CN109978778A - Convolutional neural networks medicine CT image denoising method based on residual error study - Google Patents
Convolutional neural networks medicine CT image denoising method based on residual error study Download PDFInfo
- Publication number
- CN109978778A CN109978778A CN201910166558.XA CN201910166558A CN109978778A CN 109978778 A CN109978778 A CN 109978778A CN 201910166558 A CN201910166558 A CN 201910166558A CN 109978778 A CN109978778 A CN 109978778A
- Authority
- CN
- China
- Prior art keywords
- image
- noise
- network
- layer
- training
- 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
- 239000003814 drug Substances 0.000 title claims abstract description 43
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 64
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 238000003062 neural network model Methods 0.000 claims abstract description 7
- 210000002569 neuron Anatomy 0.000 claims description 24
- 230000006870 function Effects 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 18
- 230000004913 activation Effects 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 8
- 239000012141 concentrate Substances 0.000 claims description 7
- 239000000654 additive Substances 0.000 claims description 6
- 230000000996 additive effect Effects 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 241001269238 Data Species 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 238000012986 modification Methods 0.000 claims description 3
- 230000004048 modification Effects 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 6
- 238000012876 topography Methods 0.000 abstract description 2
- 230000001737 promoting effect Effects 0.000 abstract 1
- 230000000717 retained effect Effects 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 238000002591 computed tomography Methods 0.000 description 4
- 238000003745 diagnosis Methods 0.000 description 4
- 210000005036 nerve Anatomy 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 3
- 238000007689 inspection Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000011961 computed axial tomography Methods 0.000 description 2
- 238000013170 computed tomography imaging Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000005251 gamma ray Effects 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000002604 ultrasonography Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
Classifications
-
- 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]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
Convolutional neural networks medicine CT image denoising based on residual error study, the specific steps are as follows: step 1) constructs medicine CT image model；Step 2) constructs neural network model；Step 3) trains network；Step 4) undated parameter；The denoising of step 5) medicine CT image inputs the medicine CT image of Noise, the medicine CT image after network output removal noise into the network model built.The invention has the following advantages that the knowledge in terms of proposing in conjunction with convolutional neural networks in deep learning carries out medicine CT image denoising；Come the noise in approximate study image by the way of residual error study, there is good specific aim, while promoting the training effectiveness of neural network；The method learnt using convolutional neural networks and residual error, can preferably be learnt the characteristic information in image, more topography's information are retained during image denoising.Image denoising ability also gets a promotion simultaneously.
Description
Technical field
The present invention relates to Medical Image Denoising fields, more particularly to medicine CT image, and in particular to one kind is suitable for doctor
Learn the convolutional neural networks medicine CT image denoising method based on residual error study of CT image.
Technical background
As technology develops, in the field of medical imaging, there has also been certain development, such as ultrasonic imaging, CT, MRI etc. to be imaged
Technology is widely used in medical clinic applications.(Computed Tomography, also known as " computer is disconnected for computer tomography
Layer scanning ", abbreviation CT), it is a kind of inspection of medical imageology, this technology is once referred to as computed axial tomography
(Computed Axial Tomography).CT scan utilizes the X-ray beam, gamma ray, ultrasound of Accurate collimation
Wave etc., a certain position for surrounding human body together with the detector high with sensitivity have as tomoscan one by one
The features such as sweep time is fast, image clearly can be used for the inspection of a variety of diseases.Can be divided into according to used ray difference: X is penetrated
Line CT (X-CT), ultrasound computed tomography (UCT), gamma ray CT (Y-CT) etc..
Since CT imaging technique inspection is axial imaging, tissue or device can be shown by image reconstruction, arbitrary orientation
Official shows more fully lesion, prevents from omitting；With high density resolution ratio, the small lesions for having density to change can also be shown
It shows and.There are CT the advantages such as noninvasive, imaging is fast to have become a kind of be widely used and highly safe medical diagnosis technology.
The CT value of each pixel is irregular in the image of non-uniform object in CT image, and image is in graininess, influences density
Resolving power, this phenomenon claim the noise of CT.In terms of there is detector in its source, such as detector sensitivity, pixel size, thickness
And x-ray amount etc..There are also electronic circuit and mechanical aspects, and method for reconstructing and ray at random etc. can also cause noise.In this way in reality
For the removal of CT noise, it is very important in the utilization of border.
The present invention is research object using medicine CT image, and since CT imaging will receive the influence of physical factor, spot is made an uproar
The presence of sound has seriously affected the quality of CT image, causes medicine CT image second-rate.Speckle noise is shown as on the image
Relevant different fleck in spatial domain, it will cover the characteristics of image of those gray scale difference very littles.Clinic is cured
For life, speckle noise causes certain interference to clinical diagnosis.Therefore, it from the angle of clinical application, needs to study
The denoising method of medicine CT image makes more accurately diagnosis for doctor and provides technical support, reduces the risk of Artificial Diagnosis.
In conclusion Research of Medical CT image de-noising method has very important practical significance.
Summary of the invention
For the deficiency for solving the prior art and conventional method, it is an object of the invention to improve the denoising of medicine CT image
Effect.To make medicine CT image be more clear, to allow doctor to make better diagnosis.It is proposed the convolution learnt based on residual error
Neural network medicine CT image denoising method, for solving medicine CT image denoising.
In the prior art, the filtering method of many classics has played great effect in terms of CT image filtering, but
It is that these methods often destroy the original structural information such as image border of image, or cannot obtain satisfactory
Denoising result.Due to being started for research artificial intelligence upsurge, some image de-noising methods based on deep learning are also close several
Year is suggested, and the effect of the image de-noising method based on deep learning depends on the quantity and quality of training set, if can mention
For the sufficiently high training set of quality, the denoising model obtained after training is often available more more aobvious than traditional denoising method
The denoising result of work, but the needs of the denoising method based on deep learning take a substantial amount of time in the calculating of model.It is close
Nian Lai optimizes study to model by residual error learning algorithm, can increase substantially the efficiency of study, obtained model it is pre-
It is also higher to survey efficiency, so that having certain breakthrough to the Image Denoising Technology field based on deep learning.By residual error in the present invention
Learning algorithm is used in medicine CT image denoising, has been invented with higher learning efficiency, denoising effect medicine CT image outstanding
Method for acoustic is denoised, finally by the feasibility of simulating, verifying method and the effect of optimization.
