CN106846380A - A kind of medical image registration method and equipment - Google Patents
A kind of medical image registration method and equipment Download PDFInfo
- Publication number
- CN106846380A CN106846380A CN201611220617.XA CN201611220617A CN106846380A CN 106846380 A CN106846380 A CN 106846380A CN 201611220617 A CN201611220617 A CN 201611220617A CN 106846380 A CN106846380 A CN 106846380A
- Authority
- CN
- China
- Prior art keywords
- block
- sub
- template
- training
- test
- 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
Classifications
-
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The embodiment of the present invention provides a kind of medical image registration method and equipment.Methods described includes：Test image is obtained, multiple test sub-blocks are extracted according to the first preset rules in test image；Obtain the minimum spanning tree built according to the second preset rules by template sub-block and training sub-block；First degree of correlation of test sub-block and training sub-block is calculated, and selects optimal Deformation Field corresponding with test first degree of correlation of sub-block highest training sub-block；Using optimal Deformation Field as the sparse Deformation Field of test image, and dense Deformation Field is built according to sparse Deformation Field；According to dense Deformation Field and test image to template image registration；Wherein, optimal Deformation Field is obtained from training sub-block to template sub-block registration.Minimum spanning tree and dense Deformation Field is obtained according to test sub-block that the embodiment of the present invention is built by template sub-block and training sub-block, according to dense Deformation Field and test image to template image registration, while the accuracy of registration is improve, the on-line registration time is shortened.
Description
Technical field
The present embodiments relate to technical field of image processing, more particularly to a kind of medical image registration method and equipment.
Background technology
The non-rigid registration algorithm of fast accurate is the important need of medical image analysis.By the image of patient's preoperative and postoperative
Or by the purpose of the image registration of Different Individual be to find to make the maximum optimal Deformation Field of similarity between image, make figure
The corresponding pixel reflection identical anatomic tissue information as in.
In the prior art, by calculating the optimal Deformation Field of test image, the optimal Deformation Field that will be obtained acts on test
Image, obtains intermediate die plate image (intermediate templates, abbreviation IT image), and by IT images to template image
Registration.The generating mode of wherein intermediate die plate image has intermediate die plate to generate (intermediate template
Generation, abbreviation ITG) and intermediate die plate selection (intermediate template selection, abbreviation ITS) two
Class.The purpose of ITG methods is to build new IT images using current training image collection.Conventional ITG methods, by training
The solution of corresponding relation, finds the Deformation Field of test image correspondence position between collection image and template image, will using the Deformation Field
Test image is mapped so as to solve IT images, then recycles conventional registration Algorithm that IT images are directly carried out with template image
Registration, the test image after finally being matched.Different from ITG methods, the iTREE algorithms in ITS methods are to utilize to count mould
Type generates more images on the basis of training set image, image and the training image that will be generated as IT image sets, according to
Difference between IT images sets up minimum spanning tree, and all IT images are modeled using tree structure, wherein the root section set
Point is template image, and then all IT images, to template image registration, registration are tried to achieve along the side in tree structure most
Excellent Deformation Field is preserved, and in test phase, will calculate the difference between test image and all IT images, is found and test image phase
Like the IT image BMIT for spending maximum, and will be connected between them, the minimum spanning tree that renewal is established will finally be asked the training stage
The BMIT for obtaining, as the initial deformation of test image, test image is matched somebody with somebody to template image to the Deformation Field of template image registration
It is accurate.
The problem that can be seen that the registration Algorithm facing based on IT images from ITG and ITS methods is how to build
Abundant training set sample, realization is accurately generated and selected to IT images.Only establish the training set of data rich, ability
Initial deformation is accurately predicted using training pattern, the otherwise estimation to initial deformation there will be larger mistake
Difference.And generally, abundant training image can not be obtained, make exist largely with the less training figure of template image difference
Picture, so that image difference causes greatly the precision of image registration and the problem that speed is low.
The content of the invention
For the problem that prior art is present, the embodiment of the present invention provides a kind of medical image registration method and equipment.
The embodiment of the present invention provides a kind of medical image registration method, including：
Test image is obtained, multiple test sub-blocks are extracted according to the first preset rules in the test image；
Multiple minimum spanning trees are obtained, the minimum spanning tree is by template sub-block and training sub-block according to the second preset rules
Build, wherein, the template sub-block is extracted by template image and obtained, and the training sub-block is extracted by training image
Obtain, and the template sub-block, the training sub-block are identical with the test sub-block size；
First degree of correlation of the test sub-block and the training sub-block is calculated, and is selected and the test phase of sub-block first
The corresponding optimal Deformation Field of sub-block is trained described in the degree highest of pass；
Using the optimal Deformation Field as the sparse Deformation Field of the test image, and built according to the sparse Deformation Field
Dense Deformation Field；
According to the dense Deformation Field and the test image to template image registration；Wherein, the optimal Deformation Field by
Training sub-block is obtained to template sub-block registration.
