CN109886972A - A kind of brain magnetic resonance image partition method based on multilayer dictionary - Google Patents
A kind of brain magnetic resonance image partition method based on multilayer dictionary Download PDFInfo
- Publication number
- CN109886972A CN109886972A CN201910068621.6A CN201910068621A CN109886972A CN 109886972 A CN109886972 A CN 109886972A CN 201910068621 A CN201910068621 A CN 201910068621A CN 109886972 A CN109886972 A CN 109886972A
- Authority
- CN
- China
- Prior art keywords
- dictionary
- image
- image block
- split
- magnetic resonance
- 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
Abstract
The invention discloses a kind of brain magnetic resonance image partition methods based on multilayer dictionary, by constructing intermediate dictionary Contiguous graphics domain and tag field, the weight of successive optimization label fusion, so that weight is optimal finally for label fusion, rather than the weight that the weight that the image block based on grayscale information obtains is merged as final label is used only as in conventional methods where.Through the invention, the segmentation precision of brain magnetic resonance image can be improved.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of brain magnetic resonance image based on multilayer dictionary point
Segmentation method.
Background technique
21 century is called " century of brain " by International Brain Research Organization.Since 2000, China was by annual September 16th
It is named as " brain health day ", annual September is set to " the brain health moon ", it is intended to anti-by outreach brain science knowledge and cerebral disease
Measure is controlled, attention of the whole people to brain health is caused.However due to environmental factor, rhythm of life anxiety, aging of population, traffic meaning
The influence of outer equal many factors, the disease incidence such as craniocerebral injury, intracranial tumors and cerebrovascular disease are rising, it has also become human health
The first killer；The disease incidence of the diseases such as epilepsy, brain paralysis and Parkinson's disease is also increasing year by year；The disease incidence of depression position
The 4th of 150 kinds of world common disease is occupied, and is had the tendency that further up.It is reported that China has more than ten thousand people more than 200 every year
Brain disease occurs, wherein dying of the patient of cerebral disease every year there are about 1,000,000 or more, the cerebral disease death rate is about 45%.Therefore, brain
Disease has the characteristics that the death rate is high, disability rate is high and complication is more, becomes the primary killers for endangering human health.
According to the report of Center for Disease Control and Prevention, if cerebral disease can obtain effective diagnosing and treating,
The average life expectancy of patient will improve 10-20.Existing research shows: many diseases all with cerebral diseased close association.
For this purpose, many countries have formulated the long-range planning about brain science research in the world.The U.S. starts to provide the nineties in last century
The plan of studying for a long period of time about human brain is helped, each developed country in Europe has also carried out corresponding brain science research therewith.China is same
Pay close attention to basic research, brain image treatment research and study, memory, language processing of brain function and cerebral disease etc. and brain function
Relevant Mechanism Study, associated project are the basic research projects that country preferentially subsidizes.Brain diseases prevent and treat
Treatment has become the challenge that the whole mankind faces jointly.
Nearly two during the last ten years, and medical image has become the important technical of medical diagnosis, in clinical diagnosis, treatment, hand
It plays an important role in art planning navigation.Wherein magnetic resonance imaging (Magnetic Resonance Imaging, MRI) with
Its unique imaging mode is largely effective to the intracorporal soft tissue of biology, is widely used in brain image.Use MRI technique
Soft tissue as brain is imaged, the pathological tissues inside human brain can not only be observed directly, it can also be by brain
The quantitative analysis of interior each institutional framework discloses the rule of human brain development and aging, describes brain structure variation caused by various diseases.
With the development of medical image imaging technique, the data of magnanimity need to analyze and handle, using computer carry out image analysis at
For necessity.Wherein, medical image segmentation is the key that carry out Computer Image Processing and analysis, is to restrict Medical Image Processing neck
The basis that the bottleneck and medical image of the development of domain the relevant technologies and application understand.Therefore, medical image segmentation is in biomedicine
Research, clinical diagnosis, pathological analysis etc. have great importance, it has also become clinical field and field of medical image processing are total
With a key problem of concern.
