CN108898568A - Image composition method and device - Google Patents
Image composition method and device Download PDFInfo
- Publication number
- CN108898568A CN108898568A CN201810378364.1A CN201810378364A CN108898568A CN 108898568 A CN108898568 A CN 108898568A CN 201810378364 A CN201810378364 A CN 201810378364A CN 108898568 A CN108898568 A CN 108898568A
- Authority
- CN
- China
- Prior art keywords
- image
- magnetic resonance
- resolution dictionary
- current
- synthesized
- 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
- 238000000034 method Methods 0.000 title claims abstract description 25
- 239000000203 mixture Substances 0.000 title claims abstract description 19
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 16
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 15
- 239000011159 matrix material Substances 0.000 claims description 20
- 230000004927 fusion Effects 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 7
- 241001436679 Adama Species 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 4
- 210000003484 anatomy Anatomy 0.000 abstract description 2
- 238000013507 mapping Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 241001653634 Russula vesca Species 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000002143 encouraging effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 210000003625 skull Anatomy 0.000 description 1
- 238000010189 synthetic method Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
- G06T3/4076—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution using the original low-resolution images to iteratively correct the high-resolution images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/10—Image enhancement or restoration using non-spatial domain filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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
- G06T2207/20048—Transform domain processing
- G06T2207/20052—Discrete cosine transform [DCT]
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of image composition method and devices, including:Step 1,3T and 7T magnetic resonance image sample set is obtained;Step 2, input 3T magnetic resonance image is as current 3T image;Step 3, an optional image sheet and constructs low-resolution dictionary and high-resolution dictionary as current 3T image sheet from the multiple images piece in current 3T image;Step 4, the 7T image sheet of current 3T image sheet synthesis is obtained in the spatial domainIt is with high-resolution dictionaryStep 5, the 7T image sheet of current 3T image sheet x synthesis is obtained in a frequency domainWith high-resolution dictionaryFusion obtains the 7T image sheet synthesized in spatial domainWith high-resolution dictionaryThe 7T image sheet synthesized in frequency domainWith high-resolution dictionaryStep 7, step 4 is repeated to step 6, until when obtained 7T image sheet meets the condition of convergence.The present invention can reconstruct the 7T magnetic resonance image of high-quality, effectively synthesize the anatomical structure of high-fidelity, have better subjective and objective effect.
Description
Technical field
The invention belongs to field of image processings, and in particular to a kind of image composition method and device.
Background technique
With the rapid development of mr imaging technique, resolution ratio, signal-to-noise ratio and the scanning speed of magnetic resonance image have
Larger raising.However, 7T MRI scan instrument price is extremely expensive at present, distribution is very rare, and the whole world is less than 100
Platform.With this comparison, as a kind of routine clinical selection, 3T magnetic resonance scanner is since early 20th century as the gold mark of industry
Standard is still commonly used to scientific research and clinical practice so far.In order to improve magnetic resonance image quality, the common measure of clinic be using
The lesser voxel of size carries out Image Acquisition, can obtain more image details, higher resolution ratio and contrast, but simultaneously
Cause noise relatively low and sweep time longer problem.
The task of high field intensity (such as 7T) magnetic resonance image synthesis is that the 3T image of low resolution is mapped to high-resolution
7T image, but this is not simple image super-resolution problem, because the appearance of 7T with 3T image and contrast are different.Tradition
Image composition method be that high-resolution 7T image is rebuild using image histogram matching.It is calculated although these methods have
Advantage at low cost, but the 7T image of synthesis is be easy to cause the distortion phenomenons such as fuzzy, edge-smoothing and contrast distortion occur.
In order to overcome the defect of traditional images histogram matching, it is contemplated that image degradation process, based on the method for reconstruction by image
Synthesis is considered as an inverse problem, is usually solved by applying regularization constraint.These regularization constraints often rely on nature
Image Priori Knowledge, such as edge statistics, image gradient, structure self-similarity and sparsity.Since piece-wise constant is it is assumed that total
Body variation regularization method and its mutation method tend to excess smoothness high-definition picture.Method based on rarefaction representation passes through benefit
With input picture or exterior view image set come training dictionary, learn the mapping between low-resolution image piece and high resolution graphics photo
Relationship carries out sparse coding to low-resolution image to rebuild high-definition picture.In recent years, the method based on deep learning is first led to
Mapping relations between low resolution and high-definition picture of the overfitting from internal or external image data set, then this is reflected
It penetrates and is applied to high resolution image reconstruction, achieve encouraging effect.Wherein exemplary process has convolutional neural networks mould
Type and generation confrontation network model.Conventional images synthetic method is modeled to the priori of natural image to construct regularization inverse problem
And it solves, or estimate high-definition picture using the deep learning model of data-driven.Although these image composition methods
Preferable effect is achieved, but is usually used in natural image processing, lacks the particular attribute for considering medical image, is rarely applied to magnetic
Resonance image processing.