To be more clear the object, technical solutions and advantages of the present invention, below just to technical solution of the present invention make into
The description of one step, the convolutional neural networks medicine CT image denoising method based on residual error study, the specific steps are as follows:
Step 1) constructs medicine CT image model；
The model of CT image mainly consists of two parts, both effective tissue reflection signal and invalid noise letter
Number, and noise signal then includes multiplicative noise and additive noise, wherein additive noise is for multiplicative noise to CT image
Influence it is very small.Due to considering that multiplicative noise, the universal model s (x, y) of CT electric signal indicate are as follows:
S (x, y)=r (x, y) n (x, y) (1)
Wherein, (x, y) respectively represents the transverse and longitudinal coordinate of image, and r (x, y) indicates that noise-free signal, n (x, y) indicate to be multiplied
Noise.
Step 2) constructs neural network model；
1. constructing neural network；
Great deal of nodes is coupled to each other composition network layer, node, that is, neuron, and neuron is divided into different levels, each nerve
Member is connected with other neurons of adjacent layer.Each layer of neuron have input (its input be preceding layer neuron output) and
Output.
Each cynapse has a weight in neural network, and the output valve of each neuron is previous adjacent networks nerve
The weighted input of member is simultaneously exported by activation primitive, is ReLU function, ReLU used in the convolutional neural networks in the present invention
The formula of activation primitive is as follows:
F (x) is represented as the ReLU function of input, and x is input value.
2. constructing convolutional neural networks；
Construct three network layers, input layer, hidden layer, output layer.
Input layer is the input of image, and convolution, batch normalization, activation primitive are used in hidden layer.I.e. are as follows: Conv+BN+
ReLU.The image inputted in training is gray level image；The size of convolution filter is set as 3 × 3 in network model.
The depth d of network is set as 20.
First layer is using convolution sum batch normalization (Conv+BN) in hidden layer, using 64 Convolution Filters having a size of 3 ×
3 × c generates 64 characteristic patterns.C is expressed as the quantity of image channel, because model thus is using gray level image training, institute
It is 1 with c.At the 2nd layer of hidden layer into (d-1) layer, the convolution filter for the use of 64 sizes being 3 × 3.And
Addition batch normalization between each layer of convolution sum ReLU activation primitive, some batch data is during network training
{ x1, x2 ... xn }, batch data are batch block number evidences.It is defeated that this data can be a certain layer that input is also possible among network
Out.Every batch of input data is subjected to calculation process, its distribution is made to be equal to the data distribution of whole training datas.To batch data
It is as follows that formula is normalized:
The mean μ of upper one layer of output dataβ(m is training sample batch size, xiFor the data of input):
The standard deviation of upper one layer of output data
Normalized obtains output data(wherein ε be arbitrarily close to 0 value):
To reconstructing after normalized data, y is obtainedi:
γ and β is the parameter value learnt in network, is arrived in the acquistion of network training middle school, and constantly update.
The last layer output layer realizes that image reconstruction exports using the filter that 1 size is 3 × 3 × 64.
Study in neural network model for noise is using residual error R (v) come approximate.
Step 3) trains network；
It is Gaussian noise that training data, which concentrates the noise of image addition,.Initial initial data training set is 400 big
The small gray level image for being 180 × 180, then cuts these pictures, and the image size after cutting is 50 × 50.It cuts
Image block, which is 128 × 3000, to be used to train network, and noise level σ is arranged in σ ∈ [0,55] during network training.
The network model obtained after training has the performance of Image Blind denoising.The depth of network is set as in training process
20, we use stochastic gradient descent optimization algorithm, and learning rate is set as 0.0001.
Step 4) undated parameter；
1) residual error study is combined with batch normalization in network；
Past some neural network image denoising models are that directly noise-containing image y is learnt and predicted.Pass through
Original mappings obtain the clean image removed after noise in input picture, i.e. forecast image y removes the clean image after noise.This
The design of network directly predicts noise in image v by residual error mapping R (y).This residual error mapping exists than original mappings
It is more easier to optimize in network training and learning process, be able to ascend the training effectiveness of network model in this way while promotes network mould
Type image denoising ability.
2) parameter updates and optimizes in network model；
Network model trains iteration that will be updated to network parameter every time, finally obtains with preferable medicine CT figure
As the network model of denoising performance.The update of parameter uses back-propagation algorithm (abbreviation BP algorithm) in network model.In algorithm
Main two links are that excitation is propagated and weight updates, by loop iteration repeatedly, until the sound inputted in network model
Until should reaching in scheduled target zone.It is broadly divided into forward-propagating and back-propagation process.
During forward-propagating, the image information of input enters hidden layer by input layer, successively processing and to output layer
Output.By the way that the error of output valve and desired value is carried out quadratic sum as loss function.
In back-propagation process, loss function is successively found out to the partial derivative of each neuron weight, constitutes loss function pair
The ladder amount of weight vector, as the foundation of modification weight, network model is completed parameter by multiple training iteration and is modified.Finally
Error is set to reach the desired value reached.
In training process, loss function L (Θ) formula are as follows:
What wherein Θ was represented is training parameter, and N is that training data concentrates picture total quantity, yiIt is expressed as training data concentration
One noise-containing picture, xiWhat is represented is free from noisy clean picture, and actual noise is (y in picturei-xi).So
The desired value of loss function calculates the noise residual error R (y by estimationi；Θ) with picture in actual noise square error.
For the present invention during network training, training dataset is the data set with label.
Image denoising is the level of first estimating noise of input image in general image denoising model, using corresponding noise
Level training obtains corresponding model to denoise to image.Such denoising model will receive the shadow of noise estimated accuracy
It rings, also influences whether the image denoising result of model.Medicine CT image for the angle of medicine CT image denoising, after denoising
Retain important minutia, the process of denoising will reduce the loss of minutia.
The denoising of step 5) medicine CT image；
The medicine CT image of Noise, the medicine CT after network output removal noise are inputted into the network model built
Image, network model have good blind noise removal capability.Illustrate that this method has very what is denoised to medicine CT image using upper
Good effect.