In conclusion qualitatively and quantitatively analyzing brain tissue based on magnetic resonance image, effective prevention or treatment is taken to arrange
It applies, is of great significance to cerebral disease is reduced to human health bring grave danger.
In recent decades, many kinds of for the partitioning algorithm of brain magnetic resonance image, and still emerge one after another, but still can not
Fully meet the actual demand of people.Its reason is considerably complicated, comprising: people institute can not be briefly described with mathematical model completely
The practical problem faced；Since individual and lesion difference cause medical image to have complexity and diversity；Image degenerate with
And people are different to segmentation result target etc..These, which are all determined, can not be achieved a kind of pervasive, general segmentation
Method.It can only give and reasonably select for particular problem and specific demand, in critical index such as precision, speed and robustness
On make equilibrium or stress.Therefore, the type of corresponding dividing method is also shown as various informative, is driven if any data-driven and model
Dynamic point；It is divided into based on region, based on boundary and their blending algorithm；It is divided into automatic, semi-automatic and manual；There is prison
It superintends and directs and unsupervised point；It is divided into based on model and based on feature；It is divided into soft segmentation and hard segmentation；Have based on anatomical knowledge
With point etc. based on prior probability map.Wherein, the dividing method based on multichannel chromatogram has become representative brain image segmentation side
Method.Dividing method based on multichannel chromatogram registration was proposed in 2004 by Rohlfing et al. first.The method use priori moulds
The thought of type passes through map (map includes map gray level image and its good map tag image of manual segmentation) and figure to be split
As registration, the prior information stored in map is mapped directly in image to be split, under realizing that a kind of expert's " map " is instructed
Image segmentation.Due to that can complete to divide by the mapping of mark value, and the mark of a study morphology feature can be provided
Barebone, multichannel chromatogram method are widely used in image segmentation.
In all above-mentioned existing brain magnetic resonance image partition methods based on multichannel chromatogram, weight is by between image block
Grey similarity determines.The label of determining target image is directly used in by the calculated weight of the grayscale information of image.Although its
It is simple and effective, but evidence suggests such weight is that domain is constant, that is to say, that from the image based on grayscale information
The calculated weight of block institute may not lead to optimum mark fusion results.
Summary of the invention
The purpose of the present invention is provide a kind of brain based on multilayer dictionary to solve above-mentioned the deficiencies in the prior art place
Magnetic Resonance Image Segmentation method.
In order to solve the above technical problems, one technical scheme adopted by the invention is that: it provides a kind of based on multilayer dictionary
The step of brain magnetic resonance image partition method, this method includes:
The image block to be split centered on pixel to be split is obtained, is constructed just in existing atlas image according to grey similarity
Beginning gray scale dictionary, and label dictionary corresponding with initial gray dictionary is constructed in map tag image；
Arbitrary image block is the first image block of initial gray dictionary in specified initial gray dictionary, remaining in initial gray dictionary
Image block composition is directed to the sub- dictionary of gray scale of the first image block of initial gray dictionary, marks in dictionary with the sub- dictionary of gray scale to acting in accordance with
The sub- dictionary of label of image block composition the first image block of initial gray dictionary of sequence；Using the sub- dictionary of gray scale to initial gray dictionary
First image block carries out multichannel chromatogram expression, obtains weight vectors, and fusion is marked to the sub- dictionary of label using the weight vectors
Operation, obtains new image block, executes aforementioned operation to each image block in initial gray dictionary, generates the of intermediate dictionary
One layer；
Specify centre dictionary first layer arbitrary image block for intermediate the first image block of dictionary first layer again, in intermediate dictionary first layer
Remaining image block composition be directed to intermediate the first image block of dictionary first layer the sub- dictionary of gray scale, mark dictionary in the sub- dictionary of gray scale
The image block of corresponding sequence forms the sub- dictionary of label of intermediate dictionary the first image block of first layer；Using the sub- dictionary of gray scale to centre
The first image block of dictionary first layer carries out multichannel chromatogram expression, obtains weight vectors, using the weight vectors to mark sub- dictionary into
Line flag mixing operation obtains new image block, executes aforementioned operation to each image block in intermediate dictionary first layer, and repeatedly
This step of substitute performance constructs multi-level intermediate dictionary；
Image block to be split and initial gray dictionary are subjected to multichannel chromatogram expression, obtain the first weight of image block to be split to
Amount, and the first weight vectors are marked and are merged with dictionary is marked, obtains the first probabilistic image of image block to be split, by the
One probabilistic image inputs intermediate dictionary, successively indicates to operate with each layer of progress multichannel chromatogram of intermediate dictionary, obtains weight vectors
Mixing operation is marked with label dictionary afterwards, gradually handles, obtained probabilistic image is as final label fusion results；
Binaryzation is carried out by threshold value to obtained final label fusion results, confirms that pixel is to be split in image to be split
Target or background finally obtain the segmentation result of entire image to be split.
