CN108629784A - A kind of CT image intracranial vessel dividing methods and system based on deep learning - Google Patents
A kind of CT image intracranial vessel dividing methods and system based on deep learning Download PDFInfo
- Publication number
- CN108629784A CN108629784A CN201810433128.5A CN201810433128A CN108629784A CN 108629784 A CN108629784 A CN 108629784A CN 201810433128 A CN201810433128 A CN 201810433128A CN 108629784 A CN108629784 A CN 108629784A
- Authority
- CN
- China
- Prior art keywords
- data
- layers
- layer
- deep learning
- data set
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The present invention provides a kind of CT image intracranial vessel dividing methods and system based on deep learning, including：Collect markers step：It collects more set head CTA datas and vessel position is marked, and be divided into training dataset, validation data set and test data set；Pretreatment and augmentation step：Pretreatment operation and augmentation operation are carried out to data set, pretreatment operation includes normalization operation, whitening operation, the interlamellar spacing of 3 dimensions of data set is set as identical by the method for resampling, fritter volume data is divided by set spatial window size to CT data, augmentation operation includes being rotated, being amplified to data set, reduced and symmetry transformation；Three dimensional convolution neural metwork training step：Three dimensional convolution neural network is built by structure, is trained by parameter and using training data, verify data.It is bigger this invention removes background voxels and blood vessel voxel proportional difference and cause train when the unbalanced problem of classification, to improve the accuracy rate of intracranial vessel segmentation result.
Description
Technical field
The present invention relates to technical field of medical image processing, and in particular, to a kind of CT image craniums based on deep learning
Interior blood vessel segmentation method and system.
Background technology
Blood vessel segmentation in medical image is the key technology of blood vessel imaging system, and carries out blood vessel three-dimensional visualization,
Morphological measurement and the important step of computer-aided diagnosis.With CT (Computed Tomography, computerized tomography
Scanning), MRI (Magnetic Resonance Imaging, Magnetic resonance imaging), the development of these advanced imaging techniques and
The progress of Angiography is more clear using obtained blood-vessel image, while also having promoted many domestic and foreign scholars in blood
Pipe divides the research in field.In recent years, many profession scholars also proposed many methods, include the dividing method based on region
(such as histogram threshold method, color cluster method and region-growing method), (such as edge detection method divides water to the dividing method based on edge
Ridge method), (such as dividing method based on wavelet theory, the dividing method based on Mathematical Morphology and base based on specific theory model
In the dividing method of genetic algorithm) etc..
With the fast development in recent deep learning field, deep learning and deep neural network are in traditional computer vision
Field achieves prodigious development, and in the identification of RGB image, positioning, the tasks such as semantic segmentation, deep neural network can surpass
More and close to the mankind level, therefore this also promotes deep learning to the development in medical image analysis field.
In the blood vessel segmentation field of CT images, traditional method, whether based on region or based on edge all
There are not high enough accuracy rate, algorithm sphere of action is limited to very much, and the problems such as not enough grade for the robustness of noise, and part is calculated
The operation time of method is also very long.Particularly, for the blood vessel segmentation task of encephalic, these traditional algorithms are in part blood vessel and skull
Close adjacent region simultaneously could not clearly come out blood vessel segmentation.
And for emerging deep learning method, since current deep learning method is in the general of medical image analysis field
And degree is not high, is not directed to the research that deep learning divides task in the intracranial vessel of CT images.
Deep learning is in field of medical images there are one limitation precisely due to the 3 d medical images such as CT are compared to general
Logical two-dimentional RGB image, the scale of construction is larger, and increase out a dimension makes convolutional network model all be needed in training and test
It wants prodigious video memory to carry out the operation of support model, and is difficult to find video card good enough to support the fortune of an entire CT data at present
It calculates.
Invention content
For the defects in the prior art, the object of the present invention is to provide a kind of CT image encephalic blood based on deep learning
Pipe dividing method and system.
According to a kind of CT image intracranial vessel dividing methods based on deep learning provided by the invention, including：
Collect markers step：It collects more set head CTA datas and vessel position is marked, by the head after label
CTA data is divided into training dataset, validation data set and test data set；
Pre-process augmentation step：Pretreatment operation and augmentation operation, the pretreatment operation packet are carried out to head CTA data
Normalization operation and whitening operation are included, the interlamellar spacing of 3 dimensions of data set is set as identical by the method for resampling, it is correct
Cranium CTA data is divided into fritter volume data by set spatial window size, and the augmentation operation includes being carried out to head CTA data
Rotation, amplification, diminution and symmetry transformation；
Three dimensional convolution neural metwork training step：It builds Three dimensional convolution neural network and uses the training dataset and institute
Validation data set is stated to be trained.
