CN108428225A - Image department brain image fusion identification method based on multiple dimensioned multiple features - Google Patents
Image department brain image fusion identification method based on multiple dimensioned multiple features Download PDFInfo
- Publication number
- CN108428225A CN108428225A CN201810089432.2A CN201810089432A CN108428225A CN 108428225 A CN108428225 A CN 108428225A CN 201810089432 A CN201810089432 A CN 201810089432A CN 108428225 A CN108428225 A CN 108428225A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- brain
- medical image
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 210000004556 brain Anatomy 0.000 title claims abstract description 43
- 230000004927 fusion Effects 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 34
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 22
- 239000003814 drug Substances 0.000 claims abstract description 22
- 238000003708 edge detection Methods 0.000 claims abstract description 17
- 238000012706 support-vector machine Methods 0.000 claims abstract description 10
- 238000002156 mixing Methods 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims abstract description 5
- 239000003086 colorant Substances 0.000 claims abstract description 4
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 4
- 230000000644 propagated effect Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 8
- 238000000605 extraction Methods 0.000 abstract 1
- 230000011218 segmentation Effects 0.000 description 5
- 238000003709 image segmentation Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000005192 partition Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 206010027476 Metastases Diseases 0.000 description 1
- 206010028980 Neoplasm Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 230000009401 metastasis Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000002603 single-photon emission computed tomography Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of brain image fusion identification methods of the image department based on multiple dimensioned multiple features comprising following steps：Original brain medical image is acquired as sample, gray processing processing is carried out to original brain medical image using weighted intensity algorithm, obtains gray level image；Gray level image is handled using histogram equalization, it is equalized rear gray-scale map, it adopts and edge detection is carried out to gray-scale map after equilibrium using improved Isotropic Sobel edge detection operators, obtain edge brain image, and obtain gray scale anatomic medicine image MRI and pseudo-colours functional medicine image, convolutional neural networks are initialized, training data extracts characteristic；Smoothed image and detail pictures are obtained using multi-resolution decomposition for the image after the training of medical image recognition training pattern, it is reconstructed to obtain blending image, the characteristic that convolutional neural networks model training extracts is passed in support vector machines by blending image input brain doctor and is trained, it will judge in the test feature data input medical image recognition training pattern of extraction, finally obtain accurate brain medical image recognition result.
Description
Technical field
The present invention relates to medical imaging section image processing fields, and in particular to a kind of image based on multiple dimensioned multiple features
Section's brain image fusion identification method.
Background technology
Currently, many famous scholars are devoted to the research of image segmentation blending algorithm, image partition method master both at home and abroad
There is following four：Segmentation based on threshold value, based on edge detection, based on region, based on energy.
Dividing method based on threshold value mainly has histogram concave surface analytic approach, maximum variance between clusters, threshold interpolation method etc.,
Such method is intuitively simple and efficient, but due to the complexity of image, the selection of threshold value becomes a major challenge of such method；Base
More classical algorithm has Sobel, Prewitt, Laplace and Canny operator etc. in the dividing method of edge detection, should
Class method due to the step evolution between noise spot and surrounding pixel point clearly, so being easily mistaken for edge；Based on region
Image partition method mainly have region-growing method and a split degree method, region-growing method is divided larger image and is susceptible to not
To cause over-segmentation, split degree method is easy to generate destruction to borderline region during continuous division in continuous cavity；
Image partition method based on energy includes the method based on level set, the method based on graph theory, the method etc. based on ICM.
Based on the method for level set since appearance, the hot spot for image segmentation field is become, in international top periodical and international conference
On there is a large amount of level set New image segmentation method to propose.The driving wheel of Caselles and Malladi et al. in image segmentation
Level Set Method is described in the context of wide model, but there are errors for segmentation result, and segmentation result is unstable.To understand
Certainly this problem, Osher et al. ensure the stability of level set using the method for reinitializing level set function.However,
The way reinitialized can not only cause serious problem, but also can influence the accuracy of number.Later, Li Chunming is proposed
A kind of new variation level set.In its Numerical Implementation, relatively large time step can be used in finite difference method
Length ensures enough numerical precisions to reduce iterations.Wang Xiaofeng proposes a kind of efficient, robust level set side
Multiple dimensioned segmentation thought is introduced regional area, with good efficiency and robustness by method.