Wherein, the step of constructing the initial gray dictionary and corresponding label dictionary of specified one group of brain magnetic resonance image it
Before, include the steps that specifically including the image and image to be split progress image preprocessing in initial gray dictionary:
Skull is carried out to all magnetic resonance image and image to be split to operate；
Biased field correction is carried out using N4ITK algorithm to all magnetic resonance image and image to be split；
Gray scale normalization is used to all magnetic resonance image and image to be split, image grayscale is adjusted；
All magnetic resonance image are all registrated with image to be split, including are linearly registrated and non-linear registration；
The Screening Treatment based on grey similarity all is carried out with image to be split to all magnetic resonance image.
Wherein, image to be split is after each layer of processing of intermediate dictionary, and obtained probabilistic image is by threshold value to general
Rate carries out binaryzation, and binarization result is compared with preset threshold, if more than preset threshold, then as point of image to be split
Cut result output.
Wherein, binaryzation is carried out to probabilistic image, uses 0.5 as threshold value, thinks the pixel when probability is greater than 0.5
Point is target to be split；Conversely, thinking that the pixel is background when probability value is lower than 0.5.
Wherein, the mode for marking mixing operation is non-local mean or rarefaction representation.
It is different from the prior art, the image partition method of the invention based on multilayer dictionary is by successively constructing middle word
Allusion quotation, by the weight of intermediate dictionary successive optimization label fusion, so that weight is optimal, rather than picture finally for label fusion
Weight that the image block based on grayscale information obtains is used only like that in conventional methods where to be used as to the weight for marking fusion optimal.
Through the invention, the segmentation precision of brain magnetic resonance image can be improved.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of brain magnetic resonance image partition method based on multilayer dictionary provided by the invention.
Fig. 2 is a kind of calculating logic signal of brain magnetic resonance image partition method based on multilayer dictionary provided by the invention
Figure.
Fig. 3 is to be verified in dictionary in a kind of brain magnetic resonance image partition method based on multilayer dictionary provided by the invention often
The schematic diagram of the changes of entropy situation of a element.
Specific embodiment
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention.But the present invention can be with
Much it is different from other way described herein to implement, those skilled in the art can be without prejudice to intension of the present invention the case where
Under do similar popularization, therefore the present invention is not limited to the specific embodiments disclosed below.
Secondly, the present invention is described in detail using schematic diagram, when describing the embodiments of the present invention, for purposes of illustration only, showing
It is intended to be example, the scope of protection of the invention should not be limited herein.
As depicted in figs. 1 and 2, Fig. 1 is that a kind of process of image partition method based on multilayer dictionary of the invention is illustrated
Figure, Fig. 2 is calculating logic schematic diagram of the invention.The step of this method includes:
S110: the image block to be split centered on pixel to be split is obtained, according to grey similarity in existing atlas image
Initial gray dictionary is constructed, and constructs label dictionary corresponding with initial gray dictionary in map tag image.
S120: arbitrary image block is the first image block of initial gray dictionary, initial gray in specified initial gray dictionary
Remaining image block composition is directed to the sub- dictionary of gray scale of the first image block of initial gray dictionary in dictionary, marks sub with gray scale in dictionary
Dictionary corresponds to the sub- dictionary of label of image block composition the first image block of initial gray dictionary of sequence；Using the sub- dictionary of gray scale to first
Beginning the first image block of gray scale dictionary carries out multichannel chromatogram expression, obtains weight vectors, using the weight vectors to mark sub- dictionary into
Line flag mixing operation obtains new image block, aforementioned operation is executed to each image block in initial gray dictionary, in generation
Between dictionary first layer.