Summary of the invention
In view of the deficiencies of the prior art, the present invention intends to provide a kind of image composition method and device, solution
Certainly a large amount of training image collection and corresponding label is needed to train based on the image composition method of deep learning in the prior art
Convolutional neural networks model generates confrontation network model, but the problem of lack a large amount of magnetic resonance image datas.
In order to solve the above-mentioned technical problem, the present invention is realised by adopting the following technical scheme:
A kind of image composition method, includes the following steps:
Step 1, multiple 3T magnetic resonance samples image composition 3T magnetic resonance image sample sets are obtained, multiple 7T magnetic resonance are obtained
Sample image forms 7T magnetic resonance image sample set;
It will be in the 3T magnetic resonance samples image and 7T magnetic resonance image sample set in 3T magnetic resonance image sample set
One 7T magnetic resonance samples image partners 3T-7T sample image group, Q can be obtained to 3T-7T sample image group, Q is big
In the natural number for being equal to 1;
Step 2, any 3T magnetic resonance image is inputted as current 3T image, which is not belonging to 3T magnetic resonance figure
As sample set, the current 3T image includes multiple images piece;
Step 3, an optional image sheet is as current 3T image sheet x from the multiple images piece in current 3T image, in 3T
Preceding L is selected in magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x3T,l| l=1,2 ...,
L1To get arrive low-resolution dictionary
And L is selected in 7T magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x7T, l
| l=1,2 ..., L1To get arrive high-resolution dictionary
Step 4, in the spatial domain, the 7T image sheet synthesized by current 3T image sheet x is obtained by formula (1)
In formula, x is current 3T image sheet;
For projection matrix,DLRFor low-resolution dictionary, DHRFor high-resolution
Dictionary, D 'LRFor DLRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In the spatial domain, high-resolution dictionary, the high-resolution word synthesized in spatial domain are synthesized using low-resolution dictionary
Allusion quotation isWherein
Step 5, in a frequency domain, the 7T image sheet of current 3T image sheet x synthesis is obtained by formula (2)
In formula, UHRFor DHRDiscrete cosine transform coefficient, ULRFor DLRDiscrete cosine transform coefficient, α be current 3T image
The discrete cosine transform coefficient of piece x, U 'LRFor ULRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In a frequency domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary of synthesis is
Step 6, the 7T image sheet synthesized in spatial domain is respectively obtained by formula (3) and formula (4)With high-resolution dictionary
In formula, dct () is forward discrete cosine transform (DCT) transformation, and idct () is inverse discrete cosine transform (DCT)
Transformation,Indicate the Adama product of two matrixes;
The 7T image sheet synthesized in frequency domain is respectively obtained by formula (5) and formula (6)With high-resolution dictionary
Step 7, the 7T image that will be synthesized in spatial domainAs current 3T image x, the high-resolution that will be synthesized in spatial domain
DictionaryAs the low-resolution dictionary D in spatial domainLR, the 7T image that will be synthesized in frequency domainAs current 3T image x from
Dissipate cosine transform coefficient α, the high-resolution dictionary that will be synthesized in frequency domainDiscrete cosine transform system as low-resolution dictionary
Number ULR, repeat step 4 to step 6, until obtained 7T image sheet and when with a preceding iterative image error being less than ζ, ζ≤
10-5。
Further, current 3T image described in step 2 includes multiple images piece, wherein the size of each image sheet is p
× p × p, p > 0.