Existing Medical Image Fusion is mainly to the medical image of two kinds of different modalities, not according to medical image mode
Together, Medical image fusion system can be divided into three types：Anatomic medicine image and anatomic medicine image co-registration, anatomic medicine figure
Picture and functional medicine image co-registration and functional medicine image and functional medicine image co-registration.MRI-PET and MRI-SPECT medicine
Image fusion system belongs to anatomic medicine image and functional medicine image co-registration, and the input picture of the system is that gray scale and puppet are color
Color.The MRI-PET combined integratedization external member that PHILIPS Co. releases is by commercial MRI image scannings instrument and with special shielding
PET combines, and obtained image is to the diagnosis of cancer metastasis and preoperative has important clinical value by stages.
But the processing segmentation for being directed to brain medical image at present leads to recognition capability because of its complexity and variability
Relatively low, good recognition performance is not played in inspection and treatment for doctor, and sometimes often some small variations fail
It observes and in time, lead to serious consequence.It is therefore desirable to provide a kind of accurate brain medical image recognition method.
Invention content
In view of the drawbacks described above of the prior art, that technical problem to be solved by the invention is to provide a kind of degrees of fusion is high,
Resolution greatly improves, the accuracy high image department based on multiple dimensioned multiple features brain image fusion identification method.
Technical solution of the present invention is as follows：A kind of image department based on multiple dimensioned multiple features brain image fusion recognition side
Method includes the following steps：
Original brain medical image is acquired as sample, original brain medical image is carried out using weighted intensity algorithm
Gray processing processing, obtains gray level image；Gray level image is handled using histogram equalization, is equalized rear gray-scale map,
Grey level histogram is fitted using high order spline function, the curve after the fitting has apparent valley point and peak point；
Grey level histogram divides valley section, smooth guidable matched curve is obtained on the basis of high order Spline-Fitting, and ask
The extreme point of matched curve can be led by taking this smoothly, be screened valley point according to the symbol on extreme point both sides；Using improved
Isotropic Sobel edge detection operators carry out edge detection to gray-scale map after equilibrium, obtain edge brain image, and obtain
Gray scale anatomic medicine image MRI and pseudo-colours functional medicine image are calculated being fitted improved Isotropic Sobel edge detections
Son is shown：
Binaryzation is carried out to edge image using adaptive thresholding algorithm, obtains the brain medicine figure after binaryzation
Picture；The medical image after binaryzation is handled using morphology operations operation, obtains brain medicine candidate region image, and
Candidate region image is formed into training data, initializes convolutional neural networks, the initialization convolutional neural networks are：Setting volume
Parameter in product neural network, including：The quantity of convolution kernel, the quantity of down-sampled layer, the size of convolution kernel, down-sampled layer
The range of decrease, initialize weight and the biasing of convolution kernel；Training data is assigned in batches in input convolutional neural networks, training number
According to convolutional layer, down-sampled layer, convolutional layer, down-sampled layer, multilayer perceptron is passed through respectively, propagated forward is completed；To Multilayer Perception
Device carries out error calculation and gradient calculates, and whether error in judgement restrains；If so, obtained error and gradient are passed with reversed
Algorithm is broadcast, is successively propagated by down-sampled layer, convolutional layer, down-sampled layer, convolutional layer, input layer, and successively updates network
Weight determines whether input layer, if then extracting characteristic；The characteristic that convolutional neural networks model training is extracted
It is trained according to passing in support vector machines, support vector machines is inputted by the training characteristics data of convolutional neural networks, together
When, with the optimization method of grid search come the parameter C and δ of Support Vector Machines Optimized, determines optimal supporting vector machine model, build
Vertical medical image recognition training pattern；Multi-resolution decomposition is used for the image after the training of medical image recognition training pattern
Smoothed image and detail pictures are obtained, smoothed image is merged to obtain smoothed image F using comentropyD, and for detail view
As D is merged using multiple features to obtain detail pictures FS, by smoothed image FDWith detail pictures FSIt is reconstructed to obtain fusion figure
Blending image is inputted in brain medical image recognition training pattern and is judged, finally obtained accurate medical image and melt by picture
Close recognition result.