S130: specifying intermediate dictionary first layer arbitrary image block again is intermediate the first image block of dictionary first layer, middle word
In allusion quotation first layer remaining image block composition be directed to intermediate the first image block of dictionary first layer the sub- dictionary of gray scale, mark dictionary in
The image block that the sub- dictionary of gray scale corresponds to sequence forms the sub- dictionary of label of intermediate dictionary the first image block of first layer；Use gray scale
Dictionary carries out multichannel chromatogram expression to intermediate the first image block of dictionary first layer, weight vectors is obtained, using the weight vectors to mark
Remember that mixing operation is marked in sub- dictionary, obtain new image block, before being executed to each image block in intermediate dictionary first layer
Operation is stated, and iteration executes this step, constructs multi-level intermediate dictionary.
S140: carrying out multichannel chromatogram expression for image block to be split and initial gray dictionary, obtains the of image block to be split
One weight vectors, and the first weight vectors are marked with label dictionary and are merged, obtain the first probability of image block to be split
First probabilistic image is inputted intermediate dictionary, successively indicates to operate with each layer of progress multichannel chromatogram of intermediate dictionary, obtain by image
Mixing operation is marked with label dictionary after weight vectors, gradually handles, obtained probabilistic image is melted as final label
Close result.
S150: binaryzation is carried out by threshold value to obtained final label fusion results, confirms pixel in image to be split
Point is target to be split or background, finally obtains the segmentation result of entire image to be split.
Wherein, the step of constructing the initial gray dictionary and corresponding label dictionary of specified one group of brain magnetic resonance image it
Before, include the steps that specifically including the image and image to be split progress image preprocessing in initial gray dictionary:
Skull is carried out to all magnetic resonance image and image to be split to operate；
Biased field correction is carried out using N4ITK algorithm to all magnetic resonance image and image to be split；
Gray scale normalization is used to all magnetic resonance image and image to be split, image grayscale is adjusted.
Wherein, image to be split is after each layer of processing of intermediate dictionary, and obtained probabilistic image is by threshold value to general
Rate carries out binaryzation, and binarization result is compared with preset threshold, if more than preset threshold, then as point of image to be split
Cut result output.
Wherein, binaryzation is carried out to probabilistic image, uses 0.5 as threshold value, thinks the pixel when probability is greater than 0.5
Point is target to be split；Conversely, thinking that the pixel is background when probability value is lower than 0.5.
Wherein, the mode of mixing operation is non-local mean or rarefaction representation.
The weight in gray scale domain and tag field that the image when fusion is marked to image block is found through experiments that is not
Completely the same.In order to solve this problem, the invention proposes a new labels to merge frame, by constructing in a series of
Between dictionary by weight from image area be converted to tag field finally obtain optimum mark fusion weight.These dictionaries have been connected to image area
With tag field.The known one group atlas image block based on grayscale information marked image block corresponding with them is respectively ash
Spend image dictionary and label dictionary.The intermediate dictionary of construction describes a kind of path from from fuzzy probability to label.With this side
Formula can finally obtain the weight information based on tag field by intermediate dictionary hierarchical optimization weight vectors.
The present invention provides a kind of context mechanism information, improves the robustness that image block indicates.Specifically, for each
The construction of target image block, initial dictionary is consistent with most of traditional label fusion methods, i.e., by original figure spectrum (image area)
In image block composition.Since each atlas image has its corresponding handmarking's image, can be used and initial dictionary
The corresponding label dictionary (tag field) of the identical sequential configuration of middle map.In order to reduce the gap between image area and tag field,
Firstly for each image block in initial dictionary, using remaining all image block in addition to current image block as the image block
Atlas, obtain the weight vectors of the image block using known image tagged integration technology.Then, which is made
For marking dictionary, the corresponding probabilistic image block of the image block is obtained.Each image block in initial dictionary is repeated above-mentioned
Process obtains the corresponding probabilistic image block of all gray level image blocks in entire initial dictionary.