The present embodiment additionally provides a kind of image synthesizer, including:
Sample acquisition module is obtained for obtaining multiple 3T magnetic resonance samples image composition 3T magnetic resonance image sample sets
Multiple 7T magnetic resonance samples images form 7T magnetic resonance image sample set;
It will be in the 3T magnetic resonance samples image and 7T magnetic resonance image sample set in 3T magnetic resonance image sample set
One 7T magnetic resonance samples image partners 3T-7T sample image group, Q can be obtained to 3T-7T sample image group, Q is big
In the natural number for being equal to 1;
3T image input module, for inputting any 3T magnetic resonance image as current 3T image, the current 3T image is not
Belong to 3T magnetic resonance image sample set, the current 3T image includes multiple images piece;
Dictionary constructs module, schemes for an image sheet optional from the multiple images piece in current 3T image as current 3T
Photo x selects L in 3T magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x3T,l| l=
1,2,…,L1To get arrive low-resolution dictionary
And L is selected in 7T magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x7T,l
| l=1,2 ..., L1To get arrive high-resolution dictionary
Spatial domain regression block, in the spatial domain, the 7T synthesized by current 3T image sheet x being obtained by formula (1)
Image sheet
In formula, x is current 3T image sheet;
For projection matrix,DLRFor low-resolution dictionary, DHRFor high-resolution
Dictionary, D 'LRFor DLRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In the spatial domain, high-resolution dictionary, the high-resolution word synthesized in spatial domain are synthesized using low-resolution dictionary
Allusion quotation isWherein
Frequency domain regression block, in a frequency domain, obtaining the 7T image sheet of current 3T image sheet x synthesis by formula (2)
In formula, UHRFor DHRDiscrete cosine transform coefficient, ULRFor DLRDiscrete cosine transform coefficient, α be current 3T image
The discrete cosine transform coefficient of piece x, U 'LRFor ULRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In a frequency domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary synthesized in frequency domain is
Fusion Module, for respectively obtaining the 7T image sheet synthesized in spatial domain by formula (3) and formula (4)And high-resolution
Rate dictionary
In formula, dct () is preceding to dct transform, and idct () is inverse dct transform,Indicate that the Adama of two matrixes multiplies
Product;
The 7T image sheet synthesized in frequency domain is respectively obtained by formula (5) and formula (6)With high-resolution dictionary
Iterative cycles module, the 7T image for will be synthesized in spatial domainAs current 3T image x, will be closed in spatial domain
At high-resolution dictionaryAs the low-resolution dictionary D in spatial domainLR, the 7T image that will be synthesized in frequency domainAs working as
The discrete cosine transform coefficient α of preceding 3T image x, the high-resolution dictionary that will be synthesized in frequency domainAs low-resolution dictionary
Discrete cosine transform coefficient ULR, spatial domain regression block is repeated to Fusion Module, up to obtained 7T image sheet and with before
Until when an iteration image error is less than ζ, ζ≤10-5。
Further, current 3T image described in 3T image input module includes multiple images piece, wherein each image sheet
Size be p × p × p, p > 0.
Compared with prior art, the present invention having the following technical effect that:
The present invention can reconstruct the 7T magnetic resonance image of high-quality, effectively synthesize the anatomical structure of high-fidelity, tool
There are better subjective and objective effect, usually more than existing state-of-the-art image composition method.
Detailed description of the invention
Fig. 1 is block schematic illustration of the invention;
Fig. 2 is flow diagram of the invention.
Explanation is further explained in detail to particular content of the invention below in conjunction with attached drawing.
Specific embodiment
The present invention proposes that a kind of dual domain cascades regression model to express the complex mapping relation between 3T and 7T image, this hair
The bright model is two recurrence stream that spatial domain and frequency domain are executed on multiple stages.Have in view of discrete cosine transform (DCT)
The advantages of being simple and efficient, the present invention select DCT for the conversion between spatial domain and frequency domain.For each rank of regression process
Section, executes the mapping transformation of the mode from 3T to 7T respectively.Then spatial domain and frequency domain are returned into composite result fusion as next
The input of stage regression.And the input only returned for the first time in the present invention is the 3T image for most starting input, and remaining
The input in stage is the intermediate 7T image temporarily synthesized.
Specific embodiments of the present invention are given below, it should be noted that the invention is not limited to implement in detail below
Example, all equivalent transformations made on the basis of the technical solutions of the present application each fall within protection scope of the present invention.
Embodiment 1:
As shown in Fig. 2, including the following steps the present invention provides a kind of image composition method:
Step 1, multiple 3T magnetic resonance samples image composition 3T magnetic resonance image sample sets are obtained, multiple 7T magnetic resonance are obtained
Sample image forms 7T magnetic resonance image sample set;
It will be in the 3T magnetic resonance samples image and 7T magnetic resonance image sample set in 3T magnetic resonance image sample set
One 7T magnetic resonance samples image partners 3T-7T sample image group, Q can be obtained to 3T-7T sample image group, Q is big
In the natural number for being equal to 1;
The present embodiment matches (affine registration) algorithm for 3T magnetic resonance image sample set using affine transformation
In a 3T magnetic resonance samples image and 7T magnetic resonance image sample set in a 7T magnetic resonance samples image matched
After partner 3T-7T sample image group.