Preferably, the augmentation operation is specifically included and is put to 10 degree of anticlockwise to 10 degree of right rotation to head CTA data
Greatly to 1.2 times, 0.8 times is narrowed down to, transformation symmetrical above and below and symmetrical transformation.
Preferably, the Three dimensional convolution neural network is depth residual error convolutional neural networks, input terminal directly passes through addition
It is merged into its output end, residual error connection procedure is expressed as：
xp+1=xp+F(xp,wp)
Wherein xpFor the input terminal of p layers of residual error link block, xp+1For output end, F (xp, wp) be expressed as residual error connection it is in the block
Nonlinear function layer.
Preferably, the depth residual error convolutional neural networks are 20 layers, the first 7 layers Three dimensional convolution layer for 3*3*3 is followed by criticizing
Process layer and rectification linear layer, latter 12 layers are followed by batch processing layer and rectification linear layer for empty convolutional layer, last layer is
Softmax layers.
Preferably, first 6 layers of the cavity convolutional layer are amplified 2 times, latter 6 layers are amplified 4 times.
According to a kind of CT image intracranial vessel segmenting systems based on deep learning provided by the invention, including：
Collect mark module：It collects more set head CTA datas and vessel position is marked, by the head after label
CTA data is divided into training dataset, validation data set and test data set；
Pre-process augmentation module：Pretreatment operation and augmentation operation, the pretreatment operation packet are carried out to head CTA data
Normalization operation and whitening operation are included, the interlamellar spacing of 3 dimensions of data set is set as identical by the method for resampling, it is correct
Cranium CTA data is divided into fritter volume data by set spatial window size, and the augmentation operation includes being carried out to head CTA data
Rotation, amplification, diminution and symmetry transformation；
Three dimensional convolution neural metwork training module：It builds Three dimensional convolution neural network and uses the training dataset and institute
Validation data set is stated to be trained.
Preferably, the augmentation operation is specifically included and is put to 10 degree of anticlockwise to 10 degree of right rotation to head CTA data
Greatly to 1.2 times, 0.8 times is narrowed down to, transformation symmetrical above and below and symmetrical transformation.
Preferably, the Three dimensional convolution neural network is depth residual error convolutional neural networks, input terminal directly passes through addition
It is merged into its output end, residual error connection procedure is expressed as：
xp+1=xp+F(xp,wp)
Wherein xpFor the input terminal of p layers of residual error link block, xp+1For output end, F (xp,wp) be expressed as residual error connection it is in the block
Nonlinear function layer.
Preferably, the depth residual error convolutional neural networks are 20 layers, the first 7 layers Three dimensional convolution layer for 3*3*3 is followed by criticizing
Process layer and rectification linear layer, latter 12 layers are followed by batch processing layer and rectification linear layer for empty convolutional layer, last layer is
Softmax layers.
Preferably, first 6 layers of the cavity convolutional layer are amplified 2 times, latter 6 layers are amplified 4 times.
Compared with prior art, the present invention has following advantageous effect：
1, the accuracy rate for improving intracranial vessel segmentation result, has measured 90.75% on data set
DiceSimilarity Coefficient accuracys rate, can be clear blood vessel in part blood vessel and the close adjacent region of skull
Split；
2, background voxels are eliminated and blood vessel voxel is more larger than row diversity ratio and lead to the unbalanced problem of class；
3, deeper network can be trained by using the structure of residual error network, also accelerate network convergence, improve segmentation essence
Exactness；
4, the method for using empty convolution (amplification convolution) can be such that feature is calculated in very high spatial resolution, and
And the size of receptive field can be by the amplification of arbitrary scale；
5, reduce the time needed for intracranial vessel segmentation；
6, robustness of the intracranial vessel segmentation for noise is improved.
Description of the drawings
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon：
Fig. 1 is the flow chart of the present invention；
Fig. 2 is the schematic diagram of the residual error connection of the present invention.
Specific implementation mode
With reference to specific embodiment, the present invention is described in detail.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, several changes and improvements can also be made.These belong to the present invention
Protection domain.
As shown in Figure 1, a kind of CT image intracranial vessel dividing methods based on deep learning, including：
Collect markers step：It collects more set head CTA (CT angiographies) data and vessel position is marked, will mark
Head CTA data after note is divided into training data, verify data and data set.
