuilding 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 |
| CN111091516A (en) * | 2019-12-24 | 2020-05-01 | 广州柏视医疗科技有限公司 | Anti-scattering grating method and device based on artificial intelligence |
| CN111192224A (en) * | 2020-01-13 | 2020-05-22 | 北京联合大学 | Image enhancement method and device, electronic equipment and computer readable storage medium |
| CN111325695A (en) * | 2020-02-29 | 2020-06-23 | 深圳先进技术研究院 | Low-dose image enhancement method and system based on multi-dose grade and storage medium |
| CN111951346A (en) * | 2020-08-18 | 2020-11-17 | 安徽工程大学 | 4D-CBCT reconstruction method combining motion estimation and space-time tensor enhancement expression |
| CN112258642A (en) * | 2020-12-21 | 2021-01-22 | 之江实验室 | Low-dose PET data three-dimensional iterative updating reconstruction method based on deep learning |
| CN112348936A (en) * | 2020-11-30 | 2021-02-09 | 华中科技大学 | Low-dose cone-beam CT image reconstruction method based on deep learning |
| CN113053496A (en) * | 2021-03-19 | 2021-06-29 | 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 images |
| CN113256752A (en) * | 2021-06-07 | 2021-08-13 | 太原理工大学 | Low-dose CT reconstruction method based on dual-domain interleaved 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 |
| CN113436118A (en) * | 2021-08-10 | 2021-09-24 | 安徽工程大学 | Low-dose CT image restoration method based on multi-scale convolutional coding network |
| WO2021139835A3 (en) * | 2021-03-24 | 2021-11-11 | 苏州深透智能科技有限公司 | Spect imaging prediction model creation method and apparatus, and device and storage medium |
| US11890124B2 (en) | 2021-02-01 | 2024-02-06 | Medtronic Navigation, Inc. | Systems and methods for low-dose AI-based imaging |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130208971A1 (en) * | 2010-10-27 | 2013-08-15 | Koninklijke Philips Electronics N.V. | Low dose ct denoising |
| 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 |
| US20160328842A1 (en) * | 2015-05-07 | 2016-11-10 | Korea Advanced Institute Of Science And Technology | Development of iterative reconstruction framework using analytic principle for low dose x-ray ct |
| 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 |
-
2018
- 2018-06-28 CN CN201810706749.6A patent/CN108961237B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130208971A1 (en) * | 2010-10-27 | 2013-08-15 | Koninklijke Philips Electronics N.V. | Low dose ct denoising |
| KR101412575B1 (en) * | 2012-07-02 | 2014-06-27 | 충남대학교산학협력단 | Cone beam CT apparatus using low dose x-ray |
| US20160328842A1 (en) * | 2015-05-07 | 2016-11-10 | Korea Advanced Institute Of Science And Technology | Development of iterative reconstruction framework using analytic principle for low dose x-ray ct |
| 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 (6)
| Title |
|---|
| C. ZHANG等: "CT artifact reduction via U-net CNN", 《PROC.OF SPIE》 * |
| HU CHEN等: "Low-dose CT denoising with convolutional neural network", 《2017 IEEE 14TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING》 * |
| JIN LIU 等: "Low-dose CBCT reconstruction via 3D dictionary learning", 《2016 IEEE 13TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING》 * |
| JIN LIU等: "Discriminative Feature Representation to Improve Projection Data Inconsistency for Low Dose CT Imaging", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
| 中国科学技术协会等: "《2012-2013体视学学科发展报告》", 30 April 2014 * |
| 章云港等: "基于卷积神经网络的低剂量ＣＴ图像去噪方法", 《光学学报》 * |
Cited By (24)
| 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 |
| CN109949215A (en) * | 2019-03-29 | 2019-06-28 | 浙江明峰智能医疗科技有限公司 | A kind of low-dose CT image simulation method |
| 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 |
| CN110097611A (en) * | 2019-04-28 | 2019-08-06 | 上海联影智能医疗科技有限公司 | Image rebuilding method, device, equipment and storage medium |
| 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 |
| CN111091516A (en) * | 2019-12-24 | 2020-05-01 | 广州柏视医疗科技有限公司 | Anti-scattering grating method and device based on artificial intelligence |
| CN111192224A (en) * | 2020-01-13 | 2020-05-22 | 北京联合大学 | Image enhancement method and device, electronic equipment and computer readable storage medium |
| 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 |