The present embodiment will input 3T image I using the FLIRT function in FSL kit3TWith all 3T-7T sample images
Group is registrated to MNI normed space to eliminate posture difference.Specifically, all 3T linearities are registrated to first with different templates
MNI normed space, then 7T sample image is registrated to corresponding 3T image.After co-registration, implementation deviation correction and skull stripping
From extracting brain image.In order to inhibit different scanning instrument and website to utilize Histogram Matching side to the influence of magnetic resonance image
Method is respectively normalized 3T and 7T brightness of image and it is made to zoom to tonal range [0,1].For 3T image, by institute
There is histogram of the Histogram Matching of normalized 3T sample image to normalization input 3T image.Similarly, it chooses and inputs
The corresponding 7T sample image of the nearest 3T sample image of 3T image Euclidean distance, which is used as, refers to 7T image, utilizes Histogram Matching side
Every other 7T sample image is matched to this with reference to 7T image by method.
Step 2, any 3T magnetic resonance image is inputted as current 3T image, which is not belonging to 3T magnetic resonance figure
As sample set, the current 3T image includes multiple images piece;
Specifically, the size of each image sheet is p × p × p, p > 0.P=3 in the present embodiment.
Step 3, an optional image sheet is as current 3T image sheet x from the multiple images piece in current 3T image, in 3T
Preceding L is selected in magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x3T,l| l=1,2 ...,
L1, above-mentioned sample graph photo be used to construct low-resolution dictionary to get to low-resolution dictionary in the present embodiment
L before being selected in 3T magnetic resonance image sample set in the present embodiment1A sample image similar with present image piece x
Piece set { z3T,l| l=1,2 ..., L1It is specially L before being selected in 3T magnetic resonance image sample set1A and Europe present image piece x
Formula is apart from the smallest sample graph photo set { z3T,l| l=1,2 ..., L1}。
And L before being selected in 7T magnetic resonance image sample set1A sample graph photo set similar with present image piece x
{z7T,l| l=1,2 ..., L1, above-mentioned sample graph photo be used to construct high-resolution dictionary to get to high score in the present embodiment
Resolution dictionary
The present invention returns the first stage of stream in spatial domain, proposes a kind of linear regression model (LRM) to indicate from low resolution word
Mapping relations of the allusion quotation to high-resolution dictionary:DHR=BsDLR+ ε, wherein BsIt is projection matrix, ε is error;
The present invention is solved the inverse problem of above-mentioned mapping relations using Ridge Regression Modeling Method and provides optimization form:||||2Indicate two norms;
To projection matrix BsIt can be expressed as the form of closed solution:It can obtain
To the 7T image sheet synthesized in step 4;
Step 4, in the spatial domain, the 7T image sheet synthesized by current 3T image sheet x is obtained by formula (1)
In formula, x is current 3T image sheet;
For projection matrix,DLRFor low-resolution dictionary, DHRFor high-resolution
Dictionary, D 'LRFor DLRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In the spatial domain, high-resolution dictionary, the high-resolution word synthesized in spatial domain are synthesized using low-resolution dictionary
Allusion quotation isWhereinThe high-resolution dictionary of the synthesis in the present invention isFor constructing next recurrence node
Low-resolution dictionary.
Step 5, in a frequency domain, the 7T image sheet of current 3T image sheet x synthesis is obtained by formula (2)
In formula, UHRFor DHRDiscrete cosine transform coefficient, ULRFor DLRDiscrete cosine transform coefficient, α be current 3T image
The discrete cosine transform coefficient of piece x, U 'LRFor ULRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In a frequency domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary synthesized in frequency domain is
Step 6, the 7T image sheet synthesized in spatial domain is respectively obtained by formula (3) and formula (4)With high-resolution dictionary
In formula, dct () is preceding to dct transform, and idct () is inverse dct transform,Indicate the Hadamard of two matrixes
Product;
The 7T image sheet synthesized in frequency domain is respectively obtained by formula (5) and formula (6)With high-resolution dictionary
Step 7, the 7T image sheet that will be synthesized in spatial domainAs current 3T image sheet x, the high score that will be synthesized in spatial domain
Resolution dictionaryAs the low-resolution dictionary D in spatial domainLR, the 7T image sheet that will be synthesized in frequency domainScheme as current 3T
The discrete cosine transform coefficient α of photo x, the high-resolution dictionary that will be synthesized in frequency domainAs the discrete of low-resolution dictionary
Cosine transform coefficient ULR, step 4 is repeated to step 6, is less than ζ up to obtained 7T image sheet and with a preceding iterative image error
When until, ζ≤10-5。
The present invention is calculatedWithCascade returns stream and holds respectively in spatial domain and frequency domain
Row.Specifically, the cascade in spatial domain is returned, by what is synthesized in the kth stageAnd dictionaryRespectively as kth+1
The input x and low-resolution dictionary D in stageLR.Similarly, the HR coefficient synthesized in kth stage frequency domainWith high-resolution word
Allusion quotationRespectively as the discrete cosine transform coefficient α and low-resolution dictionary U of the input picture of stage k+1LR.On the other hand,
The high-resolution dictionary D in the spatial domain and frequency domain in kth stageHRAnd UHRIt is taken as the high-resolution word in+1 stage of kth respectively
Allusion quotation.By the input of installation space domain and frequency domain, low-resolution dictionary and high-resolution dictionary, and formula (1)~(2) are executed to obtain
Obtain the regression result of current generation.Then we can utilize formula (3)~(6) fusion regression result to be sent into next stage again.
