CN107680146A - Method for reconstructing, device, equipment and the storage medium of PET image - Google Patents
Method for reconstructing, device, equipment and the storage medium of PET image Download PDFInfo
- Publication number
- CN107680146A CN107680146A CN201710821065.6A CN201710821065A CN107680146A CN 107680146 A CN107680146 A CN 107680146A CN 201710821065 A CN201710821065 A CN 201710821065A CN 107680146 A CN107680146 A CN 107680146A
- Authority
- CN
- China
- Prior art keywords
- pet image
- low resolution
- image
- resolution pet
- resolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2211/00—Image generation
- G06T2211/40—Computed tomography
- G06T2211/416—Exact reconstruction
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The applicable field of computer technology of the present invention, there is provided a kind of method for reconstructing of PET image, device, equipment and storage medium, this method include：The low resolution PET image of reception is split, according to each low resolution PET image block obtained after segmentation, low-resolution dictionary and high-resolution dictionary, generate the sparse coefficient of each low resolution PET image block, according to these sparse coefficients and high-resolution dictionary, generate high-resolution PET image corresponding to low resolution PET image, according to low resolution PET image, high-resolution PET image, fuzzy matrix and down-sampling matrix, generate and export the reconstruction image of low resolution PET image, so as to significantly reduce the time loss of PET image reconstruction, the picture quality being effectively improved after PET image reconstruction efficiency and PET image reconstruction.
Description
Technical field
The invention belongs to technical field of image processing, more particularly to a kind of method for reconstructing of PET image, device, equipment and
Storage medium.
Background technology
Positron emission computerized tomography imaging (PET) has a very high clinical value, but due to PET image-forming principles and
The limitation of PET hardware systems, the PET image that normal scan obtains is often very fuzzy, and can lost part marginal information.With
Scan obtained low dosage sampled data and carry out PET image reconstruction, the picture quality for the PET image for rebuilding to obtain can be worse, bag
Containing substantial amounts of noise, or even produce image artifacts.
At present, in terms of picture noise is suppressed, repair image edge information there is more limit in traditional PET algorithm for reconstructing
System, can not effectively improve PET image quality, and the existing algorithm rebuild to PET image uses the side of iteration optimization mostly
Formula improves the quality of PET image, can not only expend longer time, and to lack sampling data by way of iteration optimization
The suppression of image artifacts is carried out, some minutias of missing image can be caused.
The content of the invention
It is an object of the invention to provide a kind of method for reconstructing of PET image, device, equipment and storage medium, it is intended to solves
Certainly PET image reconstruction in the prior art is inefficient, and rebuilds the problem of obtained picture quality is bad.
On the one hand, the invention provides a kind of method for reconstructing of PET image, methods described to comprise the steps：
The low resolution PET image of user's input is received, the low resolution PET image is split, generates low point
Resolution PET image block；
According to the good low-resolution dictionary of each low resolution PET image block, training in advance and high-resolution dictionary,
Generate the sparse coefficient of each low resolution PET image block；
It is described low according to the sparse coefficient of each low resolution PET image block and the high-resolution dictionary, generation
High-resolution PET image corresponding to resolution PET images；
According to the low resolution PET image, the high-resolution PET image, default fuzzy matrix and it is default under
Sampling matrix, generate and export the reconstruction image of the low resolution PET image.
On the other hand, the invention provides a kind of reconstructing device of PET image, described device to include：
Image segmentation unit, for receiving the low resolution PET image of user's input, to the low resolution PET image
Split, generate low resolution PET image block；
Coefficient generation unit, for according to the good low resolution word of each low resolution PET image block, training in advance
Allusion quotation and high-resolution dictionary, generate the sparse coefficient of each low resolution PET image block；
Image generation unit, for the sparse coefficient according to each low resolution PET image block and the high-resolution
Rate dictionary, generate high-resolution PET image corresponding to the low resolution PET image；And
Image output unit, for according to the low resolution PET image, the high-resolution PET image, default mould
Matrix and default down-sampling matrix are pasted, generates and exports the reconstruction image of the low resolution PET image.
On the other hand, present invention also offers a kind of Medical Image Processing equipment, including memory, processor and storage
In the memory and the computer program that can run on the processor, computer program described in the computing device
Steps of the Shi Shixian as described in the method for reconstructing of above-mentioned PET image.
On the other hand, present invention also offers a kind of computer-readable recording medium, the computer-readable recording medium
Computer program is stored with, is realized when the computer program is executed by processor as described in the method for reconstructing of above-mentioned PET image
The step of.
The present invention is according to the low resolution PET image block after segmentation, the low-resolution dictionary trained and high-resolution word
Allusion quotation, calculates the sparse coefficient of each low resolution PET image block, and low point is calculated according to these sparse coefficients and high-resolution dictionary
High-resolution PET image corresponding to resolution PET image, by being post-processed to high-resolution PET image, obtain low resolution
PET reconstruction image, so as to be effectively improved the picture quality of PET image reconstruction, significantly reduce in PET image reconstruction
Computation complexity, and then be effectively improved the reconstruction efficiency of PET image.
Brief description of the drawings
Fig. 1 is the implementation process figure of the method for reconstructing for the PET image that the embodiment of the present invention one provides；
Fig. 2 is the implementation process figure of the method for reconstructing for the PET image that the embodiment of the present invention two provides；
Fig. 3 is the structural representation of the reconstructing device for the PET image that the embodiment of the present invention three provides；
Fig. 4 is the structural representation of the reconstructing device for the PET image that the embodiment of the present invention four provides；And
Fig. 5 is the structural representation for the Medical Image Processing equipment that the embodiment of the present invention five provides.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
It is described in detail below in conjunction with specific implementation of the specific embodiment to the present invention：
Embodiment one：
Fig. 1 shows the implementation process of the method for reconstructing for the PET image that the embodiment of the present invention one provides, for the ease of saying
It is bright, the part related to the embodiment of the present invention is illustrate only, details are as follows：
In step S101, the low resolution PET image of user's input is received, low resolution PET image is split,
Generate low resolution PET image block；
The embodiment of the present invention is applied to the image reconstruction of positron emission tomography (PET) image.Can be by default
The low resolution PET image that image segmentation algorithm inputs to the user received carries out the segmentation of image block, multiple low points of generation
Resolution PET image block, specifically, in segmentation, the size of each low resolution PET image block is identical, and front and rear adjacent low point
Overlapping region be present between resolution PET image block.