70 sets of clinical head CTA datas are had collected in the present embodiment, and by the professional person of iconography in data
Vessel position is marked, and has been separated into 50 sets for training data, and 10 sets are verify data, 10 sets of numbers for test data
According to collection.
Pre-process augmentation step：Pretreatment operation and augmentation operation, the pretreatment operation packet are carried out to head CTA data
It includes and the interlamellar spacing of 3 dimensions of head CTA data is set as identical by the method for resampling, to head CTA data according to window
Roomy small segmentation volumetric data, the augmentation operation includes being rotated, being amplified to head CTA data, reduced and symmetrically become
It changes.
Summarize in the present embodiment, since data set is collected so there are the interlamellar spacing between CTA data is different, passes through weight
The interlamellar spacing of 3 dimensions of CTA data is all set as 0.5mm by the method for sampling.When being run for the convenience and program of data processing
Convergence is accelerated, and data is normalized and whitening operation.It is one whole due to there is no enough video memorys that can support for current hardware
The operation of a CTA data is divided into CT data according to the window width size of (96,96,96) volume data of fritter.Due to data set
Very little over-fitting and convolutional neural networks itself of easily leading to be not strong to the robustness in rotational invariance, so also to data
Collection has been done to 10 degree of anticlockwise, to 10 degree of right rotation, is amplified to 1.2 times, narrows down to 0.8 times, symmetrical above and below to convert and control pair
Claim the augmentation operation of transformation.
Three dimensional convolution neural metwork training step：Build Three dimensional convolution neural network and using the training data, described
Verify data is trained.
The present embodiment uses the P3xlarge server instances of AWS (Amazon Web Service) offers, specifically matches
Be set to one it is tall and handsome reach Tesla V100GPU, video memory 16GB, 8 virtual cpus, the memory of 61GB.Experiment software：Operation system
System is crow 64 versions of class Figure 16 .04LTS, and deep learning library is Tensorflow 1.3.0 versions, and acceleration tool has used tall and handsome
The cuda9.0 and cuDNN7.1 reached.
The intracranial vessel segmentation network designed in the present invention, is 20 layers of depth residual error convolutional neural networks, just like Fig. 2 institutes
The residual error link block shown is constituted.Its core concept be identical mapping connection is created to bypass the network layer of parameter, so it
Input terminal be directed through addition and be merged into its output end.Experiments have shown that residual error connection can be information in network layer in this way
It is middle transmit it is smoother, training speed also faster, to the more deep neural network of the number of plies before solving, because the number of plies is too
It is difficult to trained problem more.
Entire residual error connection procedure can be expressed as：
xp+1=xp+F(xp, wp)
Wherein xpFor the input terminal of p layers of residual error link block, xp+1For output end, F (xp,wp) be expressed as residual error connection it is in the block
Nonlinear function layer.
In this 20 layers depth residual error convolutional neural networks, the Three dimensional convolution layer of preceding 7 layers of 3*3*3 is followed by batch processing layer and whole
Cleanliness layer, latter 12 layers are followed by batch processing layer and rectification linear layer, preceding 6 layers of amplification therein for empty convolution (amplification convolution) layer
2 times, latter 6 layers are amplified 4 times.
Wherein, compared to the structure of up-sampling and down-sampling in traditional convolutional neural networks, empty convolution (amplification volume
Product) method feature can be made to be calculated in very high spatial resolution, and the size of receptive field can be by arbitrary scale
Amplification.The expression formula of empty convolution (amplification convolution) is as follows：
Wherein r is the multiple of amplification, and I is the characteristic pattern of input, and M is its port number, and O is the characteristic pattern of output.This is empty
Hole convolution (amplification convolution) core and common 3*3*3 convolution kernels have it is identical can training parameter, but it can be with point of shelf space
Resolution provides (2r+1)3The receptive field of voxel, by adjusting r, it can theoretically provide the receptive field of arbitrary scale.
Last layer of network is softmax layers, it can calculate probability score of each voxel to each class.But by
In this typical medical image segmentation problem, there is very serious class imbalance (background voxels and blood vessel in it
Voxel is more larger than row diversity ratio and leads to the unbalanced problem of class), if this can lead to that with traditional loss function result can be allowed
The big class of comparative example has very strong bigoted.For this purpose, using losses of the Dice Similarity Coefficient as network
Function, expression formula are as follows：
Wherein { xnBe N number of voxel volume data, { ynBe classification number, δ corresponding is Dirac delta functions, and
Fc(xn) it is xnSoftmax classification for c classes probability score.