By the recurrence learning in K stage, the 7T image finally estimated is constructed by collecting the 7T image sheet of all synthesis.
Embodiment 2:
As shown in Figure 1, the present embodiment additionally provides a kind of image synthesizer, including:
Sample acquisition module is obtained for obtaining multiple 3T magnetic resonance samples image composition 3T magnetic resonance image sample sets
Multiple 7T magnetic resonance samples images form 7T magnetic resonance image sample set;
It will be in the 3T magnetic resonance samples image and 7T magnetic resonance image sample set in 3T magnetic resonance image sample set
One 7T magnetic resonance samples image partners 3T-7T sample image group, Q can be obtained to 3T-7T sample image group, Q is big
In the natural number for being equal to 1;
3T image input module, for inputting any 3T magnetic resonance image as current 3T image, the current 3T image is not
Belong to 3T magnetic resonance image sample set, the current 3T image includes multiple images piece;
Specifically, wherein the size of each image sheet is p × p × p, p > 0.P=3 in the present embodiment.
Dictionary constructs module, schemes for an image sheet optional from the multiple images piece in current 3T image as current 3T
Photo x selects L in 3T magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x3T,l| l=
1,2,…,L1To get arrive low-resolution dictionary
And L before being selected in 7T magnetic resonance image sample set1A sample graph photo set similar with present image piece x
{z7T,l| l=1,2 ..., L1To get arrive high-resolution dictionary
Spatial domain regression block, in the spatial domain, the 7T synthesized by current 3T image sheet x being obtained by formula (1)
Image sheet
In formula, x is current 3T image sheet x;
For projection matrix,DLRFor low-resolution dictionary, DHRFor high-resolution
Dictionary, D 'LRFor DLRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In the spatial domain, high-resolution dictionary, the high-resolution word synthesized in spatial domain are synthesized using low-resolution dictionary
Allusion quotation isWherein
Frequency domain regression block, in a frequency domain, obtaining the 7T image sheet of current 3T image sheet x synthesis by formula (2)
In formula, UHRFor DHRDiscrete cosine transform coefficient, ULRFor DLRDiscrete cosine transform coefficient, α be current 3T image
The discrete cosine transform coefficient of piece x, U 'LRFor ULRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In a frequency domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary synthesized in frequency domain is
Fusion Module, for respectively obtaining the 7T image sheet synthesized in spatial domain by formula (3) and formula (4)And high-resolution
Rate dictionary
In formula, dct () is preceding to dct transform, and idct () is inverse dct transform,Indicate the Hadamard of two matrixes
Product;
The 7T image sheet synthesized in frequency domain is respectively obtained by formula (5) and formula (6)With high-resolution dictionary
Iterative cycles module, the 7T image sheet for will be synthesized in spatial domainAs current 3T image sheet x, by spatial domain
The high-resolution dictionary of middle synthesisAs the low-resolution dictionary D in spatial domainLR, the 7T image sheet that will be synthesized in frequency domain
As the discrete cosine transform coefficient α of current 3T image sheet x, the high-resolution dictionary that will be synthesized in frequency domainAs low resolution
The discrete cosine transform coefficient U of rate dictionaryLR, spatial domain regression block is repeated to Fusion Module, until obtained 7T image
Piece and with a preceding iterative image error be less than ζ when until, ζ≤10-5。
Claims (4)
1. a kind of image composition method, which is characterized in that include the following steps:
Step 1, multiple 3T magnetic resonance samples image composition 3T magnetic resonance image sample sets are obtained, multiple 7T magnetic resonance samples are obtained
Image forms 7T magnetic resonance image sample set;
By one in the 3T magnetic resonance samples image and 7T magnetic resonance image sample set in 3T magnetic resonance image sample set
7T magnetic resonance samples image partners 3T-7T sample image group, Q can be obtained to 3T-7T sample image group, Q be greater than etc.