Further, described to assign to training data in input convolutional neural networks in batches, training data passes through respectively
Convolutional layer, down-sampled layer, convolutional layer, down-sampled layer, multilayer perceptron are completed propagated forward, are specifically included：First from sample set
In take a collection of sample (X, YP), wherein X is the vector of sample number, and Y is the corresponding desired values of X, and P is 0 to 9 number, and X is defeated
Enter convolutional neural networks, calculates corresponding reality output OP, OP=Fn(...F2(F1(XPW(1))W(2))W(n)), n is convolution god
N-th layer through network, W indicate weights, and wherein convolution algorithm is to do convolution algorithm in upper layer network structure with convolution filter,
Then nonlinear transformation is carried out, and down-sampled operation is operated only with maximum pondization, i.e., maximum pond sampling is filtered by one
Device extracts the characteristic of upper layer network structure, and without nonlinear operation, each filtered maximum value is that data are down-sampled
A feature afterwards.
Further, the back-propagation algorithm is specially：By minimization error method backpropagation and adjust volume
Weight matrix in product neural network calculates activation value all in convolutional neural networks first to sample batch propagated forward；
Then, for every node layer, its residual error is calculated, residual error is derivation process from back to front；Then, the partial derivative of weights is calculated,
And update weighting parameter；Finally, repetition above method iterative convolution neural network parameter makes cost function converge to one minimum
Value, final solve obtain convolutional neural networks model.
Advantageous effect：A kind of degrees of fusion of the present invention is high, resolution greatly improves, accuracy is high based on multiple dimensioned multiple features
Image department brain image fusion identification method, this method has convolutional neural networks model and supporting vector machine model
Machine combines.In conjunction with the description sample data and expected data of the identification model of convolutional neural networks and support vector machines very depth
Correlation, brain medical image could be identified after needing fusion due to its particularity, complexity, accuracy are achieved significantly
It improves, there is no any relevant reports before merges realization identification, the separating capacity that the present invention classifies to figure pattern by the two
It is very strong.And convolutional neural networks model and supporting vector machine model their being to discriminate between property of target, so that the brain of generation is cured
The output for learning image identification system is more easy to optimize.Gray scale is carried out to original brain medical image by using weighted intensity algorithm
Change is handled, and obtains gray level image, such brain medical image is just by the image for being treated as easily identifying of image；It reuses straight
Side's figure equalization handles gray level image, is equalized rear gray-scale map, is examined using the edges improved Isotropic Sobel
Measuring and calculating carries out edge detection to gray-scale map after equilibrium, obtains edge image compared to original edge detection operator, makes its side
Edge detective operators normalize, and can effectively realize medical image positioning and identification function under complex background, SVM discrimination models row
In addition to the interference in pseudo- medical image region, being accurately positioned for medical image, the inspection of the edges improved Isotropic Sobel are realized
Measuring and calculating carries out edge detection, obtains edge image, the exactly indivisible use of steps above so that the present invention is easy
In realization, brain medical image recognition effect is fine, and improves the robustness of brain image method.
Description of the drawings
Fig. 1 is the brain image fusion recognition of the image department based on multiple dimensioned multiple features that the present invention provides preferred embodiment
Method flow diagram schematic diagram.
Specific implementation mode
The invention will be further described with reference to the accompanying drawings and examples：
As shown in Figure 1, a kind of image department based on multiple dimensioned multiple features brain image fusion identification method comprising with
Lower step：
Original brain medical image is acquired as sample, original brain medical image is carried out using weighted intensity algorithm
Gray processing processing, obtains gray level image；Gray level image is handled using histogram equalization, is equalized rear gray-scale map,
Grey level histogram is fitted using high order spline function, the curve after the fitting has apparent valley point and peak point；
Grey level histogram divides valley section, smooth guidable matched curve is obtained on the basis of high order Spline-Fitting, and ask
The extreme point of matched curve can be led by taking this smoothly, be screened valley point according to the symbol on extreme point both sides；Using improved
Isotropic Sobel edge detection operators carry out edge detection to gray-scale map after equilibrium, obtain edge brain image, and obtain
Gray scale anatomic medicine image MRI and pseudo-colours functional medicine image are calculated being fitted improved Isotropic Sobel edge detections
Son is shown：
Binaryzation is carried out to edge image using adaptive thresholding algorithm, obtains the brain medicine figure after binaryzation