Following three pre-treatment steps first are executed to all experimental datas before the experiments were performed.Firstly, total to all magnetic
Vibration image has carried out that skull is gone to operate.Second, biased field correction has been carried out using N4ITK algorithm.Third uses gray scale normalization
Image grayscale is adjusted.4th pair of all magnetic resonance image are all registrated with image to be split, including are linearly registrated
And non-linear registration.5th pair of all magnetic resonance image all carry out the Screening Treatment based on grey similarity with image to be split.
Specific embodiment includes following procedure:
1, map selects
In order to improve the available time of algorithm, the present invention is based on grey similarities to be screened in advance to map and image block
Processing is selected with the most similar atlas image of test image gray scale as selected atlas image.And in order to test map number
The influence to experimental result is measured, the present invention is tested using the map of different number respectively.Experimental result is shown, with figure
The increase of quantity is composed, segmentation precision is gradually got higher, but when map quantity is more than 15, the raising of segmentation precision becomes very micro-
It is small, therefore in view of algorithm calculates the time, 15 maps are had chosen to each test image in all experiments of the invention.
2, gray scale dictionary constructs
Many researchs have been found that the image block in image with many similar structures, therefore are gone using these similar image blocks
The information that estimation testing image block obtains is more accurate.For each pixel in test image, pass through the phase between image block
Like the similitude between property replacement pixels, image block is chosen centered on the pixel, and provides a search window for each pixel
Mouthful, for each target image block, gray scale phase the most is chosen by search window in the position of same pixel point on selected map
As atlas image block, selected image block constitutes initial atlas gray scale dictionary.
3, label dictionary construction
Since each atlas image has its corresponding handmarking's image, can be used and phase in initial atlas gray scale dictionary
The same corresponding label dictionary of image block sequential configuration.
4, multilayer dictionary constructs
Assuming that having K image block in initial pictures map gray scale dictionary, then to each image block, following steps gradually structure is recycled
Make intermediate dictionary.
For each of dictionary image block, remaining all image block in the dictionary constitute new being directed to and are somebody's turn to do
The specific sub- dictionary of image block.The dimension of the sub- dictionary is.Therefore by label convergence strategy (non-local mean or
Rarefaction representation) image block is indicated using the sub- dictionary and then obtains weight vectors.Equally, the dimension of the weight vectors is。
For the tag image block that each image block in sub- dictionary has it corresponding in label dictionary, therefore pass through
Use the label dictionary of son corresponding to the available subgraph dictionary that puts in order identical with element in sub- dictionary.
The weight vectors that the first step is obtained act on the son label available new image block of dictionary in second step, should
Image block is as next layer of map gray level image block.
It all repeats the above steps for each of dictionary image block, new images corresponding to available each element
Block, and then construct with these new images blocks next layer of dictionary.
The number of plies of intermediate dictionary is most important to segmentation result, the present invention be respectively set different layers of dictionaries to image into
Row test, experimental result is shown, with the increase of the dictionary number of plies, segmentation precision is gradually got higher, when the figure dictionary number of plies is 4,
Segmentation precision reaches a stationary value.Therefore, the number of plies of intermediate dictionary is all set in all experiments of the invention as 4.
5, multilayer label fusion
As shown in Fig. 2, the present invention indicates the target figure centered on pixel using gray level image block dictionary in most initial layer
As block, to obtain initial weight vectors.Traditional is directly applied to the weight vectors based on the dividing method of multichannel chromatogram
Tag image block obtains final segmentation result.In order to optimize the weight vectors, which is acted on label by the present invention
Dictionary obtains the probabilistic image block of the initial gray image block.Then the hierarchical optimization probabilistic image block is until its close label
Until image block.Using the probabilistic image block as new target image block, the image block is indicated using new dictionary and then is obtained
New probabilistic image block.It steps be repeated alternatively until the last layer, the probabilistic image block for obtaining estimation is more next
Closer to tag image.Therefore final weight vectors will determine the mark of pixel as the optimal weights vector of label fusion
Note.