The embodiment of the present invention provides a kind of medical figure registration equipment, including：
Extraction module, for obtaining test image, multiple survey is extracted in the test image according to the first preset rules
Swab block；
First builds module, and for obtaining multiple minimum spanning trees, the minimum spanning tree is by template sub-block and training
Root tuber builds according to the second preset rules, wherein, the template sub-block is extracted by template image and obtained, the training sub-block
Extracted by training image and obtained, and the template sub-block, the training sub-block are identical with the test sub-block size；
Computing module, for calculate it is described test sub-block and it is described training sub-block first degree of correlation, and select with it is described
The corresponding optimal Deformation Field of training sub-block described in test sub-block first degree of correlation highest；
Second build module, for using the optimal Deformation Field as the test image sparse Deformation Field, and according to
The sparse Deformation Field builds dense Deformation Field；
Registration module, for registering to template image according to the dense Deformation Field and the test image；Wherein, it is described
Optimal Deformation Field is obtained from training sub-block to template sub-block registration.
A kind of medical image registration method provided in an embodiment of the present invention and equipment, by extracting template sub-block and training
Block, minimum spanning tree is built according to template sub-block and training sub-block, and sub-block is trained in calculating test sub-block and minimum spanning tree
First degree of correlation simultaneously obtains sparse Deformation Field, dense Deformation Field is obtained according to sparse Deformation Field, according to dense Deformation Field and test
Image while the accuracy of registration is improve, shortens the on-line registration time to template image registration.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
The accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, drawings in the following description are this hairs
Some bright embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, can be with root
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is a kind of medical image registration method schematic flow sheet provided in an embodiment of the present invention；
A kind of medical image registration method schematic flow sheet that Fig. 2 is provided for another embodiment of the present invention；
Fig. 3 is a kind of medical figure registration device structure schematic diagram provided in an embodiment of the present invention.
Specific embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
A part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art
The every other embodiment obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
Fig. 1 is a kind of medical image registration method schematic flow sheet provided in an embodiment of the present invention, as shown in figure 1, described
Method, including：
Step 101：Test image is obtained, multiple test are extracted according to the first preset rules in the test image
Block；
Specifically, test image is obtained, and extracts multiple test sub-blocks in test image according to the first preset rules, its
In, the size for testing sub-block can be set according to actual conditions.
Step 102：Multiple minimum spanning trees are obtained, the minimum spanning tree is by template sub-block and training sub-block according to second
Preset rules build, wherein, the template sub-block is extracted by template image and obtained, and the training sub-block is by training
What image zooming-out was obtained, and the template sub-block, the training sub-block are identical with the test sub-block size；
Specifically, template sub-block is extracted on the template image for getting, and training sub-block is extracted on training image,
Template sub-block, training sub-block are identical with above-mentioned test sub-block size, using template sub-block and training sub-block according to the second default rule
Then build minimum spanning tree, it is to be understood that have multiple template sub-block on template image, the multiple instructions of each template sub-block correspondence
Practice sub-block, and each template sub-block one minimum spanning tree of correspondence.
Step 103：First degree of correlation of the test sub-block and the training sub-block is calculated, and is selected and test
The corresponding optimal Deformation Field of sub-block is trained described in block first degree of correlation highest；
Specifically, first degree of correlation of test sub-block and all training sub-blocks is calculated using the method for normalized mutual information,
And select the corresponding optimal Deformation Field of the first degree of correlation highest training sub-block.It should be noted that optimal Deformation Field is minimum
Train what sub-block was obtained to template sub-block registration in spanning tree, and be an optimal transform matrix, by the optimal shape for getting
Variable field, line translation is entered to test image, can reduce the difference and between template image, improves the efficiency of registration.
Step 104：Using the optimal Deformation Field as the test image sparse Deformation Field, and according to the sparse shape
Variable field builds dense Deformation Field；
Specifically, each test sub-block one optimal Deformation Field of correspondence in test image, therefore in test image
All corresponding multiple optimal Deformation Fields of sub-block of testing constitute sparse Deformation Field, because test sub-block is not covered with whole test
Image, so the region having in test image does not have corresponding optimal Deformation Field, therefore referred to as sparse Deformation Field.According to interpolation
Method, calculates the corresponding optimal Deformation Field in region for not having tested sub-block to cover in test image, so as to constitute dense shape
Variable field.
Step 105：According to the dense Deformation Field and the test image to template image registration；Wherein, it is described optimal
Deformation Field is obtained from training sub-block to template sub-block registration.
Specifically, test image is entered into line translation according to dense Deformation Field, the intermediate die plate image after being converted, and will
Intermediate die plate image is to template image registration.
The embodiment of the present invention is built minimum by extracting template sub-block and training sub-block according to template sub-block and training sub-block
Spanning tree, calculates first degree of correlation of training sub-block in test sub-block and minimum spanning tree and obtains sparse Deformation Field, according to dilute
Dredge Deformation Field and obtain dense Deformation Field, according to dense Deformation Field and test image to template image registration, improve registration
While accuracy, the on-line registration time is shortened.