Picture；The medical image after binaryzation is handled using morphology operations operation, obtains brain medicine candidate region image, this
It is also the big improvement of the present invention, is exactly based on the Isotropic Sobel edge detection operators after the improved normalization,
So that the edge image being calculated is apparent, recognition capability is stronger, and candidate region image is formed training data, initialization
Convolutional neural networks, the initialization convolutional neural networks are：Parameter in convolutional neural networks is set, including：Convolution
The range of decrease of the quantity of core, the quantity of down-sampled layer, the size of convolution kernel, down-sampled layer initializes weight and the biasing of convolution kernel；
Training data is assigned in batches in input convolutional neural networks, training data passes through convolutional layer, down-sampled layer, convolution respectively
Layer, down-sampled layer, multilayer perceptron complete propagated forward；Error calculation is carried out to multilayer perceptron and gradient calculates, and is judged
Whether error restrains；If so, by obtained error and gradient back-propagation algorithm, adopted by down-sampled layer, convolutional layer, drop
Sample layer, convolutional layer, input layer are successively propagated, and successively update the weight of network, input layer are determined whether, if then extracting
Go out characteristic；The characteristic that convolutional neural networks model training extracts is passed in support vector machines and is trained, warp
The training characteristics data input support vector machines of convolutional neural networks is crossed, meanwhile, optimize branch with the optimization method of grid search
The parameter C and δ for holding vector machine, determine optimal supporting vector machine model, establish medical image recognition training pattern；For passing through
Image after the training of medical image recognition training pattern obtains smoothed image and detail pictures, smoothed image using multi-resolution decomposition
It is merged to obtain smoothed image F using comentropyD, and detail pictures D is merged to obtain detail view using multiple features
As FS, by smoothed image FDWith detail pictures FSIt is reconstructed to obtain blending image, blending image input brain medical image is known
Judged in other training pattern, finally obtains accurate Medical image fusion recognition result.
Preferably, described to assign to training data in input convolutional neural networks in batches, training data is respectively through pulleying
Lamination, down-sampled layer, convolutional layer, down-sampled layer, multilayer perceptron are completed propagated forward, are specifically included：First from sample set
Take a collection of sample (X, YP), wherein X is the vector of sample number, and Y is the corresponding desired values of X, and P is 0 to 9 number, and X is inputted
Convolutional neural networks calculate corresponding reality output OP, OP=Fn(...F2(F1(XPW(1))W(2))W(n)), n is convolutional Neural
The n-th layer of network, W indicate weights, and wherein convolution algorithm is to do convolution algorithm in upper layer network structure with convolution filter, so
After carry out nonlinear transformation, and down-sampled operation is operated only with maximum pondization, i.e., maximum pond sampling is by a filter
The characteristic for extracting upper layer network structure, without nonlinear operation, each filtered maximum value is after data are down-sampled
A feature.
Preferably, the back-propagation algorithm is specially：By minimization error method backpropagation and adjust convolution
Weight matrix in neural network calculates activation value all in convolutional neural networks first to sample batch propagated forward；So
Afterwards, for every node layer, its residual error is calculated, residual error is derivation process from back to front；Then, the partial derivative of weights is calculated, and
Update weighting parameter；Finally, repeating above method iterative convolution neural network parameter makes cost function converge to a minimum,
Final solve obtains convolutional neural networks model.
The present invention is exactly to consider that the grey level histogram corresponding to medical image has multimodal paddy, it is contemplated that converting gradation histogram
Figure can be ignored at smoothed curve, the whereby not strong peak valley of part explicitly, and the value point corresponding to useful peak valley can
To decide.High order spline function thus can be used to be fitted grey level histogram, curve has after obtained fitting
Apparent valley point and peak point.The corresponding extreme point of matched curve is obtained, by the judgement to the extreme point both sides symbol
Valley point is screened.When carrying out interval division, valley is first handled, specific practice according to slope before and after judgement, will be lacked as follows
The valley point removal of number slope unobvious, whereby so that interval division not only have it is higher intelligent, and with higher
Adaptivity.
The preferred embodiment of the present invention has been described in detail above.It should be appreciated that those skilled in the art without
It needs creative work according to the present invention can conceive and makes many modifications and variations.Therefore, all technologies in the art
Personnel are available by logical analysis, reasoning, or a limited experiment on the basis of existing technology under this invention's idea
Technical solution, all should be in the protection domain being defined in the patent claims.