6, binaryzation
Aforesaid operations are repeated to each pixel to obtain passing through threshold value after corresponding probabilistic image block and can carry out two to probability
Value, the present invention use 0.5 as threshold value, think that the pixel is target to be split when probability is greater than 0.5.Work as probability value
Think that the pixel is background when lower than 0.5.
7, multilayer dictionary entropy is verified
Minimum entropy has been widely used as a criterion of classification of assessment method.Map is for example marked when for constant distribution
Its entropy is smaller, i.e., is not less than the entropy of the image by bias effect by the entropy of the image of bias effect.Therefore for general
Rate image, the information for being included are more than label profile information, therefore probability is more unstable, and entropy is bigger.Therefore this hair
The bright evolutionary process that intermediate dictionary is verified using entropy, with the increase of the dictionary number of plies, probability should tend towards stability value, right
The entropy answered should be also gradually reduced.
Fig. 3 describes the increase with the dictionary number of plies, the changes of entropy situation of each element in dictionary.Treat segmentation figure
Image block centered on each of picture pixel, search obtains a series of similar image blocks on map and their institutes are right
The tag image block answered.Image block in red dotted line frame is dictionary among constructed multilayer, and the number of plies successively increases from top to bottom
Add.The table on right side is entropy size corresponding to each image block.It should be pointed out that the present invention has calculated separately original ash
Spend the entropy of image block and tag image block.It is consistent with foregoing description, minimum (such as rightmost side Fig. 3 bottom of the entropy of tag image block
Shown in table).On the contrary, the entropy highest of gray level image block.Since entropy is related to minimum number of bits needed for encoding strength distribution, because
The predictability of this element marking and the concept of entropy are closely related.When stochastic variable can be predicted, its entropy is lower, otherwise entropy
It is higher.As shown in figure 3, for most of image blocks, its entropy is gradually reduced when the dictionary number of plies is from the 1st layer to the 4th layer, this meaning
Its probability value and authentic signature value more closely, the label of the i.e. image block is easier to predict.With the increase of the number of plies, dictionary
Become more sharp finally close to its true binary marks, demonstrates the present invention to the validity of image segmentation.
It is different from the prior art, the image partition method of the invention based on multilayer dictionary is by successively constructing middle word
Allusion quotation, by the weight of intermediate dictionary successive optimization label fusion, so that weight is optimal, rather than picture finally for label fusion
Weight that the image block based on grayscale information obtains is used only like that in conventional methods where to be used as to the weight for marking fusion optimal.
Through the invention, the segmentation precision of brain magnetic resonance image can be improved.
Although the invention has been described by way of example and in terms of the preferred embodiments, but it is not for limiting the present invention, any this field
Technical staff without departing from the spirit and scope of the present invention, may be by the methods and technical content of the disclosure above to this hair
Bright technical solution makes possible variation and modification, therefore, anything that does not depart from the technical scheme of the invention, and according to the present invention
Technical spirit any simple modifications, equivalents, and modifications to the above embodiments, belong to technical solution of the present invention
Protection scope.