Analytical procedure：Test sample is input to Three dimensional convolution neural network, obtains the segmentation result figure of blood vessel.
On the basis of a kind of above-mentioned CT image intracranial vessel dividing methods based on deep learning, the present invention also provides
A kind of CT image intracranial vessel segmenting systems based on deep learning, including：
Collect mark module：It collects more set head CTA (CT angiographies) data and vessel position is marked, will mark
Head CTA data after note is divided into training data, verify data and data set.
70 sets of clinical head CTA datas are had collected in the present embodiment, and by the professional person of iconography in data
Vessel position is marked, and has been separated into 50 sets for training data, and 10 sets are verify data, 10 sets of numbers for test data
According to collection.
Pre-process augmentation module：Pretreatment operation and augmentation operation, the pretreatment operation packet are carried out to head CTA data
It includes and the interlamellar spacing of 3 dimensions of head CTA data is set as identical by the method for resampling, to head CTA data according to window
Roomy small segmentation volumetric data, the augmentation operation includes being rotated, being amplified to head CTA data, reduced and symmetrically become
It changes.
Summarize in the present embodiment, since data set is collected so there are the interlamellar spacing between CTA data is different, passes through weight
The interlamellar spacing of 3 dimensions of CTAA data is all set as 0.5mm by the method for sampling.When being run for the convenience and program of data processing
Convergence is accelerated, and data is normalized and whitening operation.It is one whole due to there is no enough video memorys that can support for current hardware
The operation of a CTA data is divided into CTA data according to the window width size of (96,96,96) volume data of fritter.Due to data
Collection easily leads to over-fitting and convolutional neural networks itself not strong to the robustness in rotational invariance, so going back logarithm very little
It has been done to 10 degree of anticlockwise according to collection, to 10 degree of right rotation, has been amplified to 1.2 times, narrowed down to 0.8 times, transformation symmetrical above and below and left and right
The augmentation of symmetry transformation operates.
Three dimensional convolution neural metwork training module：Build Three dimensional convolution neural network and using the training data, described
Verify data is trained.
The present embodiment uses the P3xlarge server instances of AWS (Amazon Web Service) offers, specifically matches
Be set to one it is tall and handsome reach Tesla V100GPU, video memory 16GB, 8 virtual cpus, the memory of 61GB.Experiment software：Operation system
System is crow 64 versions of class Figure 16 .04LTS, and deep learning library is Tensorflow 1.3.0 versions, and acceleration tool has used tall and handsome
The cuda9.0 and cuDNN7.1 reached.
The intracranial vessel segmentation network designed in the present invention, is 20 layers of depth residual error convolutional neural networks, just like Fig. 2 institutes
The residual error link block shown is constituted.Its core concept be identical mapping connection is created to bypass the network layer of parameter, so it
Input terminal be directed through addition and be merged into its output end.Experiments have shown that residual error connection can be information in network layer in this way
It is middle transmit it is smoother, training speed also faster, to the more deep neural network of the number of plies before solving, because the number of plies is too
It is difficult to trained problem more.
Entire residual error connection procedure can be expressed as：
xp+1=xp+F(xp,wp)
Wherein xpFor the input terminal of p layers of residual error link block, xp+1For output end, F (xp,wp) be expressed as residual error connection it is in the block
Nonlinear function layer.
In this 20 layers depth residual error convolutional neural networks, the Three dimensional convolution layer of preceding 7 layers of 3*3*3 is followed by batch processing layer and whole
Cleanliness layer, latter 12 layers are followed by batch processing layer and rectification linear layer, preceding 6 layers of amplification therein for empty convolution (amplification convolution) layer
2 times, latter 6 layers are amplified 4 times.
Wherein, compared to the structure of up-sampling and down-sampling in traditional convolutional neural networks, empty convolution (amplification volume
Product) method feature can be made to be calculated in very high spatial resolution, and the size of receptive field can be by arbitrary scale
Amplification.The expression formula of empty convolution (amplification convolution) is as follows：
Wherein r is the multiple of amplification, and I is the characteristic pattern of input, and M is its port number, and O is the characteristic pattern of output.This is empty
Hole convolution (amplification convolution) core and common 3*3*3 convolution kernels have it is identical can training parameter, but it can be with point of shelf space
Resolution provides (2r+1)3The receptive field of voxel, by adjusting r, it can theoretically provide the receptive field of arbitrary scale.