Claims (3)
1. a kind of brain Medical image fusion recognition methods of image department based on multiple dimensioned multiple features, which is characterized in that including
Following steps：
Original brain medical image is acquired as sample, gray scale is carried out to original brain medical image using weighted intensity algorithm
Change is handled, and obtains gray level image；Gray level image is handled using histogram equalization, is equalized rear gray-scale map, is used
High order spline function is fitted grey level histogram, and the curve after the fitting has apparent valley point and peak point；Gray scale
Histogram divides valley section, smooth guidable matched curve is obtained on the basis of high order Spline-Fitting, and seek this
The extreme point that matched curve can smoothly be led, screens valley point according to the symbol on extreme point both sides；Using improved
Isotropic Sobel edge detection operators carry out edge detection to gray-scale map after equilibrium, obtain edge brain image, and obtain
Gray scale anatomic medicine image MRI and pseudo-colours functional medicine image are calculated being fitted improved Isotropic Sobel edge detections
Son is shown：
Using adaptive
It answers thresholding algorithm to carry out binaryzation to edge image, obtains the brain medical image after binaryzation；It is operated using morphology operations
Medical image after binaryzation is handled, obtains brain medicine candidate region image, and candidate region image is formed and is instructed
Practice data, initialize convolutional neural networks, the initialization convolutional neural networks are：Parameter in convolutional neural networks is set,
Including：The range of decrease of the quantity of convolution kernel, the quantity of down-sampled layer, the size of convolution kernel, down-sampled layer initializes convolution kernel
Weight and biasing；Training data is assigned in batches in input convolutional neural networks, training data passes through convolutional layer, drop respectively
Sample level, convolutional layer, down-sampled layer, multilayer perceptron complete propagated forward；Error calculation and gradient are carried out to multilayer perceptron
It calculates, and whether error in judgement restrains；If so, by obtained error and gradient back-propagation algorithm, by down-sampled layer,
Convolutional layer, down-sampled layer, convolutional layer, input layer are successively propagated, and successively update the weight of network, determine whether to input
Layer, if then extracting characteristic；The characteristic that convolutional neural networks model training extracts is passed in support vector machines
It is trained, the training characteristics data by convolutional neural networks is inputted support vector machines, meanwhile, with the optimization of grid search
Method carrys out the parameter C and δ of Support Vector Machines Optimized, determines optimal supporting vector machine model, establishes medical image recognition training
Model；Smoothed image and details are obtained using multi-resolution decomposition for the image after the training of medical image recognition training pattern
Image, smoothed image are merged to obtain smoothed image F using comentropyD, and detail pictures D is melted using multiple features
Conjunction obtains detail pictures FS, by smoothed image FDWith detail pictures FSIt is reconstructed to obtain blending image, blending image is inputted into brain
Judged in portion's medical image recognition training pattern, finally obtains accurate Medical image fusion recognition result.
2. the image department according to claim 1 based on multiple dimensioned multiple features brain Medical image fusion recognition methods,
It is characterized in that, it is described by training data in batches assign to input convolutional neural networks in, training data respectively pass through convolutional layer,
Down-sampled layer, convolutional layer, down-sampled layer, multilayer perceptron are completed propagated forward, are specifically included：One is taken from sample set first
Lot sample sheet (X, YP), wherein X is the vector of sample number, and Y is the corresponding desired values of X, and P is 0 to 9 number, and X is inputted convolution
Neural network calculates corresponding reality output OP, OP=Fn(...F2(F1(XPW(1))W(2))W(n)), n is convolutional neural networks
N-th layer, W indicate weights, wherein convolution algorithm is that convolution algorithm is done in upper layer network structure with convolution filter, then into
Row nonlinear transformation, and down-sampled operation is operated only with maximum pondization, i.e., maximum pond sampling is extracted by a filter
The characteristic of upper layer network structure, without nonlinear operation, each filtered maximum value is one after data are down-sampled
A feature.