| CN111325695A (en) * | 2020-02-29 | 2020-06-23 | 深圳先进技术研究院 | Low-dose image enhancement method and system based on multi-dose grade and storage medium |
| CN111951346A (en) * | 2020-08-18 | 2020-11-17 | 安徽工程大学 | 4D-CBCT reconstruction method combining motion estimation and space-time tensor enhancement expression |
| 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 |
| CN112348936A (en) * | 2020-11-30 | 2021-02-09 | 华中科技大学 | Low-dose cone-beam CT image reconstruction method based on deep learning |
| CN112258642A (en) * | 2020-12-21 | 2021-01-22 | 之江实验室 | Low-dose PET data three-dimensional iterative updating 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 |
| CN113053496A (en) * | 2021-03-19 | 2021-06-29 | 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 images |
| WO2021139835A3 (en) * | 2021-03-24 | 2021-11-11 | 苏州深透智能科技有限公司 | Spect imaging prediction model creation method and apparatus, and device and storage medium |
| CN113256752A (en) * | 2021-06-07 | 2021-08-13 | 太原理工大学 | Low-dose CT reconstruction method based on dual-domain interleaved 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 |
| CN113436118A (en) * | 2021-08-10 | 2021-09-24 | 安徽工程大学 | Low-dose CT image restoration method based on multi-scale convolutional coding network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108961237B (en) | 2020-08-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961237A (en) | A kind of low-dose CT picture breakdown method based on convolutional neural networks | |
| CN109166161A (en) | A kind of low-dose CT image processing system inhibiting convolutional neural networks based on noise artifacts | |
| He et al. | Radon inversion via deep learning | |
| Yin et al. | Domain progressive 3D residual convolution network to improve low-dose CT imaging | |
| Wu et al. | A high-quality photon-counting CT technique based on weight adaptive total-variation and image-spectral tensor factorization for small animals imaging | |
| Huang et al. | CaGAN: A cycle-consistent generative adversarial network with attention for low-dose CT imaging | |
| CN108898642A (en) | A kind of sparse angular CT imaging method based on convolutional neural networks | |
| Wu et al. | Non-local low-rank cube-based tensor factorization for spectral CT reconstruction | |
| US11176642B2 (en) | System and method for processing data acquired utilizing multi-energy computed tomography imaging | |
| CN105118066B (en) | A kind of low-dose CT picture breakdown method based on three-dimensional distinctiveness character representation | |
| CN110462689A (en) | Tomography reconstruction based on deep learning | |
| Wu et al. | Deep learning based spectral CT imaging | |
| Wang et al. | FBP-Net for direct reconstruction of dynamic PET images | |
| CN109102550A (en) | Whole network low-dose CT imaging method and device based on convolution residual error network | |
| CN106373163B (en) | A kind of low-dose CT imaging method indicated based on three-dimensional projection's distinctive feature | |
| Zhang et al. | Accurate and robust sparse‐view angle CT image reconstruction using deep learning and prior image constrained compressed sensing (DL‐PICCS) | |
| CN113052936A (en) | Single-view CT reconstruction method integrating FDK and deep learning | |
| Wang et al. | An effective sinogram inpainting for complementary limited-angle dual-energy computed tomography imaging using generative adversarial networks | |
| Feng et al. | Dual residual convolutional neural network (DRCNN) for low-dose CT imaging | |
| CN113344876B (en) | Deformable registration method between CT and CBCT | |
| Rahim et al. | Total variant based average sparsity model with reweighted analysis for compressive sensing of computed tomography | |
| Wang et al. | One half-scan dual-energy CT imaging using the Dual-domain Dual-way Estimated Network (DoDa-Net) model | |
| Clark et al. | Deep learning based spectral extrapolation for dual‐source, dual‐energy x‐ray computed tomography | |
| Liang et al. | A model-based unsupervised deep learning method for low-dose CT reconstruction | |
| Kim et al. | CNN-based CT denoising with an accurate image domain noise insertion technique |
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 |