The invention has the following advantages that
1. the knowledge in terms of proposing in conjunction with convolutional neural networks in deep learning carries out medicine CT image denoising.
2. there is good specific aim, promoted simultaneously come the noise in approximate study image by the way of residual error study
The training effectiveness of neural network.
3. the method learnt using convolutional neural networks and residual error, can preferably learn the characteristic information in image,
Retain more topography's information during image denoising.Image denoising ability also gets a promotion simultaneously.
Detailed description of the invention
Fig. 1 is neural network structure schematic diagram；
Fig. 2 is the ReLU activation primitive schematic diagram used in present networks；
Fig. 3 is that schematic diagram of a layer structure is hidden in network；
Fig. 4 is that noise residual error learns schematic diagram in network；
Fig. 5 is that network model denoises schematic diagram to medicine CT image in the present invention；
Fig. 6 is flow chart of the method for the present invention.
Specific embodiment:
It makes a concrete explanation explanation below in conjunction with attached drawing to the present invention
Specific step is as follows for convolutional neural networks medicine CT image denoising method based on residual error study of the invention:
Step 1) constructs medicine CT image model；
The model of CT image mainly consists of two parts, both effective tissue reflection signal and invalid noise letter
Number, and noise signal then includes multiplicative noise and additive noise, wherein additive noise is for multiplicative noise to CT image
Influence it is very small.Due to considering that multiplicative noise, the universal model s (x, y) of CT electric signal indicate are as follows:
S (x, y)=r (x, y) n (x, y) (1)
Wherein, (x, y) respectively represents the transverse and longitudinal coordinate of image, and r (x, y) indicates that noise-free signal, n (x, y) indicate to be multiplied
Noise.
Step 2) constructs neural network model；
1. constructing neural network；
Construct three network layers, input layer, hidden layer, output layer, as shown in Figure 1.
Great deal of nodes is coupled to each other composition network layer, node, that is, neuron, and neuron is divided into different levels, each nerve
Member is connected with other neurons of adjacent layer.Each layer of neuron have input (its input be preceding layer neuron output) and
Output.
Each cynapse has a weight in neural network, and the output valve of each neuron is previous adjacent networks nerve
The weighted input of member is simultaneously exported by nonlinear function (activation primitive), is used in the convolutional neural networks in the present invention
ReLU function, as shown in Figure 2.The formula of ReLU activation primitive is as follows:
F (x) is represented as the ReLU function of input, and x is input value.
2. constructing convolutional neural networks；
Construct three network layers, input layer, hidden layer, output layer.
Input layer is the input of image, and convolution, batch normalization, activation primitive are used in hidden layer.I.e. are as follows: Conv+BN+
ReLU, as shown in Figure 3.The image inputted in training is gray level image；The size of convolution filter is set in network model
It is set to 3 × 3.
The depth d of network is set as 20 in present networks.It is all made of pond layer in the model of general convolutional Neural network, in Home Network
Without using pond layer in network, such model is more accurate when obtaining the feature of image, reduces some minutias in image
It loses, this is extremely important in medicine CT image.
First layer is using convolution sum batch normalization (Conv+BN) in hidden layer, using 64 Convolution Filters having a size of 3 ×
3 × c generates 64 characteristic patterns.C is expressed as the quantity of image channel, because model thus is using gray level image training, institute
It is 1 with c.At the 2nd layer of hidden layer into (d-1) layer, the convolution filter for the use of 64 sizes being 3 × 3.And
Addition batch normalization between each layer of convolution sum ReLU activation primitive, some batch data is during network training
{ x1, x2 ... xn }, batch data are batch block number evidences.It is defeated that this data can be a certain layer that input is also possible among network
Out.Every batch of input data is subjected to calculation process, its distribution is made to be equal to the data distribution of whole training datas.To batch data
It is as follows that formula is normalized:
The mean μ of upper one layer of output dataβ(m is training sample batch size, xiFor the data of input):
The standard deviation of upper one layer of output data
Normalized obtains output data(wherein ε be arbitrarily close to 0 value):
To reconstructing after normalized data, y is obtainedi:
γ and β is the parameter value learnt in network, is arrived in the acquistion of network training middle school, and constantly update.
The last layer output layer realizes that image reconstruction exports using the filter that 1 size is 3 × 3 × 64.
Study in neural network model for noise is using residual error R (v) come approximate.Pass through setting residual error network in network
It is able to ascend the training effectiveness of network with batch normalization, while being conducive to be promoted the image denoising ability of network model.
Step 3) trains network；
It is Gaussian noise that training data, which concentrates the noise of image addition,.Initial initial data training set is 400 big
The small gray level image for being 180 × 180, then cuts these pictures, and the image size after cutting is 50 × 50.It cuts
Image block, which is 128 × 3000, to be used to train network, and noise level σ is arranged in σ ∈ [0,55] during network training.
The network model obtained after training has the performance of Image Blind denoising.The depth of network is set as in training process
20, we use stochastic gradient descent optimization algorithm, and learning rate is set as 0.0001.
Step 4) undated parameter；
1) residual error study is combined with batch normalization in network；
Past some neural network image denoising models are that directly noise-containing image y is learnt and predicted.Pass through
Original mappings obtain the clean image removed after noise in input picture, i.e. forecast image y removes the clean image after noise.This
The design of network directly predicts noise in image v by residual error mapping R (y).As shown in Figure 4.This residual error maps ratio
Original mappings are more easier to optimize in network training and learning process, are able to ascend the training effectiveness of network model so simultaneously
Promote network model image denoising ability.
2) parameter updates and optimizes in network model；
Network model trains iteration that will be updated to network parameter every time, finally obtains with preferable medicine CT figure
As the network model of denoising performance.The update of parameter uses back-propagation algorithm (abbreviation BP algorithm) in network model.In algorithm
Main two links are that excitation is propagated and weight updates, by loop iteration repeatedly, until the sound inputted in network model
Until should reaching in scheduled target zone.It is broadly divided into forward-propagating and back-propagation process.
During forward-propagating, the image information of input enters hidden layer by input layer, successively processing and to output layer
Output.By the way that the error of output valve and desired value is carried out quadratic sum as loss function.