3. the image department according to claim 1 based on multiple dimensioned multiple features brain Medical image fusion recognition methods,
It is characterized in that, the back-propagation algorithm is specially：By minimization error method backpropagation and adjust convolutional Neural
Weight matrix in network calculates activation value all in convolutional neural networks first to sample batch propagated forward；Then,
For every node layer, its residual error is calculated, residual error is derivation process from back to front；Then, the partial derivative of weights is calculated, and is updated
Weighting parameter；Finally, repeating above method iterative convolution neural network parameter makes cost function converge to a minimum, finally
Solution obtains convolutional neural networks model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810089432.2A CN108428225A (en) | 2018-01-30 | 2018-01-30 | Image department brain image fusion identification method based on multiple dimensioned multiple features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810089432.2A CN108428225A (en) | 2018-01-30 | 2018-01-30 | Image department brain image fusion identification method based on multiple dimensioned multiple features |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108428225A true CN108428225A (en) | 2018-08-21 |
Family
ID=63156161
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810089432.2A Pending CN108428225A (en) | 2018-01-30 | 2018-01-30 | Image department brain image fusion identification method based on multiple dimensioned multiple features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108428225A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111882518A (en) * | 2020-06-09 | 2020-11-03 | 中海石油(中国)有限公司 | Magnetic leakage data self-adaptive pseudo-colorization method |
| CN112150564A (en) * | 2020-08-21 | 2020-12-29 | 哈尔滨理工大学 | Medical image fusion algorithm based on deep convolutional neural network |
| CN112837807A (en) * | 2021-02-05 | 2021-05-25 | 广州中医药大学第一附属医院 | An early intelligent and high-precision auxiliary diagnosis method for brain aging and cognitive impairment in T2DM |
| CN113034391A (en) * | 2021-03-19 | 2021-06-25 | 西安电子科技大学 | Multi-mode fusion underwater image enhancement method, system and application |
| CN113112511A (en) * | 2021-04-19 | 2021-07-13 | 新东方教育科技集团有限公司 | Method and device for correcting test paper, storage medium and electronic equipment |
| CN113780421A (en) * | 2021-06-07 | 2021-12-10 | 广州天鹏计算机科技有限公司 | Brain PET image identification method based on artificial intelligence |
| CN116310358A (en) * | 2023-02-23 | 2023-06-23 | 哈尔滨市科佳通用机电股份有限公司 | Method, storage medium and equipment for detecting bolt loss of railway wagon |
| TWI817884B (en) * | 2023-01-03 | 2023-10-01 | 國立中央大學 | Image detection system and operation method thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105320961A (en) * | 2015-10-16 | 2016-02-10 | 重庆邮电大学 | Handwriting numeral recognition method based on convolutional neural network and support vector machine |
| CN105447837A (en) * | 2016-02-04 | 2016-03-30 | 重庆邮电大学 | Multi-mode brain image fusion method based on adaptive cloud model |
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN107194912A (en) * | 2017-04-20 | 2017-09-22 | 中北大学 | The brain CT/MR image interfusion methods of improvement coupling dictionary learning based on rarefaction representation |
-
2018
- 2018-01-30 CN CN201810089432.2A patent/CN108428225A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105320961A (en) * | 2015-10-16 | 2016-02-10 | 重庆邮电大学 | Handwriting numeral recognition method based on convolutional neural network and support vector machine |
| CN105447837A (en) * | 2016-02-04 | 2016-03-30 | 重庆邮电大学 | Multi-mode brain image fusion method based on adaptive cloud model |
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN107194912A (en) * | 2017-04-20 | 2017-09-22 | 中北大学 | The brain CT/MR image interfusion methods of improvement coupling dictionary learning based on rarefaction representation |
Non-Patent Citations (4)
| Title |
|---|
| 刘同焰: "车牌识别系统的相关算法研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 刘欣 等: "一种基于多尺度最大信息熵和梯度的图像融合算法", 《微计算机应用》 * |
| 程琳 等: "一种基于小波熵的多尺度变换医学图像融合方法", 《中国体视学学会CT理论与应用分会、中国工程物理研究院应用电子学研究所、《CT理论与应用研究》编辑部论文集》 * |
| 赵佳: "基于自适应云模型的多模态医学图像融合方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111882518A (en) * | 2020-06-09 | 2020-11-03 | 中海石油(中国)有限公司 | Magnetic leakage data self-adaptive pseudo-colorization method |
| CN111882518B (en) * | 2020-06-09 | 2023-12-19 | 中海石油(中国)有限公司 | Self-adaptive pseudo-colorization method for magnetic flux leakage data |