Last layer of network is softmax layers, it can calculate probability score of each voxel to each class.But by
In this typical medical image segmentation problem, there is very serious class imbalance (background voxels and blood vessel in it
Voxel is more larger than row diversity ratio and leads to the unbalanced problem of class), if this can lead to that with traditional loss function result can be allowed
The big class of comparative example has very strong bigoted.For this purpose, using losses of the Dice Similarity Coefficient as network
Function, expression formula are as follows：
Wherein { xnBe N number of voxel volume data, { ynBe classification number, δ corresponding is Dirac delta functions, and
Fc(xn) it is xnSoftmax classification for c classes probability score.
Analysis module：Test sample is input to Three dimensional convolution neural network, obtains the segmentation result figure of blood vessel.
One skilled in the art will appreciate that in addition to realizing system provided by the invention in a manner of pure computer readable program code
It, completely can be by the way that method and step be carried out programming in logic come so that the present invention provides and its other than each device, module, unit
System and its each device, module, unit with logic gate, switch, application-specific integrated circuit, programmable logic controller (PLC) and embedding
Enter the form of the controller that declines etc. to realize identical function.So system provided by the invention and its every device, module, list
Member is considered a kind of hardware component, and also may be used for realizing the device of various functions, module, unit to include in it
To be considered as the structure in hardware component；It can also will be considered as realizing the device of various functions, module, unit either real
The software module of existing method can be the structure in hardware component again.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or change within the scope of the claims, this not shadow
Ring the substantive content of the present invention.In the absence of conflict, the feature in embodiments herein and embodiment can arbitrary phase
Mutually combination.
Claims (10)
1. a kind of CT image intracranial vessel dividing methods based on deep learning, which is characterized in that including：
Collect markers step：It collects more set head CTA datas and vessel position is marked, by the head CT A numbers after label
According to being divided into training dataset, validation data set and test data set；
Pre-process augmentation step：Pretreatment operation and augmentation operation are carried out to head CTA data, the pretreatment operation includes returning
One changes operation and whitening operation, the interlamellar spacing of 3 dimensions of data set is set as identical by the method for resampling, to head
CTA data is divided into fritter volume data by set spatial window size, and the augmentation operation includes being revolved to head CTA data
Turn, amplification, reduce and symmetry transformation；
Three dimensional convolution neural metwork training step：Structure Three dimensional convolution neural network simultaneously using the training dataset and described is tested
Card data set is trained.
2. the CT image intracranial vessel dividing methods according to claim 1 based on deep learning, which is characterized in that described
Augmentation operation is specifically included to be amplified to 1.2 times, narrows down to the head CTA data to 10 degree of anticlockwise to 10 degree of right rotation
It is 0.8 times, symmetrical above and below to convert and symmetrical transformation.
3. the CT image intracranial vessel dividing methods according to claim 1 based on deep learning, which is characterized in that described
Three dimensional convolution neural network is depth residual error convolutional neural networks, and input terminal is directly merged into its output end by addition, residual
Poor connection procedure is expressed as：
xp+1=xp+F(xp,wp)
Wherein xpFor the input terminal of p layers of residual error link block, xp+1For output end, F (xp,wp) be expressed as residual error connection it is in the block non-thread
Property function layer.
4. the CT image intracranial vessel dividing methods according to claim 3 based on deep learning, which is characterized in that described
Depth residual error convolutional neural networks are 20 layers, and first 7 layers are followed by batch processing layer and rectification linear layer for the Three dimensional convolution layer of 3*3*3,
It is followed by batch processing layer and rectification linear layer for empty convolutional layer for 12 layers afterwards, last layer is softmax layers.
5. the CT image intracranial vessel dividing methods according to claim 3 based on deep learning, which is characterized in that described
First 6 layers of empty convolutional layer amplify 2 times, and latter 6 layers are amplified 4 times.
6. a kind of CT image intracranial vessel segmenting systems based on deep learning, which is characterized in that including：
Collect mark module：It collects more set head CTA datas and vessel position is marked, by the head CT A numbers after label
According to being divided into training dataset, validation data set and test data set；
Pre-process augmentation module：Pretreatment operation and augmentation operation are carried out to head CTA data, the pretreatment operation includes returning
One changes operation and whitening operation, the interlamellar spacing of 3 dimensions of data set is set as identical by the method for resampling, to head
CTA data is divided into fritter volume data by set spatial window size, and the augmentation operation includes being revolved to head CTA data
Turn, amplification, reduce and symmetry transformation；
Three dimensional convolution neural metwork training module：Structure Three dimensional convolution neural network simultaneously using the training dataset and described is tested
Card data set is trained.