In back-propagation process, loss function is successively found out to the partial derivative of each neuron weight, constitutes loss function pair
The ladder amount of weight vector, as the foundation of modification weight, network model is completed parameter by multiple training iteration and is modified.Finally
Error is set to reach the desired value reached.
In training process, loss function L (Θ) formula are as follows:
What wherein Θ was represented is training parameter, and N is that training data concentrates picture total quantity, yiIt is expressed as training data concentration
One noise-containing picture, xiWhat is represented is free from noisy clean picture, and actual noise is (y in picturei-xi).So
The desired value of loss function calculates the noise residual error R (y by estimationi；Θ) with picture in actual noise square error.
For the present invention during network training, training dataset is the data set with label.
The denoising of step 5) medicine CT image；
The medicine CT image of Noise, the medicine CT after network output removal noise are inputted into the network model built
Image, as shown in Figure 5.Network model has good blind noise removal capability.Illustrate that this method is answered what is denoised to medicine CT image
With with good effect.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention
Range should not be construed as being limited to the specific forms stated in the embodiments, and protection scope of the present invention is also and in art technology
Personnel conceive according to the present invention it is conceivable that equivalent technologies mean.
Claims (1)
1. the convolutional neural networks medicine CT image denoising based on residual error study, the specific steps are as follows:
Step 1) constructs medicine CT image model；
The model of CT image mainly consists of two parts, both effective tissue reflection signal and invalid noise signal, and
Noise signal then includes multiplicative noise and additive noise, wherein influence of the additive noise for multiplicative noise to CT image
It is very small；Due to considering that multiplicative noise, the universal model s (x, y) of CT electric signal indicate are as follows:
S (x, y)=r (x, y) n (x, y) (1)
Wherein, (x, y) respectively represents the transverse and longitudinal coordinate of image, and r (x, y) indicates that noise-free signal, n (x, y) indicate multiplicative noise；
Step 2) constructs neural network model；
1. constructing neural network:
Great deal of nodes is coupled to each other composition network layer, node, that is, neuron, and neuron is divided into different levels, each neuron with
Other neurons of adjacent layer are connected；Each layer of neuron is output and input, and the input of each layer of neuron is preceding layer
Neuron output；
Each cynapse has a weight in neural network, and the output valve of each neuron is previous adjacent networks neuron
Weighted input is simultaneously exported by activation primitive, is ReLU function, the formula of ReLU activation primitive used in convolutional neural networks
It is as follows:
F (x) is represented as the ReLU function of input, and x is input value；
2. constructing convolutional neural networks:
Construct three network layers, input layer, hidden layer, output layer；
Input layer is the input of image；With convolution, batch normalization, activation primitive in hidden layer；I.e. are as follows: Conv+BN+ReLU；
The image inputted in training is gray level image；The size of convolution filter is set as 3 × 3 in network model；
The depth d of network is set as 20；
First layer is using convolution sum batch normalization (Conv+BN) in hidden layer, using 64 Convolution Filters having a size of 3 × 3 × c
To generate 64 characteristic patterns；C is expressed as the quantity of image channel, because model is using gray level image training thus, so c is
It is 1；At the 2nd layer of hidden layer into (d-1) layer, the convolution filter for the use of 64 sizes being 3 × 3；And each
Layer convolution sum ReLU activation primitive between addition batch normalization, during network training some batch data be x1,
X2 ... xn }, batch data are batch block number evidences；This data can be a certain layer output that input is also possible among network；
Every batch of input data is subjected to calculation process, its distribution is made to be equal to the data distribution of whole training datas；Batch data are carried out
Normalized processing formula is as follows:
The mean μ of upper one layer of output dataβ, m is training sample batch size, xiFor the data of input:
The standard deviation of upper one layer of output data
Normalized obtains output dataWherein ε be arbitrarily close to 0 value:
To reconstructing after normalized data, y is obtainedi:
γ and β is the parameter value learnt in network, is arrived in the acquistion of network training middle school, and constantly update；
The last layer output layer realizes that image reconstruction exports using the filter that 1 size is 3 × 3 × 64；
Study in neural network model for noise is using residual error R (v) come approximate；
Step 3) trains network；
It is Gaussian noise that training data, which concentrates the noise of image addition,；Initial initial data training set is that 400 Zhang great little are
Then 180 × 180 gray level image cuts these pictures；Image size after cutting is 50 × 50；The image of cutting
Block, which is 128 × 3000, to be used to train network；Noise level σ setting exists during network training
The network model obtained after training has the performance of Image Blind denoising；The depth of network is set as 20 in training process, adopts
With stochastic gradient descent optimization algorithm, learning rate is set as 0.0001；
Step 4) undated parameter；
41) residual error study is combined with batch normalization in network；
Noise in image v is directly predicted by residual error mapping R (v)；
42) parameter updates and optimizes in network model；
Network model trains iteration to be divided into forward-propagating and back-propagation process every time；