| CN112150564A (en) * | 2020-08-21 | 2020-12-29 | 哈尔滨理工大学 | Medical image fusion algorithm based on deep convolutional neural network |
| CN112150564B (en) * | 2020-08-21 | 2024-09-27 | 广东科技学院 | Medical image fusion algorithm based on deep convolution neural network |
| CN112837807B (en) * | 2021-02-05 | 2024-01-23 | 广州中医药大学第一附属医院 | Early intelligent high-precision auxiliary diagnosis method for T2DM brain aging cognitive impairment |
| CN112837807A (en) * | 2021-02-05 | 2021-05-25 | 广州中医药大学第一附属医院 | An early intelligent and high-precision auxiliary diagnosis method for brain aging and cognitive impairment in T2DM |
| CN113034391A (en) * | 2021-03-19 | 2021-06-25 | 西安电子科技大学 | Multi-mode fusion underwater image enhancement method, system and application |
| CN113034391B (en) * | 2021-03-19 | 2023-08-08 | 西安电子科技大学 | Multi-mode fusion underwater image enhancement method, system and application |
| CN113112511A (en) * | 2021-04-19 | 2021-07-13 | 新东方教育科技集团有限公司 | Method and device for correcting test paper, storage medium and electronic equipment |
| CN113112511B (en) * | 2021-04-19 | 2024-01-05 | 新东方教育科技集团有限公司 | Method and device for correcting test paper, storage medium and electronic equipment |
| CN113780421A (en) * | 2021-06-07 | 2021-12-10 | 广州天鹏计算机科技有限公司 | Brain PET image identification method based on artificial intelligence |
| TWI817884B (en) * | 2023-01-03 | 2023-10-01 | 國立中央大學 | Image detection system and operation method thereof |
| CN116310358A (en) * | 2023-02-23 | 2023-06-23 | 哈尔滨市科佳通用机电股份有限公司 | Method, storage medium and equipment for detecting bolt loss of railway wagon |
| CN116310358B (en) * | 2023-02-23 | 2023-12-15 | 哈尔滨市科佳通用机电股份有限公司 | Method, storage medium and equipment for detecting bolt loss of railway wagon |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108428225A (en) | Image department brain image fusion identification method based on multiple dimensioned multiple features | |
| Al-Bander et al. | Multiscale sequential convolutional neural networks for simultaneous detection of fovea and optic disc | |
| CN108389187A (en) | Image department image-recognizing method based on convolutional neural networks method and support vector machines method | |
| Soomro et al. | Impact of image enhancement technique on CNN model for retinal blood vessels segmentation | |
| Melinscak et al. | Retinal Vessel Segmentation using Deep Neural Networks. | |
| CN106651846B (en) | Segmentation method of retinal blood vessel images | |
| Wang et al. | Analysis of retinal vasculature using a multiresolution hermite model | |
| CN108198184A (en) | The method and system of contrastographic picture medium vessels segmentation | |
| CN108416360A (en) | Cancer diagnosis system and method based on breast molybdenum target calcification feature | |
| Basu et al. | Segmentation and tracing of single neurons from 3D confocal microscope images | |
| CN110517273B (en) | Cytology image segmentation method based on dynamic gradient threshold | |
| CN104794708A (en) | Atherosclerosis plaque composition dividing method based on multi-feature learning | |
| Aurangzeb et al. | An efficient and light weight deep learning model for accurate retinal vessels segmentation | |
| CN109635811A (en) | Image Analysis Methods of Space Plants | |
| CN102737379A (en) | A CT Image Segmentation Method Based on Adaptive Learning | |
| CN107909588A (en) | Partition system under MRI cortex based on three-dimensional full convolutional neural networks | |
| CN108460337A (en) | Dense fog scene aerial image fusion identification method based on adaptive cloud model | |
| CN106355584A (en) | Automatic detection method for microaneurysm in eye fundus image on basis of local entropy determining threshold | |
| CN104616289A (en) | Removal method and system for bone tissue in 3D CT (Three Dimensional Computed Tomography) image | |
| CN104463215A (en) | Tiny aneurysm occurrence risk prediction system based on retina image processing | |
| CN113870194A (en) | Deep layer characteristic and superficial layer LBP characteristic fused breast tumor ultrasonic image processing device | |
| CN108876776A (en) | A kind of method of generating classification model, eye fundus image classification method and device | |
| Sankari et al. | Automatic tumor segmentation using convolutional neural networks | |
| CN111161295A (en) | Background stripping method for dish image | |
| Maksoud et al. | MRI brain tumor segmentation system based on hybrid clustering techniques |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180821 |