In step s 102, according to the good low-resolution dictionary of each low resolution PET image block, training in advance and high score
Resolution dictionary, generate the sparse coefficient of each low resolution PET image block.
In embodiments of the present invention, low-resolution dictionary and high-resolution dictionary are the good excessively complete dictionary of training in advance,
The sparse coefficient of low resolution PET image block is the rarefaction representation of low resolution PET image block, low resolution PET image block pair
The high-resolution PET image block answered can be approximately the product of the sparse coefficient of high-resolution dictionary and low resolution PET image block,
Therefore the sparse coefficient of low resolution PET image block need to be solved, further to obtain high-resolution PET image block.
In embodiments of the present invention, the sparse coefficient of low resolution PET image block can be solved by following equation：
min||α||0, andWherein, F is default feature extraction function, and α is low resolution PET
Image block y sparse coefficient, D1For low-resolution dictionary, ε is predetermined threshold value.The sparse coefficient that the equations go out combines low point
Resolution dictionary can accurately represent low resolution PET image block, but the solution of the formula is a NP-hard problem, can should
Formula is equivalent to solve the process that L1 normal forms minimize：
Wherein, λ is parameter preset, for balancing α openness and α and y fidelity
Degree.In order that the solution for obtaining sparse coefficient α is more accurate, that is, to scheme by the high-resolution PET that sparse coefficient α recovers to obtain
As block and the degree of correlation of corresponding low resolution PET image block are higher, herein to formulaEnter
More accurate calculate is gone：
min||α||1, wherein, α is required to meet the first restricted coefficients of equation conditionWith the second restricted coefficients of equation
ConditionWherein, ε1For default first threshold, ε2It is current low for extracting for default Second Threshold, P
The overlapping region of resolution PET images block and upper low resolution PET image block, ω are that upper low resolution PET image block exists
The value of overlapping region.Finally, the sparse coefficient of every piece of low resolution PET image block is calculated.
In step s 103, it is low according to the sparse coefficient and high-resolution dictionary of each low resolution PET image block, generation
High-resolution PET image corresponding to resolution PET images.
In embodiments of the present invention, it is raw according to the sparse coefficient and high-resolution dictionary of each low resolution PET image block
Into high-resolution PET image corresponding to low resolution PET image, specifically, high score corresponding to every piece of low resolution PET image block
Resolution PET image can pass through formula x=D2α is calculated, wherein, x is high-resolution PET image block.
In step S104, according to low resolution PET image, high-resolution PET image, default fuzzy matrix and preset
Down-sampling matrix, generate and export the reconstruction image of low resolution PET image.
In embodiments of the present invention, forced due to employing approximation when solving the sparse coefficient of low resolution PET image block
Near mode, it may make it that the degree of accuracy of sparse coefficient is relatively low, the high-resolution PET image quality tried to achieve is bad, therefore generates
And need to optimize high-resolution PET image when exporting the reconstruction image of low resolution PET image.For the ease of describing,
High-resolution PET image is arranged to the first reconstruction image of low resolution PET image, according to low resolution PET image, high score
Resolution PET image, default fuzzy matrix, default down-sampling matrix and default gradient decline mode, rebuild and scheme to first
As optimizing, the second reconstruction image is generated, wherein, gradient declines mode and is represented by：
Xt+1=Xt+v[HTST(Y-SHXt)+c(Xt-X0)], wherein, XtFigure is rebuild for first during t suboptimization
Picture, Xt+1For the second reconstruction image during t suboptimization, v is that default gradient declines step-length, and H is fuzzy matrix, and S is arrow
Moment matrix, Y are low resolution PET image, X0For high-resolution PET image, c is parameter preset.Judging current optimization number t is
It is no to reach default largest optimization number, when reached, the second reconstruction image is exported, otherwise, the second reconstruction image is arranged to
First reconstruction image, current optimization number t is carried out plus one operates, and jump to execution according to low resolution PET image, high score
Resolution PET image, default fuzzy matrix, default down-sampling matrix and default gradient decline mode, rebuild and scheme to first
As the step of optimizing, so as to by being optimized to high-resolution PET image, be effectively improved low resolution PET figures
As the quality rebuild, and significantly reduce the computation complexity of low resolution PET image reconstruction.
In embodiments of the present invention, the good low-resolution dictionary of combined training and high-resolution dictionary construct the first coefficient about
Beam condition and the second restricted coefficients of equation condition, according to the first restricted coefficients of equation condition and the second restricted coefficients of equation condition, to low after segmentation
Sparse coefficient corresponding to resolution PET images block carries out approximate calculation, effectively improves low resolution PET image sparse coefficient
The degree of accuracy, the high-resolution PET image according to corresponding to sparse coefficient and high-resolution dictionary calculate low resolution PET image, and
High-resolution PET image is optimized by way of gradient decline, is effectively improved the picture quality of PET image reconstruction,
The computation complexity of PET image reconstruction is significantly reduced, so as to be effectively improved the reconstruction efficiency of PET image.
Embodiment two：
Fig. 2 shows the implementation process of the method for reconstructing for the PET image that the embodiment of the present invention two provides, for the ease of saying
It is bright, the part related to the embodiment of the present invention is illustrate only, details are as follows：
In step s 201, random initializtion is carried out to low-resolution dictionary and high-resolution dictionary.