During forward-propagating, the image information of input enters hidden layer by input layer, and successively processing is simultaneously exported to output layer；
By the way that the error of output valve and desired value is carried out quadratic sum as loss function；
In back-propagation process, loss function is successively found out to the partial derivative of each neuron weight, constitutes loss function to weight
The ladder amount of vector, as the foundation of modification weight, network model is completed parameter by multiple training iteration and is modified；Finally make to miss
Difference reaches the desired value reached；
In training process, loss function L (Θ) formula are as follows:
What wherein Θ was represented is training parameter, and N is that training data concentrates picture total quantity, and yi is expressed as training data and concentrates one
Noise-containing picture, xiWhat is represented is free from noisy clean picture, and actual noise is (y in picturei-xi)；So loss
The desired value of function calculates the noise residual error R (y by estimationi；Θ) with picture in actual noise square error；
Training dataset is the data set with label；
The denoising of step 5) medicine CT image；
The medicine CT image of Noise, the medicine CT figure after network output removal noise are inputted into the network model built
Picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910166558.XA CN109978778B (en) | 2019-03-06 | 2019-03-06 | Convolutional neural network medical CT image denoising method based on residual learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910166558.XA CN109978778B (en) | 2019-03-06 | 2019-03-06 | Convolutional neural network medical CT image denoising method based on residual learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109978778A true CN109978778A (en) | 2019-07-05 |
| CN109978778B CN109978778B (en) | 2020-11-13 |
Family
ID=67078031
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910166558.XA Active CN109978778B (en) | 2019-03-06 | 2019-03-06 | Convolutional neural network medical CT image denoising method based on residual learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109978778B (en) |
Cited By (48)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110415190A (en) * | 2019-07-25 | 2019-11-05 | 青岛科技大学 | Method, apparatus and processor based on deep learning removal compression of images noise |
| CN110490824A (en) * | 2019-08-23 | 2019-11-22 | 广东浪潮大数据研究有限公司 | The method and device of image denoising |
| CN110533610A (en) * | 2019-08-20 | 2019-12-03 | 东软医疗系统股份有限公司 | The generation method and device of image enhancement model, application method and device |
| CN110599409A (en) * | 2019-08-01 | 2019-12-20 | 西安理工大学 | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel |
| CN110782406A (en) * | 2019-10-15 | 2020-02-11 | 深圳大学 | Image denoising method and device based on information distillation network |
| CN110827250A (en) * | 2019-10-29 | 2020-02-21 | 浙江明峰智能医疗科技有限公司 | Intelligent medical image quality evaluation method based on lightweight convolutional neural network |
| CN110859642A (en) * | 2019-11-26 | 2020-03-06 | 北京华医共享医疗科技有限公司 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
| CN111028171A (en) * | 2019-12-06 | 2020-04-17 | 北京金山云网络技术有限公司 | Method, device and server for determining noise level of image |
| CN111047524A (en) * | 2019-11-13 | 2020-04-21 | 浙江工业大学 | Low-dose CT lung image denoising method based on deep convolutional neural network |
| CN111145125A (en) * | 2019-12-30 | 2020-05-12 | 南京信息工程大学 | Image denoising method based on residual learning and convolutional neural network |
| CN111242862A (en) * | 2020-01-09 | 2020-06-05 | 西安理工大学 | Multi-scale fusion parallel dense residual convolution neural network image denoising method |
| CN111242870A (en) * | 2020-01-20 | 2020-06-05 | 南开大学 | Low-light image enhancement method based on deep learning knowledge distillation technology |
| CN111260591A (en) * | 2020-03-12 | 2020-06-09 | 武汉大学 | Image self-adaptive denoising method based on attention mechanism |
| CN111311506A (en) * | 2020-01-21 | 2020-06-19 | 辽宁师范大学 | Low-dose CT tooth image denoising method based on double residual error networks |
| CN111580162A (en) * | 2020-05-21 | 2020-08-25 | 长江大学 | Seismic data random noise suppression method based on residual convolutional neural network |
| CN111612739A (en) * | 2020-04-16 | 2020-09-01 | 杭州电子科技大学 | Deep learning-based cerebral infarction classification method |
| CN111860361A (en) * | 2020-07-24 | 2020-10-30 | 吉林大学 | Green channel cargo scanning image entrainment automatic identifier and identification method |
| CN111862060A (en) * | 2020-07-27 | 2020-10-30 | 南京图格医疗科技有限公司 | CBD-Net-based medical endoscopic image denoising method |
| CN112001977A (en) * | 2020-07-08 | 2020-11-27 | 浙江工业大学 | Electrical impedance tomography image reconstruction method based on residual error network |
| CN112084982A (en) * | 2020-09-15 | 2020-12-15 | 浙江树人学院（浙江树人大学） | Double-layer tubular column pulse eddy current data denoising method based on noise model |
| CN112132768A (en) * | 2020-09-30 | 2020-12-25 | 西京学院 | MRI image noise reduction method based on multi-noise reducer joint optimization noise reduction |
| CN112184563A (en) * | 2020-08-03 | 2021-01-05 | 天津大学 | Depth learning-based human retina optical coherence tomography image batch filtering method |
| CN112308788A (en) * | 2019-07-25 | 2021-02-02 | 株式会社日立制作所 | Image processing device, image processing method, and X-ray CT device |
| CN112330575A (en) * | 2020-12-03 | 2021-02-05 | 华北理工大学 | Convolution neural network medical CT image denoising method |
| CN112381178A (en) * | 2020-12-07 | 2021-02-19 | 西安交通大学 | Medical image classification method based on multi-loss feature learning |
| CN112418408A (en) * | 2020-12-08 | 2021-02-26 | 中山大学 | Neural network interpretable method based on inhibition of irrelevant features |
| CN112419169A (en) * | 2020-10-14 | 2021-02-26 | 浙江工业大学 | CNN medical CT image denoising method based on noise prior |
| CN112651885A (en) * | 2019-10-09 | 2021-04-13 | 西门子医疗有限公司 | Method and apparatus for reducing image recording noise |