On the basis of above-described embodiment, the minimum spanning tree is by template sub-block and training sub-block according to the second default rule
Then build, including：
Template image is obtained, the corresponding curvature of pixel is obtained in the template image more than first threshold as mould
Plate key point, centered on the template key point, the template sub-block that acquisition is constituted with presetted pixel number；
Training image is obtained, by the training image to template image registration, is obtained and the mould on the template image
The corresponding training key point of plate key point, obtains the training sub-block structure of predetermined number centered on the training key point
Into training sub-block group；
Weighted average sub-block is calculated according to the training sub-block group, the template sub-block and the weighted average sub-block is calculated
Second degree of correlation, according to second degree of correlation build minimum spanning tree.
Specifically, template image is obtained, template image can be the medical image of normal, health people, also may be used
To be the medical image of typical patient, template image can be selected according to actual conditions, and the present embodiment is not especially limited.
Template image is made up of multiple pixels, obtains on template image pixel corresponding curvature more than first threshold as template
Key point, and template key point is multiple, centered on template key point, obtains the template sub-block that presetted pixel number is constituted；
Multiple training images are obtained, the training image that will be got is to template image registration, and extraction is closed with template on training image
The corresponding training key point of key point, and centered on training key point obtain predetermined number with template sub-block size identical
Training sub-block, the training sub-block composing training sub-block group of predetermined number；And training sub-block group is entered using non-area average algorithm
Row weighted average calculation, obtains weighted average sub-block, calculates second degree of correlation of weighted average sub-block and template sub-block, according to the
Two degrees of correlation build the minimum spanning tree of template sub-block and training sub-block.It is understood that the root node of minimum spanning tree is
Template sub-block, cotyledon node is to train sub-block, and template sub-block one minimum spanning tree of correspondence.It should be noted that instruction
The weighted average calculation method for practicing sub-block group is：Such as training sub-block A, training sub-block B, training sub-block C constitute training sub-block
Group, and the corresponding weights of training sub-block A are ω 1, the corresponding weights of training sub-block B are ω 2, and the corresponding weights of training sub-block C are
ω 3, the then weighted average sub-block being calculated is ω 1 × A+ ω 2 × B+, 3 × C of ω, and ω 1+ ω 2+ ω 3=1.
The embodiment of the present invention is built most by obtaining template sub-block and training sub-block according to template sub-block and training sub-block
Small spanning tree, sub-block scale is down to by registering yardstick from general image yardstick, so as to improve the accuracy of registration.
It is described to extract test according to the first preset rules in the test image on the basis of above-described embodiment
Block, including：
Using the corresponding curvature of pixel in the test image more than Second Threshold as test key point, with the survey
Centered on examination key point, the test sub-block is obtained.
Specifically, in image registration, the pixel correspondence curvature that can be chosen on image outline is more than Second Threshold
Conduct test key point, centered on the test key point, obtain the test sub-block that is constituted of pixel of predetermined number, example
Such as：Choose to test centered on key point, predetermined number is the test sub-block that 3 pixels of pixel constitute 3 × 3.Can manage
Solution, the test sub-block is identical with template sub-block and training sub-block size.
The embodiment of the present invention is used as test by selecting the corresponding curvature of pixel in test image more than Second Threshold
Key point, and obtained according to test key point and template sub-block size identical test sub-block, selecting on image, there is diagnosis to anticipate
The point of justice so that higher with punctual accuracy.
On the basis of above-described embodiment, first degree of correlation for calculating the test sub-block and the training sub-block,
Including：
First degree of correlation of the test sub-block and the training sub-block is calculated using normalized mutual information algorithm.
Specifically, using the algorithm of normalized mutual information, calculate and test all training sub-blocks in sub-block and training image
First degree of correlation, wherein first degree of correlation is similarity between characterization test sub-block and training sub-block, first degree of correlation is bigger, says
Bright test sub-block is more similar to training sub-block.
The embodiment of the present invention is built minimum by extracting template sub-block and training sub-block according to template sub-block and training sub-block
Spanning tree, calculates first degree of correlation of training sub-block in test sub-block and minimum spanning tree and obtains sparse Deformation Field, according to dilute
Dredge Deformation Field and obtain dense Deformation Field, according to dense Deformation Field and test image to template image registration, improve registration
While accuracy, the on-line registration time is shortened.
It is described that dense Deformation Field is built according to the sparse Deformation Field on the basis of above-described embodiment, including：
According to the sparse Deformation Field, dense Deformation Field is built using the method for interpolation.
Specifically, it is made up of selection optimal Deformation Field corresponding with the training sub-block of test sub-block second degree of correlation maximum
Be sparse Deformation Field, this be due to test sub-block be not covered with all of test image, therefore referred to as sparse Deformation Field, utilize
The method of interpolation, obtains the corresponding optimal Deformation Field in region for not having tested sub-block to cover in test image, so as to obtain thick
Close Deformation Field.
The embodiment of the present invention builds dense Deformation Field by the method for interpolation so that every piece of region in test image has
Corresponding optimal Deformation Field, such that it is able to realize accurately registration.