In 1 natural number;
Step 2, any 3T magnetic resonance image is inputted as current 3T image, which is not belonging to 3T magnetic resonance image sample
This collection, the current 3T image includes multiple images piece;
Step 3, an optional image sheet is total in 3T magnetic as current 3T image sheet x from the multiple images piece in current 3T image
Vibration image pattern concentration selects preceding L1A sample graph photo set { z similar with present image piece x3T,l| l=1,2 ..., L1,
Obtain low-resolution dictionary
And L is selected in 7T magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x7T,l| l=
1,2,…,L1To get arrive high-resolution dictionary
Step 4, in the spatial domain, the 7T image sheet synthesized by current 3T image sheet x is obtained by formula (1)
In formula, x is current 3T image sheet;
For projection matrix,DLRFor low-resolution dictionary, DHRFor high-resolution dictionary,
D′LRFor DLRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In the spatial domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary synthesized in spatial domain isWherein
Step 5, in a frequency domain, the 7T image sheet of current 3T image sheet x synthesis is obtained by formula (2)
In formula, UHRFor DHRDiscrete cosine transform coefficient, ULRFor DLRDiscrete cosine transform coefficient, α be current 3T image sheet x
Discrete cosine transform coefficient, U 'LRFor ULRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In a frequency domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary of synthesis is
Step 6, the 7T image sheet synthesized in spatial domain is respectively obtained by formula (3) and formula (4)With high-resolution dictionary
In formula, dct () is forward discrete cosine transform (DCT) transformation, and idct () is inverse discrete cosine transform (DCT) change
It changes,Indicate the Adama product of two matrixes;
The 7T image sheet synthesized in frequency domain is respectively obtained by formula (5) and formula (6)With high-resolution dictionary
Step 7, the 7T image that will be synthesized in spatial domainAs current 3T image x, the high-resolution dictionary that will be synthesized in spatial domainAs the low-resolution dictionary D in spatial domainLR, the 7T image that will be synthesized in frequency domainAs the discrete remaining of current 3T image x
String transformation coefficient α, the high-resolution dictionary that will be synthesized in frequency domainDiscrete cosine transform coefficient as low-resolution dictionary
ULR, repeat step 4 to step 6, until obtained 7T image sheet and when with a preceding iterative image error being less than ζ, ζ≤
10-5。
2. image composition method according to claim 1, which is characterized in that current 3T image described in step 2 includes more
A image sheet, wherein the size of each image sheet is p × p × p, p > 0.
3. a kind of image synthesizer, which is characterized in that including:
Sample acquisition module obtains multiple for obtaining multiple 3T magnetic resonance samples image composition 3T magnetic resonance image sample sets
7T magnetic resonance samples image forms 7T magnetic resonance image sample set;
By one in the 3T magnetic resonance samples image and 7T magnetic resonance image sample set in 3T magnetic resonance image sample set
7T magnetic resonance samples image partners 3T-7T sample image group, Q can be obtained to 3T-7T sample image group, Q be greater than etc.
In 1 natural number;
3T image input module, for inputting any 3T magnetic resonance image as current 3T image, which is not belonging to
3T magnetic resonance image sample set, the current 3T image includes multiple images piece;
Dictionary constructs module, for an image sheet optional from the multiple images piece in current 3T image as current 3T image sheet
X selects L in 3T magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x3T,l| l=1,
2,…,L1To get arrive low-resolution dictionary
And L is selected in 7T magnetic resonance image sample set1A sample graph photo set { z similar with present image piece x7T,l| l=
1,2,…,L1To get arrive high-resolution dictionary
Spatial domain regression block, in the spatial domain, the 7T image synthesized by current 3T image sheet x being obtained by formula (1)
Piece
In formula, x is current 3T image sheet;
For projection matrix,DLRFor low-resolution dictionary, DHRFor high-resolution dictionary,
D′LRFor DLRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In the spatial domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary synthesized in spatial domain isWherein
Frequency domain regression block, in a frequency domain, obtaining the 7T image sheet of current 3T image sheet x synthesis by formula (2)
In formula, UHRFor DHRDiscrete cosine transform coefficient, ULRFor DLRDiscrete cosine transform coefficient, α be current 3T image sheet x
Discrete cosine transform coefficient, U 'LRFor ULRTransposition, λ is regularization parameter, λ>0, E is unit matrix;