7. the CT image intracranial vessel segmenting systems according to claim 6 based on deep learning, which is characterized in that described
Augmentation operation is specifically included to be amplified to 1.2 times, narrows down to 0.8 to head CTA data to 10 degree of anticlockwise to 10 degree of right rotation
Times, transformation symmetrical above and below and symmetrical transformation.
8. the CT image intracranial vessel segmenting systems according to claim 6 based on deep learning, which is characterized in that described
Three dimensional convolution neural network is depth residual error convolutional neural networks, and input terminal is directly merged into its output end by addition, residual
Poor connection procedure is expressed as：
xp+1=xp+F(xp,wp)
Wherein xpFor the input terminal of p layers of residual error link block, xp+1For output end, F (xp,wp) be expressed as residual error connection it is in the block non-thread
Property function layer.
9. the CT image intracranial vessel segmenting systems according to claim 8 based on deep learning, which is characterized in that described
Depth residual error convolutional neural networks are 20 layers, and first 7 layers are followed by batch processing layer and rectification linear layer for the Three dimensional convolution layer of 3*3*3,
It is followed by batch processing layer and rectification linear layer for empty convolutional layer for 12 layers afterwards, last layer is softmax layers.
10. the CT image intracranial vessel segmenting systems according to claim 9 based on deep learning, which is characterized in that institute
State empty convolutional layer first 6 layers amplify 2 times, and latter 6 layers are amplified 4 times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810433128.5A CN108629784A (en) | 2018-05-08 | 2018-05-08 | A kind of CT image intracranial vessel dividing methods and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810433128.5A CN108629784A (en) | 2018-05-08 | 2018-05-08 | A kind of CT image intracranial vessel dividing methods and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108629784A true CN108629784A (en) | 2018-10-09 |
Family
ID=63695937
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810433128.5A Pending CN108629784A (en) | 2018-05-08 | 2018-05-08 | A kind of CT image intracranial vessel dividing methods and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108629784A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109448003A (en) * | 2018-10-26 | 2019-03-08 | 强联智创（北京）科技有限公司 | A kind of entocranial artery blood-vessel image dividing method and system |
| CN109754403A (en) * | 2018-11-29 | 2019-05-14 | 中国科学院深圳先进技术研究院 | Tumour automatic division method and system in a kind of CT image |
| CN109919961A (en) * | 2019-02-22 | 2019-06-21 | 北京深睿博联科技有限责任公司 | A kind of processing method and processing device for aneurysm region in encephalic CTA image |
| CN110189306A (en) * | 2019-05-14 | 2019-08-30 | 上海联影智能医疗科技有限公司 | The corresponding compressing vessels of abnormal area determine method and apparatus in brain image |
| CN110390351A (en) * | 2019-06-24 | 2019-10-29 | 浙江大学 | A kind of Epileptic focus three-dimensional automatic station-keeping system based on deep learning |
| CN111028248A (en) * | 2019-12-19 | 2020-04-17 | 杭州健培科技有限公司 | Method and device for separating static and dynamic pulses based on CT (computed tomography) image |
| CN111369567A (en) * | 2018-12-26 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Method and device for segmenting target object in three-dimensional image and electronic equipment |
| CN112561868A (en) * | 2020-12-09 | 2021-03-26 | 深圳大学 | Cerebrovascular segmentation method based on multi-view cascade deep learning network |
| CN112785581A (en) * | 2021-01-29 | 2021-05-11 | 复旦大学附属中山医院 | Training method and device for extracting and training large blood vessel CTA (computed tomography angiography) imaging based on deep learning |
| CN112822982A (en) * | 2018-10-10 | 2021-05-18 | 株式会社岛津制作所 | Image creation device, image creation method, and method for creating learned model |
| CN112907563A (en) * | 2021-03-17 | 2021-06-04 | 复旦大学附属华山医院 | CTA full-automatic collateral circulation scoring method and system based on deep learning |
| CN113269764A (en) * | 2021-06-04 | 2021-08-17 | 重庆大学 | Automatic segmentation method and system for intracranial aneurysm, sample processing method and model training method |