A kind of medical image registration method schematic flow sheet that Fig. 2 is provided for another embodiment of the present invention, as shown in Fig. 2
The specific implementation step of methods described is as follows：
Step 201：Template image；Template image is obtained, wherein template image can be the people institute according to normal health
The medical image, or the medical image of typical patient for collecting, carry out step 202；
Step 202：Extract key point；Conduct of the corresponding curvature of pixel more than first threshold is extracted on template image
Template key point, carries out step 203；
Step 203：Extract sub-block；Centered on the template key point extracted, what acquisition was constituted with presetted pixel number
Template sub-block, carries out step 204；
Step 204：Training image；Multiple training images are obtained, step 205 is carried out；
Step 205：Extract key point；By training image to template image registration, obtain and closed with the template on template image
The corresponding training key point of key point, carries out step 206；
Step 206：Extract sub-block group；Centered on training key point, the training sub-block of predetermined number, composing training are obtained
Sub-block group, wherein training sub-block is identical with template sub-block size, carries out step 207；
Step 207：Calculate weighted average sub-block；Each training sub-block of composition training sub-block group occupies different power
Weight, the weight according to each training sub-block and each training sub-block calculates weighted average sub-block using non-area average algorithm,
Carry out step 208；
Step 208：Calculate second degree of correlation；Second degree of correlation between calculation template sub-block and weighted average sub-block, with
And the third phase pass degree between weighted average sub-block；The computational methods of wherein second degree of correlation and third phase pass degree are to use normalizing
Change mutual information, second degree of correlation is bigger, shows that template sub-block is more similar to weighted average sub-block, and third phase pass degree is bigger, shows
It is more similar between weighted average sub-block, carry out step 209；
Step 209：Build minimum spanning tree；The minimum generation of template sub-block and training sub-block is built according to second degree of correlation
Tree, wherein, it is the root of each tree that each template sub-block corresponds to a minimum spanning tree, and the template sub-block from template image
Node, the training sub-block group from training image constitutes the cotyledon node of minimum spanning tree, carries out step 210；
Step 210：Sub-block registration；According to the minimum spanning tree set up, the training sub-block in cotyledon node is along minimum
The side of spanning tree, so as to obtain each corresponding optimal Deformation Field of training sub-block, is entered to the template sub-block registration in root node
Row step 211；
Step 211：Test image；Test image subject to registration is obtained, step 212 is carried out；
Step 212：Extract key point；Conduct of the corresponding curvature of pixel more than Second Threshold is extracted in test image
Test key point, carries out step 213；
Step 213：Extract sub-block；Centered on the test key point extracted, obtain and template sub-block size identical
Test sub-block, carries out step 214；
Step 214：Calculate first degree of correlation；First degree of correlation of test sub-block and training sub-block is calculated, wherein the first phase
The computational methods of Guan Du are consistent with second degree of correlation, and here is omitted, carry out step 215；
Step 215：Obtain the maximum training sub-block of first degree of correlation；In multiple first degrees of correlation for calculating, obtain
Training sub-block in the maximum corresponding minimum spanning tree of first degree of correlation, carries out step 216；
Step 216：Sparse Deformation Field；By step 210, each training sub-block is corresponding in can obtaining minimum spanning tree
Optimal Deformation Field, each test sub-block can obtain an optimal Deformation Field in test image, these optimal Deformation Field structures
Into the sparse Deformation Field of test image, step 217 is carried out；
Step 217：Dense Deformation Field；Using the method for interpolation, dense Deformation Field is obtained according to sparse Deformation Field, walked
Rapid 218；
Step 218：Registration；Test image is entered into line translation according to dense Deformation Field, intermediate die plate image is obtained, then by
Between template image to template image registration.
The embodiment of the present invention is built minimum by extracting template sub-block and training sub-block according to template sub-block and training sub-block
Spanning tree, calculates first degree of correlation of training sub-block in test sub-block and minimum spanning tree and obtains sparse Deformation Field, according to dilute
Dredge Deformation Field and obtain dense Deformation Field, according to dense Deformation Field and test image to template image registration, improve registration
While accuracy, the on-line registration time is shortened.
Fig. 3 is a kind of medical figure registration device structure schematic diagram provided in an embodiment of the present invention, as shown in figure 3, described
Equipment, including extraction module 301, first builds module 302, computing module 303, second and builds module 304 and registration module
305, wherein：
Extraction module 301 is used to obtain test image, extracts multiple according to the first preset rules in the test image
Test sub-block；First structure module 302 is used to obtain multiple minimum spanning trees, and the minimum spanning tree is by template sub-block and training
Sub-block builds according to the second preset rules, wherein, the template sub-block is extracted by template image and obtained, training
Block is extracted by training image and obtained, and the template sub-block, the training sub-block are identical with the test sub-block size；
Computing module 303 is used to calculate first degree of correlation of the test sub-block and the training sub-block, and selects and test
The corresponding optimal Deformation Field of sub-block is trained described in block first degree of correlation highest；Second structure module 304 is used for will be described optimal
Deformation Field and builds dense Deformation Field as the sparse Deformation Field of the test image according to the sparse Deformation Field；With quasi-mode
Block 305 is used for according to the dense Deformation Field and the test image to template image registration；Wherein, the optimal Deformation Field by
Training sub-block is obtained to template sub-block registration.