Claims (5)
1. a kind of brain magnetic resonance image partition method based on multilayer dictionary, which comprises the following steps:
The image block to be split centered on pixel to be split is obtained, is constructed just in existing atlas image according to grey similarity
Beginning gray scale dictionary, and label dictionary corresponding with initial gray dictionary is constructed in map tag image；
Arbitrary image block is the first image block of initial gray dictionary in specified initial gray dictionary, remaining in initial gray dictionary
Image block composition is directed to the sub- dictionary of gray scale of the first image block of initial gray dictionary, marks in dictionary with the sub- dictionary of gray scale to acting in accordance with
The sub- dictionary of label of image block composition the first image block of initial gray dictionary of sequence；Using the sub- dictionary of gray scale to initial gray dictionary
First image block carries out multichannel chromatogram expression, obtains weight vectors, and fusion is marked to the sub- dictionary of label using the weight vectors
Operation, obtains new image block, executes aforementioned operation to each image block in initial gray dictionary, generates the of intermediate dictionary
One layer；
Specify centre dictionary first layer arbitrary image block for intermediate the first image block of dictionary first layer again, in intermediate dictionary first layer
Remaining image block composition be directed to intermediate the first image block of dictionary first layer the sub- dictionary of gray scale, mark dictionary in the sub- dictionary of gray scale
The image block of corresponding sequence forms the sub- dictionary of label of intermediate dictionary the first image block of first layer；Using the sub- dictionary of gray scale to centre
The first image block of dictionary first layer carries out multichannel chromatogram expression, obtains weight vectors, using the weight vectors to mark sub- dictionary into
Line flag mixing operation obtains new image block, executes aforementioned operation to each image block in intermediate dictionary first layer, and repeatedly
This step of substitute performance constructs multi-level intermediate dictionary；
Image block to be split and initial gray dictionary are subjected to multichannel chromatogram expression, obtain the initial weight of image block to be split to
Amount, and initial weight vector is marked with label dictionary and is merged, the probability image block of image block to be split is obtained, it will
Probability image block inputs intermediate dictionary, successively indicates to operate with each layer of progress multichannel chromatogram of intermediate dictionary, obtains weight
Mixing operation is marked with label dictionary after vector, by progressive updating weight vectors, hierarchical optimization probability graph photo until
It is close to until tag image piece, and the probabilistic image finally obtained is as final label fusion results；
Binaryzation is carried out by threshold value to obtained final label fusion results, confirms that pixel is to be split in image to be split
Target or background finally obtain the segmentation result of entire image to be split.
2. the brain magnetic resonance image partition method according to claim 1 based on multilayer dictionary, which is characterized in that constructing
Before the step of initial gray dictionary and corresponding label dictionary of specified one group of brain magnetic resonance image, including to initial gray word
The step of image and image to be split in allusion quotation carry out image preprocessing, specifically includes:
Skull is carried out to all magnetic resonance image and image to be split to operate；
Biased field correction is carried out using N4ITK algorithm to all magnetic resonance image and image to be split；
Gray scale normalization is used to all magnetic resonance image and image to be split, image grayscale is adjusted；
All magnetic resonance image are all registrated with image to be split, including are linearly registrated and non-linear registration；
The Screening Treatment based on grey similarity all is carried out with image to be split to all magnetic resonance image.
3. the brain magnetic resonance image partition method according to claim 1 based on multilayer dictionary, which is characterized in that be split
For image after each layer of processing of intermediate dictionary, obtained probabilistic image carries out binaryzation, binaryzation to probability by threshold value
As a result it is compared with preset threshold, if more than preset threshold, is then exported as the segmentation result of image to be split.
4. the brain magnetic resonance image partition method according to claim 3 based on multilayer dictionary, which is characterized in that probability
Image carries out binaryzation, uses 0.5 as threshold value, thinks that the pixel is target to be split when probability is greater than 0.5；Instead
It, thinks that the pixel is background when probability value is lower than 0.5.
5. the brain magnetic resonance image partition method according to claim 1 based on multilayer dictionary, which is characterized in that label melts
The mode of closing operation is non-local mean or rarefaction representation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910068621.6A CN109886972A (en) | 2019-01-24 | 2019-01-24 | A kind of brain magnetic resonance image partition method based on multilayer dictionary |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910068621.6A CN109886972A (en) | 2019-01-24 | 2019-01-24 | A kind of brain magnetic resonance image partition method based on multilayer dictionary |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109886972A true CN109886972A (en) | 2019-06-14 |
Family
ID=66926799
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910068621.6A Pending CN109886972A (en) | 2019-01-24 | 2019-01-24 | A kind of brain magnetic resonance image partition method based on multilayer dictionary |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886972A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112561921A (en) * | 2020-11-10 | 2021-03-26 | 联想(北京)有限公司 | Image segmentation method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103714536A (en) * | 2013-12-17 | 2014-04-09 | 深圳先进技术研究院 | Sparse-representation-based multi-mode magnetic resonance image segmentation method and device |
| US20150301141A1 (en) * | 2014-04-21 | 2015-10-22 | Case Western Reserve University | Nuclear Magnetic Resonance (NMR) Fingerprinting Tissue Classification And Image Segmentation |
| CN105205810A (en) * | 2015-08-30 | 2015-12-30 | 南方医科大学 | Distance-field-fusion-based hippocampus segmentation method of MR image |
| CN106204600A (en) * | 2016-07-07 | 2016-12-07 | 广东技术师范学院 | Cerebral tumor image partition method based on multisequencing MR image related information |
| CN107392914A (en) * | 2017-08-31 | 2017-11-24 | 齐鲁工业大学 | Nuclear magnetic resonance image dividing method based on self-adaption cluster algorithm |
-
2019
- 2019-01-24 CN CN201910068621.6A patent/CN109886972A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103714536A (en) * | 2013-12-17 | 2014-04-09 | 深圳先进技术研究院 | Sparse-representation-based multi-mode magnetic resonance image segmentation method and device |
| US20150301141A1 (en) * | 2014-04-21 | 2015-10-22 | Case Western Reserve University | Nuclear Magnetic Resonance (NMR) Fingerprinting Tissue Classification And Image Segmentation |
| CN105205810A (en) * | 2015-08-30 | 2015-12-30 | 南方医科大学 | Distance-field-fusion-based hippocampus segmentation method of MR image |
| CN106204600A (en) * | 2016-07-07 | 2016-12-07 | 广东技术师范学院 | Cerebral tumor image partition method based on multisequencing MR image related information |
| CN107392914A (en) * | 2017-08-31 | 2017-11-24 | 齐鲁工业大学 | Nuclear magnetic resonance image dividing method based on self-adaption cluster algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 宋艳涛: "基于磁共振成像的脑组织及病变分割方法研究", 《中国优秀博硕士学位论文全文数据库(博士)医药卫生科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112561921A (en) * | 2020-11-10 | 2021-03-26 | 联想(北京)有限公司 | Image segmentation method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dinsdale et al. | Learning patterns of the ageing brain in MRI using deep convolutional networks | |
| Shao et al. | Brain ventricle parcellation using a deep neural network: Application to patients with ventriculomegaly | |
| CN105760874B (en) | CT image processing system and its CT image processing method towards pneumoconiosis | |
| ES2595210T3 (en) | Method and apparatus for processing medical images | |
| Bodin et al. | Anatomo-functional correspondence in the superior temporal sulcus | |
| CN111967495A (en) | Classification recognition model construction method | |
| CN106204600A (en) | Cerebral tumor image partition method based on multisequencing MR image related information | |
| CN109685804B (en) | Multi-channel head magnetic resonance imaging tissue segmentation method | |
| CN112862805B (en) | Automatic auditory neuroma image segmentation method and system | |
| CN109102512A (en) | A kind of MRI brain tumor image partition method based on DBN neural network | |
| Spitzer et al. | Parcellation of visual cortex on high-resolution histological brain sections using convolutional neural networks | |
| Kori et al. | Ensemble of fully convolutional neural network for brain tumor segmentation from magnetic resonance images | |
| CN107292103A (en) | A kind of prognostic chart picture generation method and device | |
| CN112633416A (en) | Brain CT image classification method fusing multi-scale superpixels | |
| Wegmayr et al. | Generative aging of brain MR-images and prediction of Alzheimer progression | |
| Van Buren et al. | Artificial intelligence and deep learning to map immune cell types in inflamed human tissue | |
| CN109886972A (en) | A kind of brain magnetic resonance image partition method based on multilayer dictionary | |
| CN112233805B (en) | Mining method for biomarkers based on multi-map neuroimaging data | |
| CN117036288A (en) | Tumor subtype diagnosis method for full-slice pathological image | |
| Yao et al. | Brain hematoma segmentation using active learning and an active contour model | |
| CN116486156A (en) | Full-view digital slice image classification method integrating multi-scale feature context | |
| Corso et al. | Segmentation of sub-cortical structures by the graph-shifts algorithm | |
| Xu et al. | Bi-MGAN: Bidirectional T1-to-T2 MRI images prediction using multi-generative multi-adversarial nets | |
| CN115937083A (en) | Prostate magnetic resonance image region segmentation method fusing prior information | |
| CN115496732A (en) | Semi-supervised heart semantic segmentation algorithm |
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: 20190614 |