| CN113313728A (en) * | 2021-06-16 | 2021-08-27 | 首都医科大学附属北京天坛医院 | Intracranial artery segmentation method and system |
| CN113689454A (en) * | 2020-05-18 | 2021-11-23 | 傅豪 | 3D CT vertebral body segmentation algorithm based on convolutional neural network |
| US20210382127A1 (en) * | 2019-01-25 | 2021-12-09 | Xiamen University | Method for reconstructing magnetic resonance spectrum based on deep learning |
| CN114612404A (en) * | 2022-03-04 | 2022-06-10 | 清华大学 | Blood vessel segmentation method, device, storage medium and electronic equipment |
| WO2022188695A1 (en) * | 2021-03-10 | 2022-09-15 | 腾讯科技（深圳）有限公司 | Data processing method, apparatus, and device, and medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
| CN107369189A (en) * | 2017-07-21 | 2017-11-21 | 成都信息工程大学 | The medical image super resolution ratio reconstruction method of feature based loss |
| CN107767380A (en) * | 2017-12-06 | 2018-03-06 | 电子科技大学 | A kind of compound visual field skin lens image dividing method of high-resolution based on global empty convolution |
| CN107909585A (en) * | 2017-11-14 | 2018-04-13 | 华南理工大学 | Inner membrance dividing method in a kind of blood vessel of intravascular ultrasound image |
-
2018
- 2018-05-08 CN CN201810433128.5A patent/CN108629784A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
| CN107369189A (en) * | 2017-07-21 | 2017-11-21 | 成都信息工程大学 | The medical image super resolution ratio reconstruction method of feature based loss |
| CN107909585A (en) * | 2017-11-14 | 2018-04-13 | 华南理工大学 | Inner membrance dividing method in a kind of blood vessel of intravascular ultrasound image |
| CN107767380A (en) * | 2017-12-06 | 2018-03-06 | 电子科技大学 | A kind of compound visual field skin lens image dividing method of high-resolution based on global empty convolution |
Non-Patent Citations (3)
| Title |
|---|
| JELMER M. WOLTERINK ET AL: "Dilated Convolutional Neural Networks for Cardiovascular MR Segmentation in Congenital Heart Disease", 《RAMBO 2016, HVSMR 2016: RECONSTRUCTION, SEGMENTATION, AND ANALYSIS OF MEDICAL IMAGES 》 * |
| WENQI LI ET AL: "On the Compactness, Efficiency, and Representation of 3D Convolutional Networks: Brain Parcellation as a Pretext Task", 《IPMI 2017: INFORMATION PROCESSING IN MEDICAL IMAGING 》 * |
| XINZE CHEN ET AL: "Semantic Segmentation with Modified Deep Residual Networks", 《CCPR 2016: PATTERN RECOGNITION》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112822982A (en) * | 2018-10-10 | 2021-05-18 | 株式会社岛津制作所 | Image creation device, image creation method, and method for creating learned model |
| CN112822982B (en) * | 2018-10-10 | 2023-09-15 | 株式会社岛津制作所 | Image forming apparatus, image forming method, and method for forming learning model |
| CN109448003A (en) * | 2018-10-26 | 2019-03-08 | 强联智创（北京）科技有限公司 | A kind of entocranial artery blood-vessel image dividing method and system |
| CN109754403A (en) * | 2018-11-29 | 2019-05-14 | 中国科学院深圳先进技术研究院 | Tumour automatic division method and system in a kind of CT image |
| CN111369567A (en) * | 2018-12-26 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Method and device for segmenting target object in three-dimensional image and electronic equipment |
| US11782111B2 (en) * | 2019-01-25 | 2023-10-10 | Xiamen University | Method for reconstructing magnetic resonance spectrum based on deep learning |
| US20210382127A1 (en) * | 2019-01-25 | 2021-12-09 | Xiamen University | Method for reconstructing magnetic resonance spectrum based on deep learning |
| CN109919961A (en) * | 2019-02-22 | 2019-06-21 | 北京深睿博联科技有限责任公司 | A kind of processing method and processing device for aneurysm region in encephalic CTA image |
| CN110189306B (en) * | 2019-05-14 | 2021-02-19 | 上海联影智能医疗科技有限公司 | Method and device for determining responsible blood vessels corresponding to abnormal regions in brain image |
| CN110189306A (en) * | 2019-05-14 | 2019-08-30 | 上海联影智能医疗科技有限公司 | The corresponding compressing vessels of abnormal area determine method and apparatus in brain image |
| WO2020224123A1 (en) * | 2019-06-24 | 2020-11-12 | 浙江大学 | Deep learning-based seizure focus three-dimensional automatic positioning system |
| CN110390351A (en) * | 2019-06-24 | 2019-10-29 | 浙江大学 | A kind of Epileptic focus three-dimensional automatic station-keeping system based on deep learning |