In a frequency domain, high-resolution dictionary is synthesized using low-resolution dictionary, the high-resolution dictionary synthesized in frequency domain is
Fusion Module, for respectively obtaining the 7T image sheet synthesized in spatial domain by formula (3) and formula (4)With high-resolution word
Allusion quotation
In formula, dct () is preceding to dct transform, and idct () is inverse dct transform,Indicate the Adama product of two matrixes;
The 7T image sheet synthesized in frequency domain is respectively obtained by formula (5) and formula (6)With high-resolution dictionary
Iterative cycles module, the 7T image for will be synthesized in spatial domainAs current 3T image x, by what is synthesized in spatial domain
High-resolution dictionaryAs the low-resolution dictionary D in spatial domainLR, the 7T image that will be synthesized in frequency domainAs current 3T
The discrete cosine transform coefficient α of image x, the high-resolution dictionary that will be synthesized in frequency domainAs the discrete of low-resolution dictionary
Cosine transform coefficient ULR, repeat spatial domain regression block to Fusion Module, until obtained 7T image sheet and with it is preceding primary
Until when iterative image error is less than ζ, ζ≤10-5。
4. image synthesizer according to claim 3, which is characterized in that current 3T figure described in 3T image input module
As including multiple images piece, wherein the size of each image sheet is p × p × p, p > 0.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810378364.1A CN108898568B (en) | 2018-04-25 | 2018-04-25 | Image synthesis method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810378364.1A CN108898568B (en) | 2018-04-25 | 2018-04-25 | Image synthesis method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108898568A true CN108898568A (en) | 2018-11-27 |
CN108898568B CN108898568B (en) | 2021-08-31 |
Family
ID=64342539
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810378364.1A Active CN108898568B (en) | 2018-04-25 | 2018-04-25 | Image synthesis method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108898568B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109741416A (en) * | 2019-01-04 | 2019-05-10 | 北京大学深圳医院 | Nuclear magnetic resonance image method for reconstructing, device, computer equipment and its storage medium |
CN110610458A (en) * | 2019-04-30 | 2019-12-24 | 北京联合大学 | Method and system for GAN image enhancement interactive processing based on ridge regression |
WO2020151355A1 (en) * | 2019-01-25 | 2020-07-30 | 厦门大学 | Deep learning-based magnetic resonance spectroscopy reconstruction method |
CN112348743A (en) * | 2020-11-06 | 2021-02-09 | 天津大学 | Image super-resolution method fusing discriminant network and generation network |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102968781A (en) * | 2012-12-11 | 2013-03-13 | 西北工业大学 | Image fusion method based on NSCT (Non Subsampled Contourlet Transform) and sparse representation |
CN104077791A (en) * | 2014-05-22 | 2014-10-01 | 南京信息工程大学 | Joint reconstruction method for multiple dynamic contrast enhancement nuclear magnetic resonance images |
CN104680502A (en) * | 2015-03-19 | 2015-06-03 | 四川大学 | Infrared image super-resolution reconstruction method based on sparse dictionary and non-subsample Contourlet transform |
CN105913409A (en) * | 2016-07-12 | 2016-08-31 | 常俊苹 | Image processing method based on fusion of multiple frames of images |
US20170091963A1 (en) * | 2015-09-28 | 2017-03-30 | Siemens Medical Solutions Usa, Inc. | Motion correction in a projection domain in time of flight positron emission tomography |
CN106780342A (en) * | 2016-12-28 | 2017-05-31 | 深圳市华星光电技术有限公司 | Single-frame image super-resolution reconstruction method and device based on the reconstruct of sparse domain |
EP3295202A1 (en) * | 2015-05-08 | 2018-03-21 | Max-Planck-Gesellschaft zur Förderung der Wissenschaften e.V. | Method and device for magnetic resonance imaging with improved sensitivity by noise reduction |
-
2018
- 2018-04-25 CN CN201810378364.1A patent/CN108898568B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102968781A (en) * | 2012-12-11 | 2013-03-13 | 西北工业大学 | Image fusion method based on NSCT (Non Subsampled Contourlet Transform) and sparse representation |
CN104077791A (en) * | 2014-05-22 | 2014-10-01 | 南京信息工程大学 | Joint reconstruction method for multiple dynamic contrast enhancement nuclear magnetic resonance images |
CN104680502A (en) * | 2015-03-19 | 2015-06-03 | 四川大学 | Infrared image super-resolution reconstruction method based on sparse dictionary and non-subsample Contourlet transform |
EP3295202A1 (en) * | 2015-05-08 | 2018-03-21 | Max-Planck-Gesellschaft zur Förderung der Wissenschaften e.V. | Method and device for magnetic resonance imaging with improved sensitivity by noise reduction |
US20170091963A1 (en) * | 2015-09-28 | 2017-03-30 | Siemens Medical Solutions Usa, Inc. | Motion correction in a projection domain in time of flight positron emission tomography |