| CN112700534A (en) * | 2020-12-30 | 2021-04-23 | 华东师范大学 | Ultrasonic or CT medical image three-dimensional reconstruction method based on feature migration |
| CN112733439A (en) * | 2020-12-29 | 2021-04-30 | 哈尔滨工程大学 | Method for calculating shielding material accumulation factor based on BP neural network |
| CN112767377A (en) * | 2021-01-27 | 2021-05-07 | 电子科技大学 | Cascade medical image enhancement method |
| CN112927151A (en) * | 2021-02-25 | 2021-06-08 | 西安理工大学 | Image denoising method based on edge enhancement and convolutional neural network |
| CN112991477A (en) * | 2021-01-28 | 2021-06-18 | 明峰医疗系统股份有限公司 | PET image processing method based on deep learning |
| CN113256500A (en) * | 2021-07-02 | 2021-08-13 | 北京大学第三医院（北京大学第三临床医学院） | Deep learning neural network model system for multi-modal image synthesis |
| CN113393548A (en) * | 2021-05-25 | 2021-09-14 | 中国科学院西安光学精密机械研究所 | Non-vision field imaging method based on depth residual error neural network |
| CN113436290A (en) * | 2020-03-23 | 2021-09-24 | 通用电气精准医疗有限责任公司 | Method and system for selectively removing streak artifacts and noise from images using a deep neural network |
| CN113537025A (en) * | 2021-07-08 | 2021-10-22 | 浙江工业大学 | Electromagnetic modulation signal deep denoising method and system based on self-supervision learning |
| CN113723213A (en) * | 2021-08-06 | 2021-11-30 | 武汉理工大学 | Method for realizing subway vibration signal noise reduction training based on Hilbert curve coding |
| WO2022027216A1 (en) * | 2020-08-04 | 2022-02-10 | 深圳高性能医疗器械国家研究院有限公司 | Image denoising method and application thereof |
| CN114219820A (en) * | 2021-12-08 | 2022-03-22 | 苏州工业园区智在天下科技有限公司 | Neural network generation method, denoising method and device |
| CN114548156A (en) * | 2022-01-24 | 2022-05-27 | 成都理工大学 | Distributed optical fiber temperature measurement and noise reduction method based on down-sampling and convolutional neural network |
| CN114545504A (en) * | 2022-02-18 | 2022-05-27 | 吉林大学 | Desert seismic exploration random noise modeling method based on deep learning |
| CN114596223A (en) * | 2022-02-24 | 2022-06-07 | 复旦大学 | Quantization parameter self-adaptive convolution neural network loop filter and construction method thereof |
| CN114757847A (en) * | 2022-04-24 | 2022-07-15 | 汕头市超声仪器研究所股份有限公司 | Multi-information extraction extended U-Net and application method thereof in low-dose X-ray imaging |
| CN115115022A (en) * | 2021-03-19 | 2022-09-27 | 美光科技公司 | Modular machine learning model for denoising images and systems and methods using the same |
| CN116908760A (en) * | 2023-07-11 | 2023-10-20 | 哈尔滨工业大学 | MRI bias field correction method and system based on depth decoupling network |
| CN117095074A (en) * | 2023-08-23 | 2023-11-21 | 浙江大学 | Medical image double denoising method and device based on deep learning |
| CN117392545A (en) * | 2023-10-26 | 2024-01-12 | 南昌航空大学 | SAR image target detection method based on deep learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107403419A (en) * | 2017-08-04 | 2017-11-28 | 深圳市唯特视科技有限公司 | A kind of low dose X-ray image de-noising method based on concatenated convolutional neutral net |
| CN107633486A (en) * | 2017-08-14 | 2018-01-26 | 成都大学 | Structure Magnetic Resonance Image Denoising based on three-dimensional full convolutional neural networks |
| CN108564553A (en) * | 2018-05-07 | 2018-09-21 | 南方医科大学 | Low-dose CT image noise suppression method based on convolutional neural networks |
| CN109118435A (en) * | 2018-06-15 | 2019-01-01 | 广东工业大学 | A kind of depth residual error convolutional neural networks image de-noising method based on PReLU |
| CN109242798A (en) * | 2018-09-14 | 2019-01-18 | 南昌大学 | A kind of Poisson denoising method based on three cross-talk network representations |
| CN109410127A (en) * | 2018-09-17 | 2019-03-01 | 西安电子科技大学 | A kind of image de-noising method based on deep learning and multi-scale image enhancing |
-
2019
- 2019-03-06 CN CN201910166558.XA patent/CN109978778B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107403419A (en) * | 2017-08-04 | 2017-11-28 | 深圳市唯特视科技有限公司 | A kind of low dose X-ray image de-noising method based on concatenated convolutional neutral net |
| CN107633486A (en) * | 2017-08-14 | 2018-01-26 | 成都大学 | Structure Magnetic Resonance Image Denoising based on three-dimensional full convolutional neural networks |
| CN108564553A (en) * | 2018-05-07 | 2018-09-21 | 南方医科大学 | Low-dose CT image noise suppression method based on convolutional neural networks |
| CN109118435A (en) * | 2018-06-15 | 2019-01-01 | 广东工业大学 | A kind of depth residual error convolutional neural networks image de-noising method based on PReLU |
| CN109242798A (en) * | 2018-09-14 | 2019-01-18 | 南昌大学 | A kind of Poisson denoising method based on three cross-talk network representations |
| CN109410127A (en) * | 2018-09-17 | 2019-03-01 | 西安电子科技大学 | A kind of image de-noising method based on deep learning and multi-scale image enhancing |
Cited By (75)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110415190A (en) * | 2019-07-25 | 2019-11-05 | 青岛科技大学 | Method, apparatus and processor based on deep learning removal compression of images noise |
| CN112308788A (en) * | 2019-07-25 | 2021-02-02 | 株式会社日立制作所 | Image processing device, image processing method, and X-ray CT device |
| CN112308788B (en) * | 2019-07-25 | 2023-10-20 | 富士胶片医疗健康株式会社 | Image processing device, image processing method, and X-ray CT device |
| CN110415190B (en) * | 2019-07-25 | 2023-03-14 | 青岛科技大学 | Method, device and processor for removing image compression noise based on deep learning |
| CN110599409A (en) * | 2019-08-01 | 2019-12-20 | 西安理工大学 | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel |
| CN110599409B (en) * | 2019-08-01 | 2022-02-22 | 西安理工大学 | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel |
| CN110533610A (en) * | 2019-08-20 | 2019-12-03 | 东软医疗系统股份有限公司 | The generation method and device of image enhancement model, application method and device |
| CN110490824A (en) * | 2019-08-23 | 2019-11-22 | 广东浪潮大数据研究有限公司 | The method and device of image denoising |