In step S202, according to default low resolution PET train atlas, default high-resolution PET train atlas,
The size of image block and high-resolution PET train the size of image block in atlas in low resolution PET training atlas, to low
Resolution ratio dictionary and high-resolution dictionary carry out joint training.
In embodiments of the present invention, low-resolution dictionary and high-resolution dictionary can be carried out by gaussian random matrix with
Machine initializes, because the product of high-resolution dictionary and the sparse coefficient of low resolution PET image block can be approximately high-resolution PET
Image block, and the product of the sparse coefficient of low resolution PET image block and low-resolution dictionary is low resolution PET image block,
It can be seen that to have between high-resolution dictionary and low-resolution dictionary and contact, therefore need to be to low-resolution dictionary and high-resolution dictionary
Carry out joint training.Specifically, the formula of joint training is：
Wherein, c=1,2, X1Instructed for default low resolution PET
Practice low resolution PET training images in atlas, X2Atlas middle high-resolution PET training figures are trained for default high-resolution PET
Picture, Z are default matrix variables, and N is the image block size of low resolution PET training images, and M is high-resolution PET training figures
The image block size of picture.
In step S203, the low resolution PET image of user's input is received, low resolution PET image is split,
Generate low resolution PET image block.
In embodiments of the present invention, the low resolution that can be inputted by default image segmentation algorithm to the user received
PET image carries out the segmentation of image block, generates low resolution PET image block, the size phase of each low resolution PET image block
Together, overlapping region be present between front and rear adjacent low resolution PET image block.
In step S204, according to the good low-resolution dictionary of each low resolution PET image block, training in advance and high score
Resolution dictionary, generate the sparse coefficient of each low resolution PET image block.
In embodiments of the present invention, high-resolution PET image block can be approximately high score corresponding to low resolution PET image block
Resolution dictionary and the product of the sparse coefficient of low resolution PET image block, therefore need to be to the sparse system of low resolution PET image block
Number is solved, further to obtain high-resolution PET image block.
In embodiments of the present invention, the sparse coefficient of low resolution PET image block can be solved by following equation：
min||α||0, andWherein, F is default feature extraction function, and α is low resolution PET
Image block y sparse coefficient, D1For low-resolution dictionary, ε is predetermined threshold value.The sparse coefficient that the equations go out combines low point
Resolution dictionary can accurately represent low resolution PET image block, but the solution of the formula is a NP-hard problem, can should
Formula is equivalent to solve the process that L1 normal forms minimize：
Wherein, λ is parameter preset, for balancing α openness and α and y fidelity
Degree.In order that the solution for obtaining sparse coefficient α is more accurate, that is, to scheme by the high-resolution PET that sparse coefficient α recovers to obtain
As block and the degree of correlation of corresponding low resolution PET image block are higher, herein to formulaEnter
More accurate calculate is gone：
min||α||1, wherein, α is required to meet the first restricted coefficients of equation conditionWith the second restricted coefficients of equation
ConditionWherein, ε1For default first threshold, ε2It is current low for extracting for default Second Threshold, P
The overlapping region of resolution PET images block and upper low resolution PET image block, ω are that upper low resolution PET image block exists
The value of overlapping region.Finally, the sparse coefficient of every piece of low resolution PET image block is calculated.
It is low according to the sparse coefficient and high-resolution dictionary of each low resolution PET image block, generation in step S205
High-resolution PET image corresponding to resolution PET images.
In embodiments of the present invention, high-resolution PET image corresponding to every piece of low resolution PET image block can pass through formula
X=D2α is calculated, wherein, x is high-resolution PET image block.
In step S206, according to high-resolution PET image, the first reconstruction image of initialization low resolution PET image.
In embodiments of the present invention, forced due to employing approximation when solving the sparse coefficient of low resolution PET image block
Near mode, it may make it that the degree of accuracy of sparse coefficient is relatively low, the high-resolution PET image quality tried to achieve is bad, therefore generates
And need to optimize high-resolution PET image when exporting the reconstruction image of low resolution PET image.For the ease of describing,
High-resolution PET image is arranged to the first reconstruction image of low resolution PET image.
In step S207, according to low resolution PET image, high-resolution PET image, fuzzy matrix, down-sampling matrix
Decline mode with default gradient, the first reconstruction image is optimized, the second of generation low resolution PET image rebuilds figure
Picture.
In embodiments of the present invention, according to low resolution PET image, high-resolution PET image, default fuzzy matrix,
Default down-sampling matrix and default gradient decline mode, and the first reconstruction image is optimized, and generate the second reconstruction image,
Wherein, gradient declines mode and is represented by：
Xt+1=Xt+v[HTST(Y-SHXt)+c(Xt-X0)], wherein, XtFigure is rebuild for first during t suboptimization
Picture, Xt+1For the second reconstruction image during t suboptimization, v is that default gradient declines step-length, and H is fuzzy matrix, and S is arrow
Moment matrix, Y are low resolution PET image, X0For high-resolution PET image, c is parameter preset.
In step S208, judge whether current optimization number reaches default largest optimization number.
In embodiments of the present invention, judge whether current optimization number reaches default largest optimization number, when reached,
Step S209 is performed, otherwise performs step S2010.
In step S209, the second reconstruction image is exported.
In embodiments of the present invention, when currently optimization number reaches largest optimization number, it is believed that the second reconstruction image
Optimization quality is preferable, and the second reconstruction image can be arranged to the final reconstruction image output of low resolution PET image.
In step S210, the second reconstruction image is arranged to the first reconstruction image, current optimization number is carried out adding one
Operation.
In embodiments of the present invention, when currently optimization number is not up to largest optimization number, it is believed that also need to second
Reconstruction image is optimized, and the second reconstruction image is arranged into the first reconstruction image, and current optimization number t is carried out plus one grasps
Make, and jump to execution according to low resolution PET image, high-resolution PET image, fuzzy matrix, down-sampling matrix and preset
Gradient decline mode, the step of being optimized to the first reconstruction image.
In embodiments of the present invention, the good low-resolution dictionary of combined training and high-resolution dictionary construct the first coefficient about
Beam condition and the second restricted coefficients of equation condition, according to the first restricted coefficients of equation condition and the second restricted coefficients of equation condition, to low after segmentation
Sparse coefficient corresponding to resolution PET images block carries out approximate calculation, effectively improves low resolution PET image sparse coefficient
The degree of accuracy, the high-resolution PET image according to corresponding to sparse coefficient and high-resolution dictionary calculate low resolution PET image, and
High-resolution PET image is optimized by way of gradient decline, is effectively improved the picture quality of PET image reconstruction,
The computation complexity of PET image reconstruction is significantly reduced, so as to be effectively improved the reconstruction efficiency of PET image.
Embodiment three：
Fig. 3 shows the structure of the reconstructing device for the PET image that the embodiment of the present invention three provides, for convenience of description, only
The part related to the embodiment of the present invention is shown, including：
Image segmentation unit 31, for receiving the low resolution PET image of user's input, low resolution PET image is entered
Row segmentation, generates low resolution PET image block.
In embodiments of the present invention, the low resolution that can be inputted by default image segmentation algorithm to the user received
PET image carries out the segmentation of image block, generates multiple low resolution PET image blocks, specifically, each low resolution in segmentation
The size of PET image block is identical, and overlapping region be present between front and rear adjacent low resolution PET image block.
Coefficient generation unit 32, for according to the good low-resolution dictionary of each low resolution PET image block, training in advance
With high-resolution dictionary, the sparse coefficient of each low resolution PET image block is generated.
In embodiments of the present invention, high-resolution PET image block is approximately high-resolution corresponding to low resolution PET image block
Rate dictionary and the product of the sparse coefficient of low resolution PET image block, therefore need to be to the sparse coefficient of low resolution PET image block
Solved, further to obtain high-resolution PET image block.
In embodiments of the present invention, the sparse coefficient of low resolution PET image block can be solved by following equation：
min||α||0, andWherein, F is default feature extraction function, and α is low resolution PET
Image block y sparse coefficient, D1For low-resolution dictionary, ε is predetermined threshold value.The sparse coefficient that the equations go out combines low point
Resolution dictionary can accurately represent low resolution PET image block, but the solution of the formula is a NP-hard problem, can should
Formula is equivalent to solve the process that L1 normal forms minimize：
Wherein, λ is parameter preset, for balancing α openness and α and y fidelity
Degree.In order that the solution for obtaining sparse coefficient α is more accurate, that is, to scheme by the high-resolution PET that sparse coefficient α recovers to obtain
As block and the degree of correlation of corresponding low resolution PET image block are higher, we are to formulaCarry out
It is more accurate to calculate：
min||α||1, wherein, α is required to meet the first restricted coefficients of equation conditionWith the second restricted coefficients of equation
ConditionWherein, ε1For default first threshold, ε2It is current low for extracting for default Second Threshold, P
The overlapping region of resolution PET images block and upper low resolution PET image block, ω are that upper low resolution PET image block exists
The value of overlapping region.Finally, the sparse coefficient of every piece of low resolution PET image block is calculated.
Image generation unit 33, for the sparse coefficient and high-resolution dictionary according to each low resolution PET image block,
Generate high-resolution PET image corresponding to low resolution PET image.
In embodiments of the present invention, high-resolution PET image corresponding to every piece of low resolution PET image block can pass through formula
X=D2α is calculated, wherein, x is high-resolution PET image block.
Image output unit 34, for according to low resolution PET image, high-resolution PET image, default fuzzy matrix
With default down-sampling matrix, generate and export the reconstruction image of low resolution PET image.
In embodiments of the present invention, forced due to employing approximation when solving the sparse coefficient of low resolution PET image block
Near mode, it may make it that the degree of accuracy of sparse coefficient is relatively low, the high-resolution PET image quality tried to achieve is bad, it is therefore desirable to
High-resolution PET image is optimized.For the ease of description, high-resolution PET image is arranged to low resolution PET image
The first reconstruction image, according to low resolution PET image, high-resolution PET image, default fuzzy matrix, it is default under adopt
Sample matrix and default gradient decline mode, and the first reconstruction image is optimized, and generate the second reconstruction image, wherein, gradient
Decline mode is represented by：
Xt+1=Xt+v[HTST(Y-SHXt)+c(Xt-X0)], wherein, XtFigure is rebuild for first during t suboptimization
Picture, Xt+1For the second reconstruction image during t suboptimization, v is that default gradient declines step-length, and H is fuzzy matrix, and S is arrow
Moment matrix, Y are low resolution PET image, X0For high-resolution PET image, c is parameter preset.Judging current optimization number t is
It is no to reach default largest optimization number, when reached, the second reconstruction image is exported, otherwise, the second reconstruction image is arranged to
First reconstruction image, current optimization number t is carried out plus one operates, and jump to execution according to low resolution PET image, high score
Resolution PET image, default fuzzy matrix, default down-sampling matrix and default gradient decline mode, rebuild and scheme to first
As the step of optimizing, so as to by being optimized to high-resolution PET image, be effectively improved low resolution PET figures
As the quality rebuild, and significantly reduce the computation complexity of low resolution PET image reconstruction.
In embodiments of the present invention, the good low-resolution dictionary of combined training and high-resolution dictionary construct the first coefficient about
Beam condition and the second restricted coefficients of equation condition, according to the first restricted coefficients of equation condition and the second restricted coefficients of equation condition, to low after segmentation
Sparse coefficient corresponding to resolution PET images block carries out approximate calculation, effectively improves low resolution PET image sparse coefficient
The degree of accuracy, the high-resolution PET image according to corresponding to sparse coefficient and high-resolution dictionary calculate low resolution PET image, and
High-resolution PET image is optimized by way of gradient decline, is effectively improved the picture quality of PET image reconstruction,
The computation complexity of PET image reconstruction is significantly reduced, so as to be effectively improved the reconstruction efficiency of PET image.
Example IV：
Fig. 4 shows the structure of the reconstructing device for the PET image that the embodiment of the present invention four provides, for convenience of description, only
The part related to the embodiment of the present invention is shown, including：
Dictionary initialization unit 41, for carrying out random initializtion to low-resolution dictionary and high-resolution dictionary.
Dictionary training unit 42, for training atlas, default high-resolution PET to instruct according to default low resolution PET
Practice the chi that the size of image block and high-resolution PET in atlas, low resolution PET training atlas train image block in atlas
It is very little, joint training is carried out to low-resolution dictionary and high-resolution dictionary.
In embodiments of the present invention, low-resolution dictionary and high-resolution dictionary can be carried out by gaussian random matrix with
Machine initializes, because the product of high-resolution dictionary and the sparse coefficient of low resolution PET image block can be approximately high-resolution PET
Image block, and the product of the sparse coefficient of low resolution PET image block and low-resolution dictionary is low resolution PET image block,
It can be seen that to have between high-resolution dictionary and low-resolution dictionary and contact, therefore need to be to low-resolution dictionary and high-resolution dictionary
Carry out joint training.Specifically, the formula of joint training is：
Wherein, c=1,2, X1Instructed for default low resolution PET
Practice low resolution PET training images in atlas, X2Atlas middle high-resolution PET training figures are trained for default high-resolution PET
Picture, Z are default matrix variables, and N is the image block size of low resolution PET training images, and M is high-resolution PET training figures
The image block size of picture.
Image segmentation unit 43, for receiving the low resolution PET image of user's input, low resolution PET image is entered
Row segmentation, generates low resolution PET image block.
In embodiments of the present invention, the low resolution that can be inputted by default image segmentation algorithm to the user received
PET image carries out the segmentation of image block, generates low resolution PET image block, the size phase of each low resolution PET image block
Together, overlapping region be present between front and rear adjacent low resolution PET image block.
Coefficient generation unit 44, for according to the good low-resolution dictionary of each low resolution PET image block, training in advance
With high-resolution dictionary, the sparse coefficient of each low resolution PET image block is generated.
In embodiments of the present invention, high-resolution PET image block is approximately high-resolution corresponding to low resolution PET image block
Rate dictionary and the product of the sparse coefficient of low resolution PET image block, therefore need to be to the sparse coefficient of low resolution PET image block
Solved, further to obtain high-resolution PET image block.
In embodiments of the present invention, the sparse coefficient of low resolution PET image block can be solved by following equation：
min||α||0, andWherein, F is default feature extraction function, and α is low resolution PET
Image block y sparse coefficient, D1For low-resolution dictionary, ε is predetermined threshold value.The sparse coefficient that the equations go out combines low point
Resolution dictionary can accurately represent low resolution PET image block, but the solution of the formula is a NP-hard problem, can should
Formula is equivalent to solve the process that L1 normal forms minimize：
Wherein, λ is parameter preset, for balancing α openness and α and y fidelity
Degree.In order that the solution for obtaining sparse coefficient α is more accurate, that is, to scheme by the high-resolution PET that sparse coefficient α recovers to obtain
As block and the degree of correlation of corresponding low resolution PET image block are higher, herein to formulaEnter
More accurate calculate is gone：
min||α||1, wherein, α is required to meet the first restricted coefficients of equation conditionWith the second restricted coefficients of equation
ConditionWherein, ε1For default first threshold, ε2It is current low for extracting for default Second Threshold, P
The overlapping region of resolution PET images block and upper low resolution PET image block, ω are that upper low resolution PET image block exists
The value of overlapping region.Finally, the sparse coefficient of every piece of low resolution PET image block is calculated.
Image generation unit 45, for the sparse coefficient and high-resolution dictionary according to each low resolution PET image block,
Generate high-resolution PET image corresponding to low resolution PET image.
In embodiments of the present invention, high-resolution PET image corresponding to every piece of low resolution PET image block can pass through formula
X=D2α is calculated, wherein, x is high-resolution PET image block.
Image output unit 46, for according to low resolution PET image, high-resolution PET image, default fuzzy matrix
With default down-sampling matrix, generate and export the reconstruction image of low resolution PET image.
In embodiments of the present invention, forced due to employing approximation when solving the sparse coefficient of low resolution PET image block
Near mode, it may make it that the degree of accuracy of sparse coefficient is relatively low, the high-resolution PET image quality tried to achieve is bad, therefore generates
And need to optimize high-resolution PET image when exporting the reconstruction image of low resolution PET image.For the ease of describing,
High-resolution PET image is arranged to the first reconstruction image of low resolution PET image, according to low resolution PET image, high score
Resolution PET image, default fuzzy matrix, default down-sampling matrix and default gradient decline mode, rebuild and scheme to first
As optimizing, the second reconstruction image is generated, wherein, gradient declines mode and is represented by：
Xt+1=Xt+v[HTST(Y-SHXt)+c(Xt-X0)], wherein, XtFigure is rebuild for first during t suboptimization
Picture, Xt+1For the second reconstruction image during t suboptimization, v is that default gradient declines step-length, and H is fuzzy matrix, and S is arrow
Moment matrix, Y are low resolution PET image, X0For high-resolution PET image, c is parameter preset.Judging current optimization number t is
It is no to reach default largest optimization number, when reached, the second reconstruction image is exported, otherwise, the second reconstruction image is arranged to
First reconstruction image, current optimization number t is carried out plus one operates, and jump to execution according to low resolution PET image, high score
Resolution PET image, default fuzzy matrix, default down-sampling matrix and default gradient decline mode, rebuild and scheme to first
As the step of optimizing, so as to by being optimized to high-resolution PET image, be effectively improved low resolution PET figures
As the quality rebuild, and significantly reduce the computation complexity of low resolution PET image reconstruction.
Preferably, coefficient generation unit 44 includes the first constraint construction unit 441, second constraint construction unit 442 and is
Number computing unit 443, wherein：
First constraint construction unit 441, for according to low resolution PET image block, low-resolution dictionary, default feature
Function and preset first threshold value are extracted, builds the first restricted coefficients of equation condition；
Second constraint construction unit 442, for according to low resolution PET image block and low resolution PET image block
Overlapping region, high-resolution dictionary and the default Second Threshold of one low resolution PET image block, build the second restricted coefficients of equation
Condition；And
Coefficient calculation unit 443, for according to default coefficient formulas, calculate the first restricted coefficients of equation condition that meets and
Second restricted coefficients of equation condition, low resolution PET image block sparse coefficient.
Preferably, image output unit 46 includes reconstruction image initialization unit 461, reconstruction image optimizes unit 462, again
Build judging unit 463 and rebuild output unit 464, wherein：
Reconstruction image initialization unit 461, for according to high-resolution PET image, initializing low resolution PET image
First reconstruction image；
Reconstruction image optimize unit 462, for according to low resolution PET image, high-resolution PET image, fuzzy matrix,
Down-sampling matrix and default gradient decline mode, and the first reconstruction image is optimized, and the of generation low resolution PET image
Two reconstruction images；
Judging unit 463 is rebuild, for judging currently to optimize whether number reaches default largest optimization number；And
Output unit 464 is rebuild, for when currently optimization number reaches largest optimization number, figure to be rebuild in output second
Picture, otherwise, the second reconstruction image is arranged to the first reconstruction image, current optimization number is carried out plus one operates, and by rebuilding
Image optimization unit 462 performs the operation optimized to the first reconstruction image.
In embodiments of the present invention, the good low-resolution dictionary of combined training and high-resolution dictionary construct the first coefficient about
Beam condition and the second restricted coefficients of equation condition, according to the first restricted coefficients of equation condition and the second restricted coefficients of equation condition, to low after segmentation
Sparse coefficient corresponding to resolution PET images block carries out approximate calculation, effectively improves low resolution PET image sparse coefficient
The degree of accuracy, the high-resolution PET image according to corresponding to sparse coefficient and high-resolution dictionary calculate low resolution PET image, and
High-resolution PET image is optimized by way of gradient decline, is effectively improved the picture quality of PET image reconstruction,
The computation complexity of PET image reconstruction is significantly reduced, so as to be effectively improved the reconstruction efficiency of PET image.
In embodiments of the present invention, each unit of equipment for reconstructing image can be realized by corresponding hardware or software unit, respectively
Unit can be independent soft and hardware unit, can also be integrated into a soft and hardware unit, herein not limiting the present invention.
Embodiment five：
Fig. 5 shows the structure for the Medical Image Processing equipment that the embodiment of the present invention five provides, and for convenience of description, only shows
The part related to the embodiment of the present invention is gone out.
The Medical Image Processing equipment 5 of the embodiment of the present invention includes processor 50, memory 51 and is stored in memory
In 51 and the computer program 52 that can be run on processor 50.The processor 50 is realized above-mentioned each when performing computer program 52
Step in individual embodiment of the method, such as the step S101 to S104 shown in Fig. 1.Or processor 50 performs computer program
The function of each unit in above-mentioned each device embodiment, such as the function of unit 31 to 34 shown in Fig. 3 are realized when 52.
In embodiments of the present invention, according to the low resolution PET image block after segmentation, the low-resolution dictionary trained and
High-resolution dictionary, the sparse coefficient of each low resolution PET image block is calculated, according to these sparse coefficients and high-resolution word
High-resolution PET image corresponding to allusion quotation calculating low resolution PET image, by being post-processed to high-resolution PET image, is obtained
To low resolution PET reconstruction image, so as to be effectively improved the picture quality of PET image reconstruction, PET is significantly reduced
Computation complexity in image reconstruction, and then it is effectively improved the reconstruction efficiency of PET image.
Embodiment six：
In embodiments of the present invention, there is provided a kind of computer-readable recording medium, the computer-readable recording medium are deposited
Computer program is contained, the computer program realizes the step in above-mentioned each embodiment of the method when being executed by processor, for example,
Step S101 to S104 shown in Fig. 1.Or the computer program is realized in above-mentioned each device embodiment when being executed by processor
The function of each unit, such as the function of unit 31 to 34 shown in Fig. 3.
In embodiments of the present invention, according to the low resolution PET image block after segmentation, the low-resolution dictionary trained and
High-resolution dictionary, the sparse coefficient of each low resolution PET image block is calculated, according to these sparse coefficients and high-resolution word
High-resolution PET image corresponding to allusion quotation calculating low resolution PET image, by being post-processed to high-resolution PET image, is obtained
To low resolution PET reconstruction image, so as to be effectively improved the picture quality of PET image reconstruction, PET is significantly reduced
Computation complexity in image reconstruction, and then it is effectively improved the reconstruction efficiency of PET image.
The computer-readable recording medium of the embodiment of the present invention can include that any of computer program code can be carried
Entity or device, recording medium, for example, the memory such as ROM/RAM, disk, CD, flash memory.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
All any modification, equivalent and improvement made within refreshing and principle etc., should be included in the scope of the protection.
Claims (10)
1. a kind of method for reconstructing of PET image, it is characterised in that methods described comprises the steps：
The low resolution PET image of user's input is received, the low resolution PET image is split, generates low resolution
PET image block；
According to the good low-resolution dictionary of each low resolution PET image block, training in advance and high-resolution dictionary, generation
The sparse coefficient of each low resolution PET image block；
According to the sparse coefficient of each low resolution PET image block and the high-resolution dictionary, the low resolution is generated
High-resolution PET image corresponding to rate PET image；
According to the low resolution PET image, the high-resolution PET image, default fuzzy matrix and default down-sampling
Matrix, generate and export the reconstruction image of the low resolution PET image.
2. the method as described in claim 1, it is characterised in that according to each low resolution PET image block, training in advance
Good low-resolution dictionary and high-resolution dictionary, the step of generating the sparse coefficient of each low resolution PET image block,
Including：
According to the low resolution PET image block, the low-resolution dictionary, default feature extraction function and default first threshold
Value, build the first restricted coefficients of equation condition；
According to the low resolution PET image block and the upper low resolution PET image block of the low resolution PET image block
Overlapping region, the high-resolution dictionary and default Second Threshold, build the second restricted coefficients of equation condition；
According to default coefficient formulas, calculating meets the first restricted coefficients of equation condition and the second restricted coefficients of equation condition
, the sparse coefficient of the low resolution PET image block.
3. the method as described in claim 1, it is characterised in that according to the low resolution PET image, the high-resolution
PET image, default fuzzy matrix and default down-sampling matrix, generate and export the reconstruction of the low resolution PET image
The step of image, including：
According to the high-resolution PET image, the first reconstruction image of the low resolution PET image is initialized；
According to the low resolution PET image, the high-resolution PET image, the fuzzy matrix, the down-sampling matrix and
Default gradient declines mode, and first reconstruction image is optimized, generates the second weight of the low resolution PET image
Build image；
Judge whether current optimization number reaches default largest optimization number；
When the current optimization number reaches the largest optimization number, second reconstruction image is exported, otherwise, by described in
Second reconstruction image is arranged to first reconstruction image, and the current optimization number is carried out plus one operates, and jumps to and holds
The step of row optimizes to first reconstruction image.
4. the method as described in claim 1, it is characterised in that receive user input low resolution PET image the step of it
Before, methods described also includes：
Random initializtion is carried out to the low-resolution dictionary and the high-resolution dictionary；
Atlas, default high-resolution PET training atlas, low resolution PET instructions are trained according to default low resolution PET
Practice the size that the size of image block and the high-resolution PET in atlas train image block in atlas, to the low resolution
Dictionary and the high-resolution dictionary carry out joint training.
5. a kind of reconstructing device of PET image, it is characterised in that described device includes：
Image segmentation unit, for receiving the low resolution PET image of user's input, the low resolution PET image is carried out
Segmentation, generate low resolution PET image block；
Coefficient generation unit, for according to the good low-resolution dictionary of each low resolution PET image block, training in advance and
High-resolution dictionary, generate the sparse coefficient of each low resolution PET image block；
Image generation unit, for the sparse coefficient according to each low resolution PET image block and the high-resolution word
Allusion quotation, generate high-resolution PET image corresponding to the low resolution PET image；And
Image output unit, for according to the low resolution PET image, the high-resolution PET image, default fuzzy square
Battle array and default down-sampling matrix, generate and export the reconstruction image of the low resolution PET image.
6. device as claimed in claim 5, it is characterised in that the coefficient generation unit includes：
First constraint construction unit, for according to the low resolution PET image block, the low-resolution dictionary, default spy
Sign extraction function and preset first threshold value, build the first restricted coefficients of equation condition；
Second constraint construction unit, for according to the low resolution PET image block and the low resolution PET image block
The overlapping region of one low resolution PET image block, the high-resolution dictionary and default Second Threshold, build the second coefficient
Constraints；And
Coefficient calculation unit, for meeting the first restricted coefficients of equation condition and institute according to default coefficient formulas, calculating
State the second restricted coefficients of equation condition, the low resolution PET image block sparse coefficient.
7. device as claimed in claim 5, it is characterised in that described image output unit includes：
Reconstruction image initialization unit, for according to the high-resolution PET image, initializing the low resolution PET image
The first reconstruction image；
Reconstruction image optimizes unit, for according to the low resolution PET image, the high-resolution PET image, described fuzzy
Matrix, the down-sampling matrix and default gradient decline mode, and first reconstruction image is optimized, and generation is described low
Second reconstruction image of resolution PET images；
Judging unit is rebuild, for judging currently to optimize whether number reaches default largest optimization number；And
Output unit is rebuild, for when the current optimization number reaches the largest optimization number, exporting second weight
Image is built, otherwise, second reconstruction image is arranged to first reconstruction image, the current optimization number is added
One operation, and the operation optimized to first reconstruction image is performed by reconstruction image optimization unit.
8. device as claimed in claim 5, it is characterised in that described device also includes：
Dictionary initialization unit, for carrying out random initializtion to the low-resolution dictionary and the high-resolution dictionary；With
And
Dictionary training unit, for according to default low resolution PET train atlas, default high-resolution PET train atlas,
The size of image block and the high-resolution PET train the chi of image block in atlas in the low resolution PET training atlas
It is very little, joint training is carried out to the low-resolution dictionary and the high-resolution dictionary.
9. a kind of Medical Image Processing equipment, including memory, processor and it is stored in the memory and can be described
The computer program run on processor, it is characterised in that realize such as right described in the computing device during computer program
It is required that the step of any one of 1 to 4 methods described.
10. a kind of computer-readable recording medium, the computer-readable recording medium storage has computer program, and its feature exists
In when the computer program is executed by processor the step of realization such as any one of Claims 1-4 methods described.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710821065.6A CN107680146A (en) | 2017-09-13 | 2017-09-13 | Method for reconstructing, device, equipment and the storage medium of PET image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710821065.6A CN107680146A (en) | 2017-09-13 | 2017-09-13 | Method for reconstructing, device, equipment and the storage medium of PET image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107680146A true CN107680146A (en) | 2018-02-09 |
Family
ID=61135897
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710821065.6A Pending CN107680146A (en) | 2017-09-13 | 2017-09-13 | Method for reconstructing, device, equipment and the storage medium of PET image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107680146A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109741255A (en) * | 2018-12-12 | 2019-05-10 | 深圳先进技术研究院 | PET image super-resolution reconstruction method, device, equipment and medium based on decision tree |
| WO2020118844A1 (en) * | 2018-12-14 | 2020-06-18 | 深圳先进技术研究院 | Reconstruction method for pet image, computer storage medium, and computer device |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105844590A (en) * | 2016-03-23 | 2016-08-10 | 武汉理工大学 | Image super-resolution reconstruction method and system based on sparse representation |
| CN106227015A (en) * | 2016-07-11 | 2016-12-14 | 中国科学院深圳先进技术研究院 | A kind of hologram image super-resolution reconstruction method and system based on compressive sensing theory |
| CN106491151A (en) * | 2016-01-25 | 2017-03-15 | 上海联影医疗科技有限公司 | PET image acquisition methods and system |
| 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 |
| CN106846430A (en) * | 2014-11-21 | 2017-06-13 | 上海联影医疗科技有限公司 | A kind of image rebuilding method |
| CN106934766A (en) * | 2017-03-15 | 2017-07-07 | 西安理工大学 | A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation |
| CN106952317A (en) * | 2017-03-23 | 2017-07-14 | 西安电子科技大学 | Based on the high spectrum image method for reconstructing that structure is sparse |
-
2017
- 2017-09-13 CN CN201710821065.6A patent/CN107680146A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106846430A (en) * | 2014-11-21 | 2017-06-13 | 上海联影医疗科技有限公司 | A kind of image rebuilding method |
| CN106491151A (en) * | 2016-01-25 | 2017-03-15 | 上海联影医疗科技有限公司 | PET image acquisition methods and system |
| CN105844590A (en) * | 2016-03-23 | 2016-08-10 | 武汉理工大学 | Image super-resolution reconstruction method and system based on sparse representation |
| CN106227015A (en) * | 2016-07-11 | 2016-12-14 | 中国科学院深圳先进技术研究院 | A kind of hologram image super-resolution reconstruction method and system based on compressive sensing theory |
| 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 |
| CN106934766A (en) * | 2017-03-15 | 2017-07-07 | 西安理工大学 | A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation |
| CN106952317A (en) * | 2017-03-23 | 2017-07-14 | 西安电子科技大学 | Based on the high spectrum image method for reconstructing that structure is sparse |
Non-Patent Citations (1)
| Title |
|---|
| 郭坦: "基于稀疏表示的超分辨率图像重建算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109741255A (en) * | 2018-12-12 | 2019-05-10 | 深圳先进技术研究院 | PET image super-resolution reconstruction method, device, equipment and medium based on decision tree |
| WO2020118844A1 (en) * | 2018-12-14 | 2020-06-18 | 深圳先进技术研究院 | Reconstruction method for pet image, computer storage medium, and computer device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443867B (en) | CT image super-resolution reconstruction method based on generation countermeasure network | |
| CN108460726B (en) | Magnetic resonance image super-resolution reconstruction method based on enhanced recursive residual network | |
| CN111445390B (en) | Wide residual attention-based three-dimensional medical image super-resolution reconstruction method | |
| US20220092742A1 (en) | Learning Method of Generative Adversarial Network with Multiple Generators for Image Denoising | |
| CN113012172B (en) | AS-UNet-based medical image segmentation method and system | |
| WO2021027571A1 (en) | Artificial intelligence-based medical image processing method, medical device and storage medium | |
| Han et al. | Flow field reduction via reconstructing vector data from 3-D streamlines using deep learning | |
| CN110097611A (en) | Image rebuilding method, device, equipment and storage medium | |
| Ziabari et al. | 2.5 D deep learning for CT image reconstruction using a multi-GPU implementation | |
| CN113516659A (en) | Medical image automatic segmentation method based on deep learning | |
| CN112070752B (en) | Auricle segmentation method and device for medical image and storage medium | |
| CN116682120A (en) | Multilingual mosaic image text recognition method based on deep learning | |
| CN109215014A (en) | Training method, device, equipment and the storage medium of CT image prediction model | |
| CN111373439A (en) | Method for image segmentation using CNN | |
| CN111915626B (en) | Automatic segmentation method, device and storage medium for heart ultrasonic image ventricular region | |
| WO2024051018A1 (en) | Pet parameter image enhancement method and apparatus, device, and storage medium | |
| Xie et al. | Artifact removal using GAN network for limited-angle CT reconstruction | |
| CN113870327A (en) | Medical image registration method based on multi-level deformation field prediction | |
| CN107680146A (en) | Method for reconstructing, device, equipment and the storage medium of PET image | |
| US20210074034A1 (en) | Methods and apparatus for neural network based image reconstruction | |
| CN116894783A (en) | Metal artifact removal method for countermeasure generation network model based on time-varying constraint | |
| CN116912344A (en) | List mode TOF-PET reconstruction method based on original-dual network | |
| CN110782396A (en) | Light-weight image super-resolution reconstruction network and reconstruction method | |
| CN112634224B (en) | Focus detection method and device based on target image | |
| CN115330600A (en) | Lung CT image super-resolution method based on improved SRGAN |
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 |