| US11645748B2 (en) | 2019-06-24 | 2023-05-09 | Zhejiang University | Three-dimensional automatic location system for epileptogenic focus based on deep learning |
| CN111028248A (en) * | 2019-12-19 | 2020-04-17 | 杭州健培科技有限公司 | Method and device for separating static and dynamic pulses based on CT (computed tomography) image |
| CN113689454A (en) * | 2020-05-18 | 2021-11-23 | 傅豪 | 3D CT vertebral body segmentation algorithm based on convolutional neural network |
| CN112561868A (en) * | 2020-12-09 | 2021-03-26 | 深圳大学 | Cerebrovascular segmentation method based on multi-view cascade deep learning network |
| CN112785581A (en) * | 2021-01-29 | 2021-05-11 | 复旦大学附属中山医院 | Training method and device for extracting and training large blood vessel CTA (computed tomography angiography) imaging based on deep learning |
| WO2022188695A1 (en) * | 2021-03-10 | 2022-09-15 | 腾讯科技（深圳）有限公司 | Data processing method, apparatus, and device, and medium |
| CN112907563A (en) * | 2021-03-17 | 2021-06-04 | 复旦大学附属华山医院 | CTA full-automatic collateral circulation scoring method and system based on deep learning |
| CN112907563B (en) * | 2021-03-17 | 2023-12-01 | 复旦大学附属华山医院 | CTA full-automatic collateral circulation scoring method and system based on deep learning |
| CN113269764A (en) * | 2021-06-04 | 2021-08-17 | 重庆大学 | Automatic segmentation method and system for intracranial aneurysm, sample processing method and model training method |
| CN113313728A (en) * | 2021-06-16 | 2021-08-27 | 首都医科大学附属北京天坛医院 | Intracranial artery segmentation method and system |
| CN114612404A (en) * | 2022-03-04 | 2022-06-10 | 清华大学 | Blood vessel segmentation method, device, storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108629784A (en) | A kind of CT image intracranial vessel dividing methods and system based on deep learning | |
| US10867384B2 (en) | System and method for automatically detecting a target object from a 3D image | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| US9968257B1 (en) | Volumetric quantification of cardiovascular structures from medical imaging | |
| Wang et al. | A multi-view deep convolutional neural networks for lung nodule segmentation | |
| Saikumar et al. | A novel implementation heart diagnosis system based on random forest machine learning technique. | |
| US9959615B2 (en) | System and method for automatic pulmonary embolism detection | |
| WO2020260936A1 (en) | Medical image segmentation using an integrated edge guidance module and object segmentation network | |
| EP3714467A2 (en) | Content based image retrieval for lesion analysis | |
| EP3629898A1 (en) | Automated lesion detection, segmentation, and longitudinal identification | |
| CN110934606A (en) | Cerebral apoplexy early-stage flat-scan CT image evaluation system and method and readable storage medium | |
| CN112258530A (en) | Neural network-based computer-aided lung nodule automatic segmentation method | |
| CN112150428A (en) | Medical image segmentation method based on deep learning | |
| CN112263217B (en) | Improved convolutional neural network-based non-melanoma skin cancer pathological image lesion area detection method | |
| CN112862756A (en) | Method for identifying pathological change type and gene mutation in thyroid tumor pathological image | |
| Gamage et al. | Instance-based segmentation for boundary detection of neuropathic ulcers through Mask-RCNN | |
| CN110570419A (en) | Method and device for acquiring characteristic information and storage medium | |
| CN114494215A (en) | Transformer-based thyroid nodule detection method | |
| CN114600155A (en) | Weakly supervised multitask learning for cell detection and segmentation | |
| Irene et al. | Segmentation and approximation of blood volume in intracranial hemorrhage patients based on computed tomography scan images using deep learning method | |
| He et al. | Automatic left ventricle segmentation from cardiac magnetic resonance images using a capsule network | |
| Furuzuki et al. | A detection method for liver cancer region based on faster R-CNN | |
| CN111476802B (en) | Medical image segmentation and tumor detection method, equipment and readable storage medium | |
| CN116958705A (en) | Medical image classifying system based on graph neural network | |
| CN111507431A (en) | Medical image classification method, device, system and readable medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181009 |
|
| RJ01 | Rejection of invention patent application after publication |