| US11983847B2 (en) * | 2019-10-09 | 2024-05-14 | Siemens Healthineers Ag | Method and device for noise reduction in image recordings |
| CN112651885A (en) * | 2019-10-09 | 2021-04-13 | 西门子医疗有限公司 | Method and apparatus for reducing image recording noise |
| US20210110517A1 (en) * | 2019-10-09 | 2021-04-15 | Siemens Healthcare Gmbh | Method and device for noise reduction in image recordings |
| CN110782406A (en) * | 2019-10-15 | 2020-02-11 | 深圳大学 | Image denoising method and device based on information distillation network |
| CN110782406B (en) * | 2019-10-15 | 2022-10-11 | 深圳大学 | Image denoising method and device based on information distillation network |
| CN110827250A (en) * | 2019-10-29 | 2020-02-21 | 浙江明峰智能医疗科技有限公司 | Intelligent medical image quality evaluation method based on lightweight convolutional neural network |
| CN111047524A (en) * | 2019-11-13 | 2020-04-21 | 浙江工业大学 | Low-dose CT lung image denoising method based on deep convolutional neural network |
| CN110859642B (en) * | 2019-11-26 | 2024-01-23 | 北京华医共享医疗科技有限公司 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
| CN110859642A (en) * | 2019-11-26 | 2020-03-06 | 北京华医共享医疗科技有限公司 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
| CN111028171B (en) * | 2019-12-06 | 2023-04-18 | 北京金山云网络技术有限公司 | Method, device and server for determining noise level of image |
| CN111028171A (en) * | 2019-12-06 | 2020-04-17 | 北京金山云网络技术有限公司 | Method, device and server for determining noise level of image |
| CN111145125A (en) * | 2019-12-30 | 2020-05-12 | 南京信息工程大学 | Image denoising method based on residual learning and convolutional neural network |
| CN111242862B (en) * | 2020-01-09 | 2022-02-22 | 西安理工大学 | Multi-scale fusion parallel dense residual convolution neural network image denoising method |
| CN111242862A (en) * | 2020-01-09 | 2020-06-05 | 西安理工大学 | Multi-scale fusion parallel dense residual convolution neural network image denoising method |
| CN111242870A (en) * | 2020-01-20 | 2020-06-05 | 南开大学 | Low-light image enhancement method based on deep learning knowledge distillation technology |
| CN111242870B (en) * | 2020-01-20 | 2023-05-05 | 南开大学 | Low-light image enhancement method based on deep learning knowledge distillation technology |
| CN111311506A (en) * | 2020-01-21 | 2020-06-19 | 辽宁师范大学 | Low-dose CT tooth image denoising method based on double residual error networks |
| CN111311506B (en) * | 2020-01-21 | 2023-05-09 | 辽宁师范大学 | Low-dose CT (computed tomography) tooth image denoising method based on double residual error network |
| CN111260591B (en) * | 2020-03-12 | 2022-04-26 | 武汉大学 | Image self-adaptive denoising method based on attention mechanism |
| CN111260591A (en) * | 2020-03-12 | 2020-06-09 | 武汉大学 | Image self-adaptive denoising method based on attention mechanism |
| CN113436290A (en) * | 2020-03-23 | 2021-09-24 | 通用电气精准医疗有限责任公司 | Method and system for selectively removing streak artifacts and noise from images using a deep neural network |
| CN111612739B (en) * | 2020-04-16 | 2023-11-03 | 杭州电子科技大学 | Deep learning-based cerebral infarction classification method |
| CN111612739A (en) * | 2020-04-16 | 2020-09-01 | 杭州电子科技大学 | Deep learning-based cerebral infarction classification method |
| CN111580162A (en) * | 2020-05-21 | 2020-08-25 | 长江大学 | Seismic data random noise suppression method based on residual convolutional neural network |
| CN112001977A (en) * | 2020-07-08 | 2020-11-27 | 浙江工业大学 | Electrical impedance tomography image reconstruction method based on residual error network |
| CN111860361A (en) * | 2020-07-24 | 2020-10-30 | 吉林大学 | Green channel cargo scanning image entrainment automatic identifier and identification method |
| CN111862060A (en) * | 2020-07-27 | 2020-10-30 | 南京图格医疗科技有限公司 | CBD-Net-based medical endoscopic image denoising method |
| CN112184563A (en) * | 2020-08-03 | 2021-01-05 | 天津大学 | Depth learning-based human retina optical coherence tomography image batch filtering method |
| WO2022027216A1 (en) * | 2020-08-04 | 2022-02-10 | 深圳高性能医疗器械国家研究院有限公司 | Image denoising method and application thereof |
| CN112084982A (en) * | 2020-09-15 | 2020-12-15 | 浙江树人学院（浙江树人大学） | Double-layer tubular column pulse eddy current data denoising method based on noise model |
| CN112084982B (en) * | 2020-09-15 | 2024-07-02 | 浙江树人学院（浙江树人大学） | Double-layer tubular column pulse vortex data denoising method based on noise model |
| CN112132768A (en) * | 2020-09-30 | 2020-12-25 | 西京学院 | MRI image noise reduction method based on multi-noise reducer joint optimization noise reduction |
| CN112419169B (en) * | 2020-10-14 | 2024-04-02 | 浙江工业大学 | CNN medical CT image denoising method based on noise priori |
| CN112419169A (en) * | 2020-10-14 | 2021-02-26 | 浙江工业大学 | CNN medical CT image denoising method based on noise prior |
| CN112330575A (en) * | 2020-12-03 | 2021-02-05 | 华北理工大学 | Convolution neural network medical CT image denoising method |
| CN112330575B (en) * | 2020-12-03 | 2022-10-14 | 华北理工大学 | Convolution neural network medical CT image denoising method |
| CN112381178B (en) * | 2020-12-07 | 2023-05-02 | 西安交通大学 | Medical image classification method based on multi-loss feature learning |
| CN112381178A (en) * | 2020-12-07 | 2021-02-19 | 西安交通大学 | Medical image classification method based on multi-loss feature learning |
| CN112418408A (en) * | 2020-12-08 | 2021-02-26 | 中山大学 | Neural network interpretable method based on inhibition of irrelevant features |
| CN112418408B (en) * | 2020-12-08 | 2023-09-05 | 中山大学 | Neural network interpretability method based on uncorrelated feature suppression |
| CN112733439A (en) * | 2020-12-29 | 2021-04-30 | 哈尔滨工程大学 | Method for calculating shielding material accumulation factor based on BP neural network |
| CN112700534A (en) * | 2020-12-30 | 2021-04-23 | 华东师范大学 | Ultrasonic or CT medical image three-dimensional reconstruction method based on feature migration |
| CN112767377A (en) * | 2021-01-27 | 2021-05-07 | 电子科技大学 | Cascade medical image enhancement method |
| CN112991477B (en) * | 2021-01-28 | 2023-04-18 | 明峰医疗系统股份有限公司 | PET image processing method based on deep learning |
| CN112991477A (en) * | 2021-01-28 | 2021-06-18 | 明峰医疗系统股份有限公司 | PET image processing method based on deep learning |