Specifically, extraction module 301 obtains test image, and extracts multiple in test image according to the first preset rules
Test sub-block, wherein, the size for testing sub-block can be set according to actual conditions.First structure module 302 is getting
Template image on extract template sub-block, and on training image extract training sub-block, template sub-block, training sub-block with it is above-mentioned
Test sub-block size is identical, and minimum spanning tree, Ke Yili are built according to the second preset rules using template sub-block and training sub-block
Solution, there is multiple template sub-block on template image, the multiple training sub-blocks of each template sub-block correspondence, and each template sub-block pair
Answer a minimum spanning tree.Computing module 303 calculates test sub-block and all training sub-blocks using the method for normalized mutual information
First degree of correlation, and select the first degree of correlation highest training corresponding optimal Deformation Field of sub-block.It should be noted that optimal
Deformation Field is to train in minimum spanning tree sub-block to be obtained to template sub-block registration, and is an optimal transform matrix.Test chart
Each test sub-block one optimal Deformation Field of correspondence as in, therefore, second builds module 304 according to the institute in test image
There are the corresponding multiple optimal Deformation Fields of test sub-block to constitute sparse Deformation Field, because test sub-block is not covered with whole test chart
Picture, so the region having in test image does not have corresponding optimal Deformation Field, therefore referred to as sparse Deformation Field.According to the side of interpolation
Method, calculates the corresponding optimal Deformation Field in region for not having tested sub-block to cover in test image, so as to constitute dense deformation
.Test image is entered line translation by registration module 305 according to dense Deformation Field, the intermediate die plate image after being converted, and will
Intermediate die plate image is to template image registration.
Medical figure registration equipment provided in an embodiment of the present invention be used for perform the above method, its specific embodiment with
The implementation method of method is consistent, and here is omitted.
The embodiment of the present invention is built minimum by extracting template sub-block and training sub-block according to template sub-block and training sub-block
Spanning tree, calculates first degree of correlation of training sub-block in test sub-block and minimum spanning tree and obtains sparse Deformation Field, according to dilute
Dredge Deformation Field and obtain dense Deformation Field, according to dense Deformation Field and test image to template image registration, improve registration
While accuracy, the on-line registration time is shortened.
On the basis of above-described embodiment, it is described first build module specifically for：
Template image is obtained, the corresponding curvature of pixel is obtained in the template image more than first threshold as mould
Plate key point, centered on the template key point, the template sub-block that acquisition is constituted with presetted pixel number；
Training image is obtained, by the training image to template image registration, is obtained and the mould on the template image
The corresponding training key point of plate key point, obtains the training sub-block structure of predetermined number centered on the training key point
Into training sub-block group；
Weighted average sub-block is calculated according to the training sub-block group, the template sub-block and the weighted average sub-block is calculated
Second degree of correlation, minimum spanning tree is built according to second degree of correlation, wherein the template sub-block is the minimum generation
The root node of tree, the training sub-block is the cotyledon node of the minimum spanning tree.
Specifically, first the acquisition template image of module 302 is built, template image can be normal, health a people
Medical image, or the medical image of typical patient, template image can select according to actual conditions, this implementation
Example is not especially limited.Template image is made up of multiple pixels, obtains on template image the corresponding curvature of pixel more than the
One threshold value as template key point, and template key point is multiple, centered on template key point, obtains presetted pixel number
The template sub-block of composition；Multiple training images are obtained, the training image that will be got is registering to template image, and in training image
It is upper to extract the training key point corresponding with template key point, and centered on training key point acquisition predetermined number and template
Sub-block size identical trains sub-block, the training sub-block composing training sub-block group of predetermined number；And to training sub-block group using non-
Zone leveling algorithm is weighted average computation, obtains weighted average sub-block, calculates the of weighted average sub-block and template sub-block
Two degrees of correlation, the minimum spanning tree of template sub-block and training sub-block is built according to second degree of correlation.It is understood that most your pupil
The root node of Cheng Shu is template sub-block, and cotyledon node is to train sub-block, and template sub-block one minimum spanning tree of correspondence.
The embodiment of the present invention is built most by obtaining template sub-block and training sub-block according to template sub-block and training sub-block
Small spanning tree, sub-block scale is down to by registering yardstick from general image yardstick, so as to improve the accuracy of registration.
On the basis of above-described embodiment, the extraction module, specifically for：
Using the corresponding curvature of pixel in the test image more than Second Threshold as test key point, with the survey
Centered on examination key point, the test sub-block is obtained.
Specifically, in image registration, the pixel correspondence curvature that extraction module 301 can be chosen on image outline is big
Key point is tested in the conduct of Second Threshold, centered on the test key point, the pixel for obtaining predetermined number is constituted
Test sub-block, for example：Choose to test centered on key point, predetermined number is the test that 1 pixel of pixel constitutes 3 × 3
Sub-block.It is understood that the test sub-block is identical with template sub-block and training sub-block size.
The embodiment of the present invention is used as test by selecting the corresponding curvature of pixel in test image more than Second Threshold
Key point, and obtained according to test key point and template sub-block size identical test sub-block, selecting has diagnosis on image
The point of meaning so that higher with punctual accuracy.
On the basis of above-described embodiment, the computing module, specifically for：
First degree of correlation of the test sub-block and the training sub-block is calculated using normalized mutual information algorithm.
Specifically, computing module 303 calculates test sub-block all with training image using the algorithm of normalized mutual information
First degree of correlation of sub-block is trained, wherein first degree of correlation is similarity, the first phase between characterization test sub-block and training sub-block
Guan Du is bigger, illustrates that test sub-block is more similar to training sub-block.
The embodiment of the present invention is built minimum by extracting template sub-block and training sub-block according to template sub-block and training sub-block
Spanning tree, calculates first degree of correlation of training sub-block in test sub-block and minimum spanning tree and obtains sparse Deformation Field, according to dilute
Dredge Deformation Field and obtain dense Deformation Field, according to dense Deformation Field and test image to template image registration, improve registration
While accuracy, the on-line registration time is shortened.
On the basis of above-described embodiment, described second builds module, specifically for：
According to the sparse Deformation Field, dense Deformation Field is built using the method for interpolation.
Specifically, it is made up of selection optimal Deformation Field corresponding with the training sub-block of test sub-block second degree of correlation maximum
Be sparse Deformation Field, this be due to test sub-block be not covered with all of test image, therefore referred to as sparse Deformation Field, utilize
The method of interpolation, obtains the corresponding optimal Deformation Field in region for not having tested sub-block to cover in test image, so as to obtain thick
Close Deformation Field.
The embodiment of the present invention builds dense Deformation Field by the method for interpolation so that every piece of region in test image has
Corresponding optimal Deformation Field, such that it is able to realize accurately registration.
Medical figure registration equipment provided in an embodiment of the present invention be used for perform the above method, its specific embodiment with
The implementation method of method is consistent, and here is omitted.
The embodiment of the present invention is built minimum by extracting template sub-block and training sub-block according to template sub-block and training sub-block
Spanning tree, calculates first degree of correlation of training sub-block in test sub-block and minimum spanning tree and obtains sparse Deformation Field, according to dilute
Dredge Deformation Field and obtain dense Deformation Field, according to dense Deformation Field and test image to template image registration, improve registration
While accuracy, the on-line registration time is shortened.
Device embodiment described above is only schematical, wherein the unit illustrated as separating component can
To be or may not be physically separate, the part shown as unit can be or may not be physics list
Unit, you can with positioned at a place, or can also be distributed on multiple NEs.It can according to the actual needs be selected
In some or all of module realize the purpose of this embodiment scheme.Those of ordinary skill in the art are not paying creativeness
Work in the case of, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each implementation method can
Realized by the mode of software plus required general hardware platform, naturally it is also possible to by hardware.Based on such understanding, on
Stating the part that technical scheme substantially contributes to prior art in other words can be embodied in the form of software product, should
Computer software product can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including some fingers
Order is used to so that a computer equipment (can be personal computer, server, or network equipment etc.) performs each implementation
Method described in some parts of example or embodiment.
Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
The present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those within the art that：It still may be used
Modified with to the technical scheme described in foregoing embodiments, or equivalent is carried out to which part technical characteristic；
And these modification or replace, do not make appropriate technical solution essence depart from various embodiments of the present invention technical scheme spirit and
Scope.
Claims (10)
1. a kind of medical image registration method, it is characterised in that including：
Test image is obtained, multiple test sub-blocks are extracted according to the first preset rules in the test image；
Multiple minimum spanning trees are obtained, the minimum spanning tree is by template sub-block and training sub-block according to the second preset rules structure
Build, wherein, the template sub-block is the root node of the minimum spanning tree, and the training sub-block is the son of the minimum spanning tree
Leaf node, the template sub-block is extracted by template image and obtained, and the training sub-block is extracted by training image
Arrive, and the template sub-block, the training sub-block are identical with the test sub-block size；
First degree of correlation of the test sub-block and the training sub-block is calculated, and is selected and the test degree of correlation of sub-block first
The corresponding optimal Deformation Field of sub-block is trained described in highest；
Using the optimal Deformation Field as the sparse Deformation Field of the test image, and built according to the sparse Deformation Field dense
Deformation Field；
According to the dense Deformation Field and the test image to template image registration；Wherein, the optimal Deformation Field is by training
Sub-block is obtained to template sub-block registration.
2. method according to claim 1, it is characterised in that the minimum spanning tree is by template sub-block and training sub-block root
Built according to the second preset rules, including：
Template image is obtained, the corresponding curvature of pixel is obtained in the template image and is closed as template more than first threshold
Key point, centered on the template key point, the template sub-block that acquisition is constituted with presetted pixel number；
Training image is obtained, by the training image to template image registration, is obtained and is closed with the template on the template image
The corresponding training key point of key point, the training sub-block that predetermined number is obtained centered on the training key point is constituted instruction
Practice sub-block group；
Weighted average sub-block is calculated according to the training sub-block group, the of the template sub-block and the weighted average sub-block is calculated
Two degrees of correlation, minimum spanning tree is built according to second degree of correlation.