CN105913409A (en) * | 2016-07-12 | 2016-08-31 | 常俊苹 | Image processing method based on fusion of multiple frames of images |
CN106780342A (en) * | 2016-12-28 | 2017-05-31 | 深圳市华星光电技术有限公司 | Single-frame image super-resolution reconstruction method and device based on the reconstruct of sparse domain |
Non-Patent Citations (4)
Title |
---|
KHOSRO BAHRAMI ET AL.: "Reconstruction of 7T-Like Images From 3T MRI", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
KHOSRO BAHRAMI: "7T-guided super-resolution of 3T MRI", 《MEDICAL PHYSICS》 * |
YONGQIN ZHANG ET AL.: "Image Super-Resolution Based on Structure-Modulated Sparse Representation", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
龙超: "图像超分辨率重建算法综述", 《科技视界》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109741416A (en) * | 2019-01-04 | 2019-05-10 | 北京大学深圳医院 | Nuclear magnetic resonance image method for reconstructing, device, computer equipment and its storage medium |
WO2020151355A1 (en) * | 2019-01-25 | 2020-07-30 | 厦门大学 | Deep learning-based magnetic resonance spectroscopy reconstruction method |
US11782111B2 (en) | 2019-01-25 | 2023-10-10 | Xiamen University | Method for reconstructing magnetic resonance spectrum based on deep learning |
CN110610458A (en) * | 2019-04-30 | 2019-12-24 | 北京联合大学 | Method and system for GAN image enhancement interactive processing based on ridge regression |
CN110610458B (en) * | 2019-04-30 | 2023-10-20 | 北京联合大学 | GAN image enhancement interaction processing method and system based on ridge regression |
CN112348743A (en) * | 2020-11-06 | 2021-02-09 | 天津大学 | Image super-resolution method fusing discriminant network and generation network |
CN112348743B (en) * | 2020-11-06 | 2023-01-31 | 天津大学 | Image super-resolution method fusing discriminant network and generation network |
Also Published As
Publication number | Publication date |
---|---|
CN108898568B (en) | 2021-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108898568A (en) | Image composition method and device | |
CN108090871B (en) | Multi-contrast magnetic resonance image reconstruction method based on convolutional neural network | |
CN105844635B (en) | A kind of rarefaction representation depth image method for reconstructing based on structure dictionary | |
CN105550649B (en) | Extremely low resolution ratio face identification method and system based on unity couping local constraint representation | |
CN105631807B (en) | The single-frame image super-resolution reconstruction method chosen based on sparse domain | |
CN109214989B (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
Song et al. | MSDNet for medical image fusion | |
CN107845065B (en) | Super-resolution image reconstruction method and device | |
Du et al. | Accelerated super-resolution MR image reconstruction via a 3D densely connected deep convolutional neural network | |
CN112734764A (en) | Unsupervised medical image segmentation method based on countermeasure network | |
CN107301630B (en) | CS-MRI image reconstruction method based on ordering structure group non-convex constraint | |
CN114170088A (en) | Relational reinforcement learning system and method based on graph structure data | |
CN111487573B (en) | Enhanced residual error cascade network model for magnetic resonance undersampling imaging | |
CN111080591A (en) | Medical image segmentation method based on combination of coding and decoding structure and residual error module | |
CN104574456A (en) | Graph regularization sparse coding-based magnetic resonance super-undersampled K data imaging method | |
CN111161405A (en) | Three-dimensional reconstruction method for animal hair | |
Jacobsen et al. | Stylometry of paintings using hidden Markov modelling of contourlet transforms | |
CN115511767A (en) | Self-supervised learning multi-modal image fusion method and application thereof | |
CN109544488B (en) | Image synthesis method based on convolutional neural network | |
Lila et al. | Statistical analysis of functions on surfaces, with an application to medical imaging | |
CN114299185A (en) | Magnetic resonance image generation method, magnetic resonance image generation device, computer equipment and storage medium | |
CN116051494A (en) | Fourier-based meta-learning field generalization less-sample medical image segmentation method | |
CN107103592B (en) | Multi-pose face image quality enhancement method based on dual-core norm regularization | |
CN117333750A (en) | Spatial registration and local global multi-scale multi-modal medical image fusion method | |
CN116630964A (en) | Food image segmentation method based on discrete wavelet attention network |
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 | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20230317 Address after: Room 101-1, Cultural and Creative Industries, Lanpan South 1983, No. 235, Xujiang Road, Gusu District, Suzhou City, Jiangsu Province, 215002 Patentee after: Suzhou Kujia Technology Co.,Ltd. Address before: 710069 No. 229 Taibai North Road, Shaanxi, Xi'an Patentee before: NORTHWEST University |