| CN112927151A (en) * | 2021-02-25 | 2021-06-08 | 西安理工大学 | Image denoising method based on edge enhancement and convolutional neural network |
| CN112927151B (en) * | 2021-02-25 | 2023-09-05 | 西安理工大学 | Image denoising method based on edge enhancement and convolutional neural network |
| CN115115022A (en) * | 2021-03-19 | 2022-09-27 | 美光科技公司 | Modular machine learning model for denoising images and systems and methods using the same |
| CN113393548B (en) * | 2021-05-25 | 2022-12-09 | 中国科学院西安光学精密机械研究所 | Non-vision field imaging method based on depth residual error neural network |
| CN113393548A (en) * | 2021-05-25 | 2021-09-14 | 中国科学院西安光学精密机械研究所 | Non-vision field imaging method based on depth residual error neural network |
| CN113256500A (en) * | 2021-07-02 | 2021-08-13 | 北京大学第三医院（北京大学第三临床医学院） | Deep learning neural network model system for multi-modal image synthesis |
| CN113537025B (en) * | 2021-07-08 | 2024-03-22 | 浙江工业大学 | Electromagnetic modulation signal deep denoising method and system based on self-supervision learning |
| CN113537025A (en) * | 2021-07-08 | 2021-10-22 | 浙江工业大学 | Electromagnetic modulation signal deep denoising method and system based on self-supervision learning |
| CN113723213A (en) * | 2021-08-06 | 2021-11-30 | 武汉理工大学 | Method for realizing subway vibration signal noise reduction training based on Hilbert curve coding |
| CN113723213B (en) * | 2021-08-06 | 2023-11-10 | 武汉理工大学 | Method for realizing noise reduction training of subway vibration signals based on Hilbert curve coding |
| CN114219820A (en) * | 2021-12-08 | 2022-03-22 | 苏州工业园区智在天下科技有限公司 | Neural network generation method, denoising method and device |
| CN114548156A (en) * | 2022-01-24 | 2022-05-27 | 成都理工大学 | Distributed optical fiber temperature measurement and noise reduction method based on down-sampling and convolutional neural network |
| CN114545504B (en) * | 2022-02-18 | 2023-05-09 | 吉林大学 | Desert seismic prospecting random noise modeling method based on deep learning |
| CN114545504A (en) * | 2022-02-18 | 2022-05-27 | 吉林大学 | Desert seismic exploration random noise modeling method based on deep learning |
| CN114596223A (en) * | 2022-02-24 | 2022-06-07 | 复旦大学 | Quantization parameter self-adaptive convolution neural network loop filter and construction method thereof |
| CN114757847A (en) * | 2022-04-24 | 2022-07-15 | 汕头市超声仪器研究所股份有限公司 | Multi-information extraction extended U-Net and application method thereof in low-dose X-ray imaging |
| CN116908760A (en) * | 2023-07-11 | 2023-10-20 | 哈尔滨工业大学 | MRI bias field correction method and system based on depth decoupling network |
| CN116908760B (en) * | 2023-07-11 | 2024-05-14 | 哈尔滨工业大学 | MRI bias field correction method and system based on depth decoupling network |
| CN117095074B (en) * | 2023-08-23 | 2024-04-09 | 浙江大学 | Medical image double denoising method and device based on deep learning |
| CN117095074A (en) * | 2023-08-23 | 2023-11-21 | 浙江大学 | Medical image double denoising method and device based on deep learning |
| CN117392545B (en) * | 2023-10-26 | 2024-02-09 | 南昌航空大学 | SAR image target detection method based on deep learning |
| CN117392545A (en) * | 2023-10-26 | 2024-01-12 | 南昌航空大学 | SAR image target detection method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109978778B (en) | 2020-11-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109978778A (en) | Convolutional neural networks medicine CT image denoising method based on residual error study | |
| JP2020503095A (en) | Machine learning of anatomical model parameters | |
| CN110074813A (en) | A kind of ultrasonic image reconstruction method and system | |
| CN103985099B (en) | Dispersion tensor magnetic resonance image tensor domain non-local mean denoising method | |
| CN110473150A (en) | CNN medicine CT image denoising method based on multi-feature extraction | |
| CN110010229B (en) | Ultrasonic image simulation method based on generation of countermeasure network | |
| Zhao et al. | Ultrasound transmission tomography image reconstruction with a fully convolutional neural network | |
| CN103229210A (en) | Image registration apparatus | |
| Song et al. | Denoising of MR and CT images using cascaded multi-supervision convolutional neural networks with progressive training | |
| Luchies et al. | Training improvements for ultrasound beamforming with deep neural networks | |
| Peng et al. | A real-time medical ultrasound simulator based on a generative adversarial network model | |
| Prasad et al. | DeepUCT: Complex cascaded deep learning network for improved ultrasound tomography | |
| Fan et al. | Model-data-driven image reconstruction with neural networks for ultrasound computed tomography breast imaging | |
| Brickson et al. | Reverberation noise suppression in the aperture domain using 3D fully convolutional neural networks | |
| Zhang et al. | Photoacoustic digital brain and deep-learning-assisted image reconstruction | |
| CN113947662B (en) | Ultrasonic simulation method and system based on medical tomography | |
| Gao et al. | Medical image segmentation algorithm for three-dimensional multimodal using deep reinforcement learning and big data analytics | |
| CN117115452B (en) | Controllable medical ultrasonic image denoising method, system and computer storage medium | |
| Long et al. | Deep Learning Ultrasound Computed Tomography under Sparse Sampling | |
| CN115778412B (en) | Method and device for optimizing contrast agent dosage in X-ray photo-acoustic imaging and storage medium | |
| Abd-Elmoniem et al. | Direct pixel to pixel principal strain mapping from tagging MRI using end to end deep convolutional neural network (DeepStrain) | |
| Danesh et al. | Automatic production of synthetic labelled OCT images using an active shape model | |
| FLOYD Jr et al. | An artificial neural network for lesion detection on single-photon emission computed tomographic images | |
| Singh et al. | An advanced technique of de-noising medical images using ANFIS | |
| Lyu et al. | Photoacoustic digital brain: numerical modelling and image reconstruction via deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |