CN109741255A - PET image super-resolution reconstruction method, device, equipment and medium based on decision tree - Google Patents
PET image super-resolution reconstruction method, device, equipment and medium based on decision tree Download PDFInfo
- Publication number
- CN109741255A CN109741255A CN201811519205.5A CN201811519205A CN109741255A CN 109741255 A CN109741255 A CN 109741255A CN 201811519205 A CN201811519205 A CN 201811519205A CN 109741255 A CN109741255 A CN 109741255A
- Authority
- CN
- China
- Prior art keywords
- image block
- decision tree
- image
- pet image
- 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
- 238000003066 decision tree Methods 0.000 title claims abstract description 141
- 238000000034 method Methods 0.000 title claims abstract description 39
- 239000011159 matrix material Substances 0.000 claims abstract description 44
- 238000013507 mapping Methods 0.000 claims abstract description 43
- 238000007637 random forest analysis Methods 0.000 claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims description 104
- 230000006870 function Effects 0.000 claims description 23
- 238000004590 computer program Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 11
- 230000015654 memory Effects 0.000 claims description 8
- 239000000203 mixture Substances 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 238000003745 diagnosis Methods 0.000 abstract description 8
- 238000005516 engineering process Methods 0.000 abstract description 4
- 238000002059 diagnostic imaging Methods 0.000 abstract description 2
- 238000012636 positron electron tomography Methods 0.000 description 144
- 239000000284 extract Substances 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000012879 PET imaging Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000002595 magnetic resonance imaging Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000002591 computed tomography Methods 0.000 description 2
- 230000007850 degeneration Effects 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 238000002600 positron emission tomography Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001727 in vivo Methods 0.000 description 1
- 230000002503 metabolic effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 229940121896 radiopharmaceutical Drugs 0.000 description 1
- 239000012217 radiopharmaceutical Substances 0.000 description 1
- 230000002799 radiopharmaceutical effect Effects 0.000 description 1
- 238000003325 tomography Methods 0.000 description 1
Classifications
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention is applicable in medical imaging technology field, provide a kind of PET image super-resolution reconstruction method based on decision tree, device, equipment and medium, this method comprises: the target PET image to pending super-resolution rebuilding carries out image block extraction, obtain corresponding PET image block, image block similar features cluster is carried out to PET image block by every decision tree in random forest, obtain what PET image block decision in every decision tree went out, corresponding image block mapping matrix, gone out according to every decision tree decision, corresponding image block mapping matrix predicts the corresponding high-definition picture block of PET image block, the corresponding high-resolution PET image of target PET image is reconstructed according to high-definition picture block, to improve the accuracy for carrying out super-resolution rebuilding to target PET image, it mentions simultaneously It is high to rebuild the resolution ratio and clarity of obtained high-resolution PET image, and then improve the accuracy and accuracy of the clinic diagnosis based on PET image.
Description
Technical field
The invention belongs to medical imaging technology field more particularly to a kind of PET image Super-resolution Reconstructions based on decision tree
Method, apparatus, equipment and medium.
Background technique
Positron emission tomography (Positron Emission Tomography, abbreviation PET) is a kind of emission type
Imaging technique (Emission Tomography, abbreviation ET), it is shown not by the method that radiopharmaceutical is injected in vivo
With the metabolic situation of tissue.PET technology be after computer tomography (Computed Tomography, abbreviation CT) and
It is applied to a kind of clinical New video skill after magnetic resonance imaging (Magnetic Resonance Imaging, abbreviation MRI)
Art, clinical value are unquestionable.The diagnosis accuracy of doctor can be improved in the PET image of high quality, and measures image
One important indicator of quality is image resolution ratio, therefore, in field of image processing, improves image resolution ratio into industry one
A important goal.
Image resolution ratio describes the pixel number that image includes, and more pixels mean the information content that image includes
Bigger, image can carry more scene details, and such image has higher application value.In the 1960s,
The concept of super-resolution (Super-Resolution) reconstruction technique first appears, these technologies are advocated from piece image or sequence
Restore high-definition picture in column picture frame, obtains developing rapidly and widely under the actively promoting of Tasi and Huang.It is curing
Imaging field is learned, such as PET imaging, super-resolution rebuilding algorithm can make up the limitation of PET physics imaging device, REINFORCED PET
Image detail information, improves the quality of medical image, to assist doctor to be preferably detected and analyzed to the state of an illness, to make
Correctly diagnosis has extensive important application value.
However, in existing super-resolution rebuilding algorithm, the method based on reconstruction be by be added priori knowledge come
To high-definition picture, the image resolution ratio that such methods are rebuild is improved less obvious, especially for feature structure
Inconsistent PET image, and sample set needed for the method based on study is excessive, reconstruction takes a long time, and also needs manually to divide
Image pattern collection is clustered, cluster feature is also not quite identical, and therefore, the effect for the PET image rebuild is also not satisfactory.
Summary of the invention
The PET image super-resolution reconstruction method that the purpose of the present invention is to provide a kind of based on decision tree, device, equipment and
Medium, it is intended to solve that a kind of effective PET image super-resolution reconstruction method can not be provided due to the prior art, reconstruction is caused to obtain
PET image the bad problem of picture quality.
On the one hand, the PET image super-resolution reconstruction method based on decision tree that the present invention provides a kind of, the method includes
Following step:
When the super-resolution rebuilding for receiving target PET image is requested, target PET image progress image block is mentioned
It takes, obtains corresponding PET image block；
The similar spy of image block is carried out to the PET image block by every decision tree in preparatory trained random forest
Sign cluster obtains image block mapping matrix that PET image block decision in every decision tree goes out, corresponding, wherein
The random forest is made of more independent decision trees；
Described image block mapping matrix gone out according to every decision tree decision, corresponding predicts the PET image block
Corresponding high-definition picture block；
The corresponding high-resolution PET image of the target PET image is reconstructed according to the high-definition picture block.
On the other hand, the present invention provides a kind of PET image Super-resolution Reconstruction device based on decision tree, described device packet
It includes:
Image block extraction unit, for when receive target PET image super-resolution rebuilding request when, by the target
PET image carries out image block extraction, obtains corresponding PET image block；
Mapping matrix obtaining unit, for passing through every decision tree in preparatory trained random forest to the PET
Image block carries out image block similar features cluster, obtains that PET image block decision in every decision tree goes out, corresponding
Image block mapping matrix, wherein the random forest is made of more independent decision trees；
Image block predicting unit maps square for described image block go out according to every decision tree decision, corresponding
Battle array predicts the corresponding high-definition picture block of the PET image block；And
Image reconstruction unit, for reconstructing the corresponding height of the target PET image according to the high-definition picture block
Resolution PET images.
On the other hand, the present invention also provides a kind of image processing equipment, including memory, processor and it is stored in institute
The computer program that can be run in memory and on the processor is stated, the processor executes real when the computer program
The now step as described in the PET image super-resolution reconstruction method above-mentioned based on decision tree.
On the other hand, the present invention also provides a kind of computer readable storage medium, the computer readable storage mediums
It is stored with computer program, is realized when the computer program is executed by processor as the above-mentioned PET image based on decision tree is super
Step described in resolved reconstruction method.
The present invention carries out image block extraction to the target PET image of pending super-resolution rebuilding, obtains corresponding PET figure
As block, image block similar features cluster is carried out to PET image block by every decision tree in random forest, obtains PET image
Image block mapping matrix that block decision in every decision tree goes out, corresponding, figure gone out according to every decision tree decision, corresponding
As the corresponding high-definition picture block of block mapping matrix prediction PET image block, target PET is reconstructed according to high-definition picture block
The corresponding high-resolution PET image of image, to improve the accuracy for carrying out super-resolution rebuilding to target PET image, together
When improve the resolution ratio and clarity of high-resolution PET image rebuild and obtained, and then improve the clinic based on PET image
The accuracy of diagnosis and treatment.
Detailed description of the invention
Fig. 1 is the implementation process for the PET image super-resolution reconstruction method based on decision tree that the embodiment of the present invention one provides
Figure；
Fig. 2 is the implementation process of the PET image super-resolution reconstruction method provided by Embodiment 2 of the present invention based on decision tree
Figure；
Fig. 3 is the structural representation for the PET image Super-resolution Reconstruction device based on decision tree that the embodiment of the present invention three provides
Figure；
Fig. 4 is the structural representation for the PET image Super-resolution Reconstruction device based on decision tree that the embodiment of the present invention four provides
Figure；And
Fig. 5 is the structural schematic diagram for the image processing equipment that the embodiment of the present invention five provides.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
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.
Specific implementation of the invention is described in detail below in conjunction with specific embodiment:
Embodiment one:
Fig. 1 shows the realization of the PET image super-resolution reconstruction method based on decision tree of the offer of the embodiment of the present invention one
Process, for ease of description, only parts related to embodiments of the present invention are shown, and details are as follows:
In step s101, when receive target PET image super-resolution rebuilding request when, by target PET image into
Row image block extracts, and obtains corresponding PET image block.
The embodiment of the present invention is suitable for Medical Image Processing platform, system or equipment, such as personal computer, server
Deng.Target PET image can be by PET imaging device collection in worksite or download from the medical information database pre-established
, noise more or less can all occur in the image of PET imaging device acquisition, target PET image quality is influenced, and network downloading is difficult
Exempt from that original PET image can be compressed with downscaled images volume, so that causing the image resolution ratio of target PET image reduces,
In order to not influence diagnosis and treatment, need to carry out super-resolution rebuilding to target PET image.When the super-resolution for receiving target PET image
When rate reconstruction request, according to the picture size of target PET image and preset image number of blocks, image block is carried out to PET image
It extracts, obtains corresponding quantity PET image block.
In step s 102, figure is carried out to PET image block by every decision tree in preparatory trained random forest
As block similar features cluster, image block mapping matrix that PET image block decision in every decision tree goes out, corresponding is obtained.
In embodiments of the present invention, PET image block is separately input to every decision of preparatory trained random forest
Image block similar features cluster is carried out in tree, every decision tree goes out corresponding image block mapping matrix according to PET image block decision,
I.e. since the root node of decision tree, node PET image block being recursively assigned in decision tree, until reaching the decision tree
Leaf node, the corresponding image block mapping matrix of the leaf node be PET image block in the decision tree decision go out figure
As block mapping matrix.Wherein, random forest is made of more independent decision trees, and decision tree is made of each node, including root section
Point, non-leaf nodes and leaf node, non-leaf nodes include left and right child node again.
In embodiments of the present invention, it is preferable that set 10 for the quantity of the decision tree in random forest, thus improve with
The accuracy of the machine forest result of decision.
In step s 103, image block mapping matrix gone out according to every decision tree decision, corresponding predicts PET image block
Corresponding high-definition picture block.
In embodiments of the present invention, according to preset leaf node mapping model ml(y)=Wlφ (y) and every decision
Image block mapping matrix that tree decision goes out, corresponding predicts PET image block, obtains PET image block in every decision tree
Prediction result (prediction result is initial high-resolution image block) that middle prediction obtains, corresponding, further according to modelThe prediction result of comprehensive all decision trees, i.e., be averaged the prediction result of all decision trees, obtain most
Whole high-definition picture blockWherein, WlFor the image block mapping of PET image block sort first of leaf node into decision tree
Matrix, φ (y)=y are preset basic function, and y is PET image block, and T is the quantity of decision tree in random forest, and l (t) is PET
Image block y is split off first of leaf node to the t decision tree,It is the high-definition picture block that prediction obtains.
In step S104, the corresponding high-resolution PET of target PET image is reconstructed according to high-definition picture block and is schemed
Picture.
In embodiments of the present invention, all high-definition picture blocks prediction obtainedIt is stacked, so that image dimension
Become larger, and then reconstructs the corresponding high-resolution PET image of target PET image.
In embodiments of the present invention, image block extraction is carried out to the target PET image of pending super-resolution rebuilding, obtained
Corresponding PET image block carries out image block similar features cluster to PET image block by every decision tree in random forest,
Image block mapping matrix that PET image block decision in every decision tree goes out, corresponding is obtained, is gone out according to every decision tree decision
, the corresponding high-definition picture block of corresponding image block mapping matrix prediction PET image block, according to high-definition picture block weight
The corresponding high-resolution PET image of target PET image is built out, super-resolution rebuilding is carried out to target PET image to improve
Accuracy, while improving the resolution ratio and clarity of high-resolution PET image rebuild and obtained, and then improve and be based on
The accuracy of the clinic diagnosis of PET image.
Embodiment two:
The realization for the PET image super-resolution reconstruction method based on decision tree that Fig. 2 shows provided by Embodiment 2 of the present invention
Process, for ease of description, only parts related to embodiments of the present invention are shown, and details are as follows:
In step s 201, according to preset training sample and preset division function since the root node of decision tree into
The division of row recurrence is determined until terminating the building of decision tree when the depth of division reaches preset decision tree depth threshold with realizing
The image block similar features of plan tree cluster training.
In embodiments of the present invention, since first node (i.e. root node) of decision tree, according to preset division letter
The sample for meeting left division result in training sample is assigned in left child node by several division results, meets right division result
Sample is assigned in right child node, then again respectively to the corresponding left and right child node sample of left and right child node according to division function
Division result is divided again, successively carries out recurrence division according to this, until the corresponding sub- training sample of child node is inseparable, i.e.,
When generation leaf node or the depth of division reach preset decision tree depth threshold, decision tree just stops growing, that is, terminates
The building of decision tree, while the left and right child node of recurrence division divides and clusters all training samples, namely completes decision tree
Image block similar features cluster training.All decision tree training process are independent mutually in random forest, to improve by certainly
The stability of the Random Forest model of plan tree composition.The training sample of decision tree is taken out using resampling mode in random forest
It takes, i.e. decision tree the case where there are repeated samples, the quantity of the training sample of all decision trees is identical, and training sample is
The instruction being made of high-resolution training image blocks and low resolution training image blocks corresponding with the high-resolution training image blocks
Practice image block to composition, to improve the training time of decision tree.
Recurrence division is being carried out since the root node of decision tree according to preset training sample and preset division function
Before, it is preferable that PET high-resolution training image is carried out by degeneration processing by preset image degeneration formula, is corresponded to
PET low resolution training image, image is carried out respectively to PET high-resolution training image and PET low resolution training image
Block extracts, and corresponding high-resolution training image blocks and low resolution training image blocks is obtained, by high-resolution training image blocks
Set is set as with the training image blocks of low resolution training image blocks corresponding with high-resolution training image blocks composition
Training sample, to improve the accuracy of follow-up decision tree prediction result.
In embodiments of the present invention, it is preferable that dividing function isDivide function
The low resolution training image blocks y that value is 0 is split off left child node, the corresponding high-resolution instruction of low resolution training image blocks
Practice image block and be also split off left child node, is expressed asOtherwise it is split off right sub- section
Point, is expressed asWherein, rθ(y)=y [θ]-θth, θ is that low resolution is instructed in training sample
Practice the feature of image block y, rθ(y) be feature θ receptance function, y [θ] is the one-dimensional vector of the eigenmatrix of y, θthIt is default threshold
Value, { H, L } are expressed as the training figure of high-resolution training image blocks H and corresponding low resolution training image blocks L composition
As block pair, to improve the similitude of the image block characteristics clustered in training sample.
Recurrence division is being carried out since the root node of decision tree according to preset training sample and preset division function
When, it is preferable that recurrence division is realized by following step:
1) the low resolution training image in the corresponding present node training sample of the present node to be divided to decision tree
Block carries out image block characteristics extraction, until the image block characteristics extracted meet preset Feature Selection formula, chooses image block
Feature is as node split feature, wherein present node training sample belongs to training sample.
In embodiments of the present invention, first corresponding to present node when needing the present node to decision tree to divide
The image block characteristics of present node training sample carry out traversal extraction, until the image block characteristics extracted meet preset spy
Sign chooses formula, then chooses the image block characteristics as node split feature, wherein present node training sample belongs to trained sample
This.
In embodiments of the present invention, it is preferable that Feature Selection formula is
It is optimal characteristics to choose the feature θ that formula judgement is extracted according to this feature, i.e., is made to calculate error according to this feature θ
It is minimum, wherein It is n-th of training sample ynMean value (i.e.
Pixel mean value on the same spatial position), k is pre-set hyper parameter, and N is the quantity of training sample, XC、YCIt splits into
The high-resolution training image blocks and low resolution training image blocks of decision tree child nodes, m (yn) it is decision tree to ynIt carries out pre-
The prediction result of survey, xnIt is training sample y to be reconstructednCorresponding high-definition picture block, Le are left child node, and Ri is right son
Node.
2) left and right child node division is carried out to present node training sample according to node split feature and division function.
In embodiments of the present invention, division function is according to selection, optimal node split feature to present node training
Sample carries out left and right child node division.
The division for 1) and 2) realizing decision tree present node through the above steps, to improve the figure clustered in training sample
As the similitude of block feature.
Low resolution training image in the present node corresponding present node training sample to be divided to decision tree
Block carried out before the step of image block characteristics extraction, it is preferable that when the depth where present node reaches preset decision tree depth
The quantity for spending low resolution training image blocks in threshold value or present node training sample is less than preset image block amount threshold
When (for example, each one piece of high-resolution and low-resolution image block), stopping carries out left and right child node division to present node training sample, will
Present node is set as the leaf node of decision tree, and learns image block mapping matrix by present node training sample, thus
Reduce the probability of decision tree over-fitting.
In embodiments of the present invention, it is preferable that 15 are set by decision tree depth threshold, to reduce decision tree over-fitting
Probability.
In the leaf node for setting present node to decision tree, and image block is learnt by present node training sample and is reflected
When penetrating matrix, it is preferable that in each leaf node of decision tree, according to the corresponding present node training sample of leaf node,
Use formula WlT=(φ (Y)Tφ(Y)+λI)-1φ(Y)TY learns the corresponding image block mapping matrix of leaf node, wherein
WlFor the corresponding image block mapping matrix of leaf node, I is preset unit matrix, λ be preset regularization parameter (for example,
0.01), φ (Y) is preset basic function, and Y is low resolution training image blocks, so that the image block mapping square learnt
Battle array can more accurately reflect the mapping relations between high-low resolution training image blocks.
In step S202, when receive target PET image super-resolution rebuilding request when, by target PET image into
Row image block extracts, and obtains corresponding PET image block.
In step S203, figure is carried out to PET image block by every decision tree in preparatory trained random forest
As block similar features cluster, image block mapping matrix that PET image block decision in every decision tree goes out, corresponding is obtained.
In step S204, image block mapping matrix gone out according to every decision tree decision, corresponding predicts PET image block
Corresponding high-definition picture block.
In step S205, the corresponding high-resolution PET of target PET image is reconstructed according to high-definition picture block and is schemed
Picture.
In embodiments of the present invention, the step of specific embodiment of step S202- step S205 can refer to embodiment one
The description of S101- step S104, details are not described herein.
In embodiments of the present invention, firstly, according to preset training sample and preset division root of the function from decision tree
Node starts to carry out recurrence division, until the depth of division reaches preset decision tree depth threshold, to complete the structure of decision tree
It builds, realizes the image block similar features cluster training of decision tree, target PET image is carried out further according to trained random forest
Super-resolution rebuilding obtains corresponding high-resolution PET image, to improve the training time of decision tree, and improves
By the stability for the Random Forest model that decision tree forms, and then improve the resolution for rebuilding obtained high-resolution PET image
Rate and clarity.
Embodiment three:
Fig. 3 shows the structure of the PET image Super-resolution Reconstruction device based on decision tree of the offer of the embodiment of the present invention three,
For ease of description, only parts related to embodiments of the present invention are shown, including:
Image block extraction unit 31, for when receive target PET image super-resolution rebuilding request when, by target
PET image carries out image block extraction, obtains corresponding PET image block；
Mapping matrix obtaining unit 32, for being schemed by every decision tree in preparatory trained random forest to PET
As block progress image block similar features cluster, obtains image block that PET image block decision in every decision tree goes out, corresponding and reflect
Penetrate matrix；
Image block predicting unit 33 is predicted for image block mapping matrix gone out according to every decision tree decision, corresponding
The corresponding high-definition picture block of PET image block；And
Image reconstruction unit 34, for reconstructing the corresponding high-resolution of target PET image according to high-definition picture block
PET image.
In embodiments of the present invention, each unit of the PET image Super-resolution Reconstruction device based on decision tree can be by corresponding
Hardware or software unit realize that each unit can be independent soft and hardware unit, also can integrate as a soft and hardware unit,
Herein not to limit the present invention.Specifically, the embodiment of each unit can refer to the description of previous embodiment one, herein no longer
It repeats.
Example IV:
Fig. 4 shows the structure of the PET image Super-resolution Reconstruction device based on decision tree of the offer of the embodiment of the present invention four,
For ease of description, only parts related to embodiments of the present invention are shown, including:
Decision tree construction unit 41, for according to preset training sample and preset division root section of the function from decision tree
Point starts to carry out recurrence division, until terminate the building of decision tree when the depth of division reaches preset decision tree depth threshold,
To realize the image block similar features cluster training of decision tree；
Image block extraction unit 42, for when receive target PET image super-resolution rebuilding request when, by target
PET image carries out image block extraction, obtains corresponding PET image block；
Mapping matrix obtaining unit 43, for being schemed by every decision tree in preparatory trained random forest to PET
As block progress image block similar features cluster, obtains image block that PET image block decision in every decision tree goes out, corresponding and reflect
Penetrate matrix；
Image block predicting unit 44 is predicted for image block mapping matrix gone out according to every decision tree decision, corresponding
The corresponding high-definition picture block of PET image block；And
Image reconstruction unit 45, for reconstructing the corresponding high-resolution of target PET image according to high-definition picture block
PET image.
Wherein it is preferred to which decision tree construction unit 41 includes:
Feature Selection unit 411, for in the corresponding present node training sample of decision tree present node to be divided
Low resolution training image blocks carry out image block characteristics extraction, until the image block characteristics that extract meet preset feature choosing
Formula is taken, chooses image block characteristics as node split feature, present node training sample belongs to training sample；
Node split unit 412, for being carried out according to node split feature and division function to present node training sample
Left and right child node division；
Node split stop unit 413 reaches decision tree depth threshold for the depth where working as present node, or works as
When the quantity of low resolution training image blocks is less than preset image block amount threshold in front nodal point training sample, stop to current
Node training sample carries out left and right child node division；And
Mapping matrix unit 414, for setting present node to the leaf node of decision tree, and by working as prosthomere
Point training sample learns image block mapping matrix.
In embodiments of the present invention, each unit of the PET image Super-resolution Reconstruction device based on decision tree can be by corresponding
Hardware or software unit realize that each unit can be independent soft and hardware unit, also can integrate as a soft and hardware unit,
Herein not to limit the present invention.Specifically, the embodiment of each unit can refer to the description of preceding method embodiment, herein not
It repeats again.
Embodiment five:
The structure that Fig. 5 shows the image processing equipment of the offer of the embodiment of the present invention five illustrates only for ease of description
Part related to the embodiment of the present invention.
The image processing equipment 5 of the embodiment of the present invention includes processor 50, memory 51 and is stored in memory 51
And the computer program 52 that can be run on processor 50.The processor 50 is realized above-mentioned based on certainly when executing computer program 52
Step in the PET image super-resolution reconstruction method embodiment of plan tree, such as step S101 to S104 shown in FIG. 1.Alternatively, place
Reason device 50 realizes the function of each unit in above-mentioned each Installation practice, such as unit 31 shown in Fig. 3 when executing computer program 52
To 34 function.
In embodiments of the present invention, image block extraction is carried out to the target PET image of pending super-resolution rebuilding, obtained
Corresponding PET image block carries out image block similar features cluster to PET image block by every decision tree in random forest,
Image block mapping matrix that PET image block decision in every decision tree goes out, corresponding is obtained, is gone out according to every decision tree decision
, the corresponding high-definition picture block of corresponding image block mapping matrix prediction PET image block, according to high-definition picture block weight
The corresponding high-resolution PET image of target PET image is built out, super-resolution rebuilding is carried out to target PET image to improve
Accuracy, while improving the resolution ratio and clarity of high-resolution PET image rebuild and obtained, and then improve and be based on
The accuracy of the clinic diagnosis of PET image.
The image processing equipment of the embodiment of the present invention can be personal computer, server.In the image processing equipment 5
The step of reason device 50 is realized when realizing the PET image super-resolution reconstruction method based on decision tree when executing computer program 52 can join
The description of embodiment of the method is stated before examination, and details are not described herein.
Embodiment six:
In embodiments of the present invention, a kind of computer readable storage medium is provided, which deposits
Computer program is contained, which realizes the above-mentioned PET image super-resolution weight based on decision tree when being executed by processor
Step in construction method embodiment, for example, step S101 to S104 shown in FIG. 1.Alternatively, the computer program is held by processor
The function of each unit in above-mentioned each Installation practice, such as the function of unit 31 to 34 shown in Fig. 3 are realized when row.
In embodiments of the present invention, image block extraction is carried out to the target PET image of pending super-resolution rebuilding, obtained
Corresponding PET image block carries out image block similar features cluster to PET image block by every decision tree in random forest,
Image block mapping matrix that PET image block decision in every decision tree goes out, corresponding is obtained, is gone out according to every decision tree decision
, the corresponding high-definition picture block of corresponding image block mapping matrix prediction PET image block, according to high-definition picture block weight
The corresponding high-resolution PET image of target PET image is built out, super-resolution rebuilding is carried out to target PET image to improve
Accuracy, while improving the resolution ratio and clarity of high-resolution PET image rebuild and obtained, and then improve and be based on
The accuracy of the clinic diagnosis of PET image.
The computer readable storage medium of the embodiment of the present invention may include can carry computer program code any
Entity or device, recording medium, for example, the memories 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 in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
1. a kind of PET image super-resolution reconstruction method based on decision tree, which is characterized in that the method includes the following steps:
When the super-resolution rebuilding for receiving target PET image is requested, the target PET image is subjected to image block extraction,
Obtain corresponding PET image block；
It is poly- that image block similar features are carried out to the PET image block by every decision tree in preparatory trained random forest
Class obtains image block mapping matrix that PET image block decision in every decision tree goes out, corresponding, wherein described
Random forest is made of more independent decision trees；
Described image block mapping matrix gone out according to every decision tree decision, corresponding predicts that the PET image block is corresponding
High-definition picture block；
The corresponding high-resolution PET image of the target PET image is reconstructed according to the high-definition picture block.
2. the method as described in claim 1, which is characterized in that in the step that the target PET image is carried out to image block extraction
Before rapid, the method also includes:
Recurrence division is carried out since the root node of the decision tree according to preset training sample and preset division function, directly
When the depth of the extremely division reaches preset decision tree depth threshold, terminate the building of the decision tree, to realize described determine
The image block similar features of plan tree cluster training, wherein the training sample by high-resolution training image blocks and with the height
The corresponding low resolution training image blocks composition of resolution ratio training image blocks.
3. method according to claim 2, which is characterized in that according to preset training sample and preset division function from institute
The root node for stating decision tree starts the step of carrying out recurrence division, comprising:
To the low resolution training image blocks in the corresponding present node training sample of present node that the decision tree is to be divided
Image block characteristics extraction is carried out, until the image block characteristics extracted meet preset Feature Selection formula, chooses described image
Block feature belongs to the training sample as node split feature, the present node training sample；
Left and right child node point is carried out to the present node training sample according to the node split feature and the division function
It splits.
4. method as claimed in claim 3, which is characterized in that work as to decision tree present node to be divided is corresponding
Before the step of low resolution training image blocks in front nodal point training sample carry out image block characteristics extraction, the method is also wrapped
It includes:
When the depth where the present node reaches preset decision tree depth threshold or the present node training sample
When the quantity of middle low resolution training image blocks is less than preset image block amount threshold, stop to present node training sample
This carries out left and right child node division；
It sets the present node to the leaf node of the decision tree, and institute is learnt by the present node training sample
State image block mapping matrix.
5. a kind of PET image Super-resolution Reconstruction device based on decision tree, which is characterized in that described device includes:
Image block extraction unit, for when receive target PET image super-resolution rebuilding request when, by the target PET
Image carries out image block extraction, obtains corresponding PET image block；
Mapping matrix obtaining unit, for passing through every decision tree in preparatory trained random forest to the PET image
Block carries out image block similar features cluster, obtains figure that PET image block decision in every decision tree goes out, corresponding
As block mapping matrix, wherein the random forest is made of more independent decision trees；
Image block predicting unit, for being gone out according to every decision tree decision, corresponding described image block mapping matrix to be pre-
Survey the corresponding high-definition picture block of the PET image block；And
Image reconstruction unit, for reconstructing the corresponding high-resolution of the target PET image according to the high-definition picture block
Rate PET image.
6. device as claimed in claim 5, which is characterized in that described device further include:
Decision tree construction unit, for according to preset training sample and preset division root node of the function from the decision tree
Start to carry out recurrence division, until terminating the decision tree when depth of the division reaches preset decision tree depth threshold
Building, with realize the decision tree image block similar features cluster training, wherein the training sample is instructed by high-resolution
Practice image block and low resolution training image blocks corresponding with high-resolution training image blocks composition.
7. device as claimed in claim 6, which is characterized in that the decision tree construction unit includes:
Feature Selection unit, for low in the corresponding present node training sample of decision tree present node to be divided
Resolution ratio training image blocks carry out image block characteristics extraction, until the image block characteristics extracted meet preset Feature Selection public affairs
Formula chooses described image block feature as node split feature, and the present node training sample belongs to the training sample；With
And
Node split unit is used for according to the node split feature and the division function to the present node training sample
Carry out left and right child node division.
8. device as claimed in claim 5, which is characterized in that described device further include:
Node split stop unit reaches the decision tree depth threshold for the depth where working as the present node, or
When the quantity of low resolution training image blocks is less than preset image block amount threshold in the present node training sample, stop
Left and right child node division is carried out to the present node training sample；And
Mapping matrix unit, for setting the present node to the leaf node of the decision tree, and by described
Present node training sample learns described image block mapping matrix.
9. a kind of image processing equipment, including memory, processor and storage are in the memory and can be in the processing
The computer program run on device, which is characterized in that the processor realizes such as claim 1 when executing the computer program
The step of to any one of 4 the method.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In when the computer program is executed by processor the step of any one of such as Claims 1-4 of realization the method.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811519205.5A CN109741255A (en) | 2018-12-12 | 2018-12-12 | PET image super-resolution reconstruction method, device, equipment and medium based on decision tree |
| PCT/CN2019/071109 WO2020118829A1 (en) | 2018-12-12 | 2019-01-10 | Decision-tree-based pet image super-resolution reconstruction method, apparatus and device, and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811519205.5A CN109741255A (en) | 2018-12-12 | 2018-12-12 | PET image super-resolution reconstruction method, device, equipment and medium based on decision tree |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109741255A true CN109741255A (en) | 2019-05-10 |
Family
ID=66358982
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811519205.5A Pending CN109741255A (en) | 2018-12-12 | 2018-12-12 | PET image super-resolution reconstruction method, device, equipment and medium based on decision tree |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109741255A (en) |
| WO (1) | WO2020118829A1 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110580689A (en) * | 2019-08-19 | 2019-12-17 | 深圳先进技术研究院 | image reconstruction method and device |
| CN112825271A (en) * | 2019-11-21 | 2021-05-21 | 通用电气精准医疗有限责任公司 | System and method for reconstructing medical images using a deep neural network and recursive extraction of measurement data |
| CN113674335A (en) * | 2021-08-19 | 2021-11-19 | 北京的卢深视科技有限公司 | Depth imaging method, electronic device, and storage medium |
| WO2023000526A1 (en) * | 2021-07-22 | 2023-01-26 | 苏州大学 | Image interpolation model training method based on residual-guided policy |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113128467B (en) * | 2021-05-11 | 2022-03-29 | 临沂大学 | Low-resolution face super-resolution and recognition method based on face priori knowledge |
| CN116188805B (en) * | 2023-04-26 | 2023-08-04 | 青岛尘元科技信息有限公司 | Image content analysis method and device for massive images and image information network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110268334A1 (en) * | 2010-04-30 | 2011-11-03 | Korean Advanced Institute Of Science And Technology | Apparatus for Improving Image Resolution and Apparatus for Super-Resolution Photography Using Wobble Motion and Point Spread Function (PSF), in Positron Emission Tomography |
| CN107680146A (en) * | 2017-09-13 | 2018-02-09 | 深圳先进技术研究院 | Method for reconstructing, device, equipment and the storage medium of PET image |
| CN107767337A (en) * | 2017-09-29 | 2018-03-06 | 深圳先进技术研究院 | A kind of image super-resolution rebuilding method, device, mobile terminal and storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018053340A1 (en) * | 2016-09-15 | 2018-03-22 | Twitter, Inc. | Super resolution using a generative adversarial network |
| CN107092884B (en) * | 2017-04-20 | 2020-09-25 | 宋佳颖 | Rapid coarse-fine cascade pedestrian detection method |
| CN107977930A (en) * | 2017-12-09 | 2018-05-01 | 北京花开影视制作有限公司 | A kind of image super-resolution method and its system |
| CN108416734A (en) * | 2018-02-08 | 2018-08-17 | 西北大学 | Text image super resolution ratio reconstruction method and device based on edge driving |
-
2018
- 2018-12-12 CN CN201811519205.5A patent/CN109741255A/en active Pending
-
2019
- 2019-01-10 WO PCT/CN2019/071109 patent/WO2020118829A1/en active Application Filing
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110268334A1 (en) * | 2010-04-30 | 2011-11-03 | Korean Advanced Institute Of Science And Technology | Apparatus for Improving Image Resolution and Apparatus for Super-Resolution Photography Using Wobble Motion and Point Spread Function (PSF), in Positron Emission Tomography |
| CN107680146A (en) * | 2017-09-13 | 2018-02-09 | 深圳先进技术研究院 | Method for reconstructing, device, equipment and the storage medium of PET image |
| CN107767337A (en) * | 2017-09-29 | 2018-03-06 | 深圳先进技术研究院 | A kind of image super-resolution rebuilding method, device, mobile terminal and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 卢欢: "基于随机森林的单幅图像超分辨重建算法研究", 《中国优秀硕士学位论文全文数据库(信息科技辑)》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110580689A (en) * | 2019-08-19 | 2019-12-17 | 深圳先进技术研究院 | image reconstruction method and device |
| CN112825271A (en) * | 2019-11-21 | 2021-05-21 | 通用电气精准医疗有限责任公司 | System and method for reconstructing medical images using a deep neural network and recursive extraction of measurement data |
| WO2023000526A1 (en) * | 2021-07-22 | 2023-01-26 | 苏州大学 | Image interpolation model training method based on residual-guided policy |
| CN113674335A (en) * | 2021-08-19 | 2021-11-19 | 北京的卢深视科技有限公司 | Depth imaging method, electronic device, and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020118829A1 (en) | 2020-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109741255A (en) | PET image super-resolution reconstruction method, device, equipment and medium based on decision tree | |
| JP7179757B2 (en) | Dose Reduction for Medical Imaging Using Deep Convolutional Neural Networks | |
| Gong et al. | PET image denoising using a deep neural network through fine tuning | |
| Häggström et al. | DeepPET: A deep encoder–decoder network for directly solving the PET image reconstruction inverse problem | |
| CN108898560A (en) | Rock core CT image super-resolution rebuilding method based on Three dimensional convolution neural network | |
| CN111462020B (en) | Method, system, storage medium and apparatus for motion artifact correction of cardiac images | |
| JP2020537555A (en) | Image reconstruction using machine learning regularizer | |
| CN104700438B (en) | Image rebuilding method and device | |
| Du et al. | Accelerated super-resolution MR image reconstruction via a 3D densely connected deep convolutional neural network | |
| KR101977067B1 (en) | Method for reconstructing diagnosis map by deep neural network-based feature extraction and apparatus using the same | |
| CN110415311B (en) | PET image reconstruction method, system, readable storage medium and apparatus | |
| CN110503699A (en) | A kind of CT projection path reduce in the case of CT image rebuilding method | |
| Shen et al. | Rapid reconstruction of highly undersampled, non‐Cartesian real‐time cine k‐space data using a perceptual complex neural network (PCNN) | |
| CN112070752A (en) | Method, device and storage medium for segmenting auricle of medical image | |
| Zeng et al. | Full-spectrum-knowledge-aware tensor model for energy-resolved CT iterative reconstruction | |
| CN112819911A (en) | Four-dimensional cone beam CT reconstruction image enhancement algorithm based on N-net and CycN-net network structures | |
| JP7139397B2 (en) | Systems and methods for reconstructing medical images using deep neural networks and recursive decimation of measured data | |
| Song et al. | Super-resolution PET using a very deep convolutional neural network | |
| Galve et al. | Super-iterative image reconstruction in PET | |
| CN112102251A (en) | Method and device for segmenting image, electronic equipment and storage medium | |
| EP4071705A1 (en) | Medical image processing method, medical image processing apparatus, and computer readable, non-volatile storage medium storing medical image processing program | |
| US20240062061A1 (en) | Methods for training a cnn and for processing an inputted perfusion sequence using said cnn | |
| CN106462988A (en) | Multi-view tomographic reconstruction | |
| CN117203671A (en) | Iterative image reconstruction improvement based on machine learning | |
| Xie et al. | 3D few-view CT image reconstruction with deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190510 |