3. method according to claim 1, it is characterised in that it is described in the test image according to the first preset rules
Test sub-block is extracted, including：
Using the corresponding curvature of pixel in the test image more than Second Threshold as test key point, closed with the test
Centered on key point, the test sub-block is obtained.
4. method according to claim 2, it is characterised in that the calculating test sub-block and the training sub-block
First degree of correlation, including：
First degree of correlation of the test sub-block and the training sub-block is calculated using normalized mutual information algorithm.
5. method according to claim 2, it is characterised in that described that dense deformation is built according to the sparse Deformation Field
, including：
According to the sparse Deformation Field, dense Deformation Field is built using the method for interpolation.
6. a kind of medical figure registration equipment, it is characterised in that including：
Extraction module, for obtaining test image, multiple test is extracted in the test image according to the first preset rules
Block；
First builds module, and for obtaining multiple minimum spanning trees, the minimum spanning tree is by template sub-block and training sub-block root
Built according to the second preset rules, wherein, the template sub-block is extracted by template image and obtained, and the training sub-block is logical
Cross training image and extract what is obtained, and the template sub-block, the training sub-block are identical with the test sub-block size；
Computing module, for calculating first degree of correlation of the test sub-block and the training sub-block, and selects and the test
The corresponding optimal Deformation Field of sub-block is trained described in sub-block first degree of correlation highest；
Second builds module, for using the optimal Deformation Field as the test image sparse Deformation Field, and according to described
Sparse Deformation Field builds dense Deformation Field；
Registration module, for registering to template image according to the dense Deformation Field and the test image；Wherein, it is described optimal
Deformation Field is obtained from training sub-block to template sub-block registration.
7. medical figure registration equipment according to claim 6, it is characterised in that the first structure module is specifically used
In：
Template image is obtained, the corresponding curvature of pixel is obtained in the template image and is closed as template more than first threshold
Key point, centered on the template key point, the template sub-block that acquisition is constituted with presetted pixel number；
Training image is obtained, by the training image to template image registration, is obtained and is closed with the template on the template image
The corresponding training key point of key point, the training sub-block that predetermined number is obtained centered on the training key point is constituted instruction
Practice sub-block group；
Weighted average sub-block is calculated according to the training sub-block group, the of the template sub-block and the weighted average sub-block is calculated
Two degrees of correlation, build minimum spanning tree, wherein the template sub-block is the minimum spanning tree according to second degree of correlation
Root node, the training sub-block is the cotyledon node of the minimum spanning tree.
8. medical figure registration equipment according to claim 6, it is characterised in that the extraction module, specifically for：
Using the corresponding curvature of pixel in the test image more than Second Threshold as test key point, closed with the test
Centered on key point, the test sub-block is obtained.
9. medical figure registration equipment according to claim 7, it is characterised in that the computing module, specifically for：
First degree of correlation of the test sub-block and the training sub-block is calculated using normalized mutual information algorithm.
10. medical figure registration equipment according to claim 7, it is characterised in that described second builds module, specific to use
In：
According to the sparse Deformation Field, dense Deformation Field is built using the method for interpolation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611220617.XA CN106846380B (en) | 2016-12-26 | 2016-12-26 | A kind of medical image registration method and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611220617.XA CN106846380B (en) | 2016-12-26 | 2016-12-26 | A kind of medical image registration method and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106846380A true CN106846380A (en) | 2017-06-13 |
| CN106846380B CN106846380B (en) | 2019-09-03 |
Family
ID=59136740
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611220617.XA Active CN106846380B (en) | 2016-12-26 | 2016-12-26 | A kind of medical image registration method and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106846380B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110148150A (en) * | 2019-06-20 | 2019-08-20 | 深圳大学 | It is dense to connect asymmetric hierarchical network training method and heart movement field estimation method |
| CN110536142A (en) * | 2019-08-30 | 2019-12-03 | 天津大学 | A kind of interframe interpolation method for non-rigid image sequence |
| US10726555B2 (en) | 2018-06-06 | 2020-07-28 | International Business Machines Corporation | Joint registration and segmentation of images using deep learning |
| US10799183B2 (en) | 2018-11-07 | 2020-10-13 | General Electric Company | Methods and systems for whole body imaging |
| CN113393498A (en) * | 2021-05-26 | 2021-09-14 | 上海联影医疗科技股份有限公司 | Image registration method and device, computer equipment and storage medium |
| CN115641365A (en) * | 2022-08-25 | 2023-01-24 | 北京微视威信息科技有限公司 | Point cloud registration method, system, device and medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1892703A (en) * | 2005-06-22 | 2007-01-10 | 西门子共同研究公司 | Combining the apriori information from pre-alignment image to EMST-based image registration |
| CN104036452A (en) * | 2013-03-06 | 2014-09-10 | 株式会社东芝 | Image processing device, image processing method and medical image equipment |
| CN104077781A (en) * | 2014-07-07 | 2014-10-01 | 上海电力学院 | Medical image registration algorithm evaluation method based on deformation field |
| CN104424647A (en) * | 2013-09-04 | 2015-03-18 | 三星电子株式会社 | Method and apparatus for registering medical images |
| CN105096306A (en) * | 2014-05-12 | 2015-11-25 | 三星电子株式会社 | Method and apparatus for registering medical images |
| CN105427308A (en) * | 2015-11-20 | 2016-03-23 | 中国地质大学（武汉） | Sparse and dense characteristic matching combined image registration method |
-
2016
- 2016-12-26 CN CN201611220617.XA patent/CN106846380B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1892703A (en) * | 2005-06-22 | 2007-01-10 | 西门子共同研究公司 | Combining the apriori information from pre-alignment image to EMST-based image registration |
| CN104036452A (en) * | 2013-03-06 | 2014-09-10 | 株式会社东芝 | Image processing device, image processing method and medical image equipment |
| CN104424647A (en) * | 2013-09-04 | 2015-03-18 | 三星电子株式会社 | Method and apparatus for registering medical images |
| CN105096306A (en) * | 2014-05-12 | 2015-11-25 | 三星电子株式会社 | Method and apparatus for registering medical images |
| CN104077781A (en) * | 2014-07-07 | 2014-10-01 | 上海电力学院 | Medical image registration algorithm evaluation method based on deformation field |
| CN105427308A (en) * | 2015-11-20 | 2016-03-23 | 中国地质大学（武汉） | Sparse and dense characteristic matching combined image registration method |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10726555B2 (en) | 2018-06-06 | 2020-07-28 | International Business Machines Corporation | Joint registration and segmentation of images using deep learning |
| US10799183B2 (en) | 2018-11-07 | 2020-10-13 | General Electric Company | Methods and systems for whole body imaging |
| CN110148150A (en) * | 2019-06-20 | 2019-08-20 | 深圳大学 | It is dense to connect asymmetric hierarchical network training method and heart movement field estimation method |
| CN110148150B (en) * | 2019-06-20 | 2021-07-02 | 深圳大学 | Dense connection asymmetric hierarchical network training method and cardiac motion field estimation method |
| CN110536142A (en) * | 2019-08-30 | 2019-12-03 | 天津大学 | A kind of interframe interpolation method for non-rigid image sequence |
| CN110536142B (en) * | 2019-08-30 | 2021-11-09 | 天津大学 | Interframe interpolation method for non-rigid image sequence |
| CN113393498A (en) * | 2021-05-26 | 2021-09-14 | 上海联影医疗科技股份有限公司 | Image registration method and device, computer equipment and storage medium |
| CN115641365A (en) * | 2022-08-25 | 2023-01-24 | 北京微视威信息科技有限公司 | Point cloud registration method, system, device and medium |
| CN115641365B (en) * | 2022-08-25 | 2023-08-04 | 北京微视威信息科技有限公司 | Point cloud registration method, system, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106846380B (en) | 2019-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106846380A (en) | A kind of medical image registration method and equipment | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| CN108806792B (en) | Deep learning face diagnosis system | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| CN109543526B (en) | True and false facial paralysis recognition system based on depth difference characteristics | |
| CN109858540B (en) | Medical image recognition system and method based on multi-mode fusion | |
| CN110276745B (en) | Pathological image detection algorithm based on generation countermeasure network | |
| CN107832802A (en) | Quality of human face image evaluation method and device based on face alignment | |
| CN107292884A (en) | The method and device of oedema and hemotoncus in a kind of identification MRI image | |
| CN108564026A (en) | Network establishing method and system for Thyroid Neoplasms smear image classification | |
| CN102567734B (en) | Specific value based retina thin blood vessel segmentation method | |
| CN108511055A (en) | Ventricular premature beat identifying system and method based on Multiple Classifier Fusion and diagnostic rule | |
| CN109242825A (en) | A kind of steel surface defect identification method and device based on depth learning technology | |
| CN113095382B (en) | Interpretable tuberculosis classification network identification method based on CT image | |
| CN108564123A (en) | A kind of Thyroid Neoplasms smear image classification method and its device | |
| CN107330883A (en) | A kind of medical image lesion region positioning and sorting technique | |
| Rajput et al. | An accurate and noninvasive skin cancer screening based on imaging technique | |
| CN109767459A (en) | Novel ocular base map method for registering | |
| CN106446890A (en) | Candidate area extraction method based on window scoring and superpixel segmentation | |
| CN103984954B (en) | Image combining method based on multi-feature fusion | |
| CN110119746A (en) | A kind of characteristic recognition method and device, computer readable storage medium | |
| CN113782184A (en) | Cerebral apoplexy auxiliary evaluation system based on facial key point and feature pre-learning | |
| CN106169084A (en) | A kind of SVM mammary gland sorting technique based on Gauss kernel parameter selection | |
| CN113506274B (en) | Detection system for human cognitive condition based on visual saliency difference map | |
| CN105551042B (en) | A kind of scanning bed mark point method for determining position and device |
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 |