CN110517222A - A kind of brain image dividing method based on multichannel chromatogram - Google Patents

A kind of brain image dividing method based on multichannel chromatogram Download PDF

Info

Publication number
CN110517222A
CN110517222A CN201910619691.6A CN201910619691A CN110517222A CN 110517222 A CN110517222 A CN 110517222A CN 201910619691 A CN201910619691 A CN 201910619691A CN 110517222 A CN110517222 A CN 110517222A
Authority
CN
China
Prior art keywords
indicate
region
value
label
formula
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
Application number
CN201910619691.6A
Other languages
Chinese (zh)
Inventor
颜成钢
陈安琪
朱嘉凯
孙垚棋
张继勇
张勇东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN201910619691.6A priority Critical patent/CN110517222A/en
Publication of CN110517222A publication Critical patent/CN110517222A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30016—Brain

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

The invention discloses a kind of brain image dividing method based on multichannel chromatogram.The present invention includes the following steps: step 1: the calculating of mask, introduces Mask function, for the label of the map marked in advance, does not need to calculate tissue points by the omission of Mask function；Use the label data of L maps as sample, when carrying out tag fusion, actual treatment is the labeled all tissue points of L maps；Step 2: a region of search is set in advance；Calculate the similarity of block in region of search；Vector standardization, calculating finally obtain weighted value；Step 3: carrying out the ballot of label value using the weighted value obtained, obtain final tag fusion result；Step 4: utilizing Overlapping Calculation segmentation precision.The present invention not only allows for the similarity in the region of corresponding position, while also contemplating corresponding position and nearby putting the Regional Similarity of position, therefore searching for more like block in certain field and may obtain better effect.

Description

A kind of brain image dividing method based on multichannel chromatogram
Technical field
The present invention relates to brain images to divide field, and in particular to label blending algorithm in the image segmentation based on multichannel chromatogram Improvement.
Background technique
Many Neuscience and clinical research have studied the shape of certain structures, such as hippocampus, they and some brains Disease, such as Alzheimer disease have close relationship.However, manual demarcate of interested structure is a cumbersome times Business, more particularly to the research of large data sets.Due to the development of technology, the amount of images to be analyzed is sharply increased, so that by hand Divide less effective in clinical practice.Therefore, the exploitation of automatic partition tools is the key that promote Medical imaging research.
Currently, the accurate segmentation to brain MR image has become the emphasis studied both at home and abroad.Due to every in human brain The region that a structure has him to fix, and these structures are also very similar in shape, therefore, the segmentation based on multichannel chromatogram Technology just grows up therewith.Cutting techniques based on multichannel chromatogram contain image registration and tag fusion two mostly important Process.
The common algorithm of tag fusion process has majority voting method (MV), local weighted ballot method (LWV) etc., traditional calculation Method is feasible, but still has the thinking of many innovatory algorithms, so as to promote the precision of segmentation, carrys out auxiliary diagnosis.
Summary of the invention
The present invention proposes a kind of brain image dividing method based on multichannel chromatogram, carries out to the algorithm during tag fusion It improves, the precision of brain image segmentation can be improved in this method.
A kind of brain image dividing method based on multichannel chromatogram, includes the following steps:
Step 1: the calculating (area-of-interest) of mask:
1-1. introduces Mask function, for the label of the map marked in advance, does not need to count by the omission of Mask function Calculate tissue points；
1-2. uses the label data of L maps as sample, and when carrying out tag fusion, actual treatment is L map marks All tissue points recorded a demerit；
Step 2: a region of search is set in advance；
2-1. calculates the similarity of block in region of search, and calculating formula of similarity is as follows:
Wherein, x indicates that input picture, y indicate map.Indicate region unit Pl(x) mean value,Indicate region unit Pl(x) variance.Indicate corresponding region blockMean value,Indicate corresponding region blockVariance.
By formula (1) it is found that if variance and mean value are closer, the value of s (x, y) is closer to 1, otherwise just close to 0, To obtain more like region unit；
The standardization of 2-2. vector, formula are as follows:
Wherein, N indicate regional area to be calculated all tissue points obtain number, IkIndicate that each tissue points obtains gray scale Value.
The formula that 2-3. finally obtains weighted value is as follows:
Wherein, Pk' indicate that vector of the input picture after regional area standardization, σ indicate constant.
Step 3: carrying out the ballot of label value using the weighted value obtained, obtain final tag fusion result；
Step 4: utilizing degree of overlapping Dice (Sa,Sb) calculation formula calculate segmentation precision:
Wherein Dice (Sa,Sb) indicate block degree of overlapping, wherein Sa∩SbIndicate the overlapping region of two width figures.|Sa| indicate input The label of image, | Sb| indicate the label of image after tag fusion.
It is concluded that from formula (4) if the two image is completely coincident, Dice (Sa,Sb) value will level off to 1, If the two picture registration is all very low, Dice (Sa,Sb) value will be intended to 0.
The present invention has the beneficial effect that:
Defect present in the conventional labels of being solved blending algorithm of the invention, first enhances gray level image, then On the basis of local weighted ballot method, the similarity in the region of corresponding position is not only allowed for, while also contemplating corresponding position The Regional Similarity for setting point position nearby, that is, think the registration process of front there are error, similar block not necessarily in corresponding position, And it there may be certain deviation, therefore search for more like block in certain field and may obtain better effect.
Detailed description of the invention
Fig. 1 is innovatory algorithm schematic diagram；
Fig. 2 is segmentation result schematic diagram；
Fig. 3 is experiment flow figure；
Specific embodiment
The present invention is further described in detail with reference to the accompanying drawings and embodiments.
Fig. 1 is the innovatory algorithm schematic diagram of LWV, a region of search is arranged in advance, such as the chain-dotted line region of Fig. 1, chain-dotted line Region just represents region of search.In chain-dotted line region unit, it is understood that there may be it is more more similar with input picture than dashed region block Voxel areas, therefore by being scanned for each tissue points in region of search, the better point of effect is found, is exactly the algorithm Main thought.
This method is the improvement proposed on the basis of local weighted ballot method.
Existing local weighted ballot method (Local Weighted Voting) is described as follows:
Under the basis of majority voting method, further boosting algorithm, since majority voting method does not utilize gray level image Gray value or mutual information between information, such as image etc., so local weighted ballot method is just come into being.
On the basis of majority voting method, the grayscale information near tissue points is calculated, and give according to the information of gray scale Each map assigns different weighted values, that is to say, that each map be not it is of equal value see generation, this and majority voting method Equivalence see generation, completely it is different.
The weighted value of local weighted ballot method must obtain as follows:
Gray value is obtained firstly the need of each tissue points are counted, and these gray values are added up to obtain m:
Wherein N indicates that the regional area all tissue points to be calculated obtain number, IkIndicate the gray scale of each tissue points Value.Expression is averaged,
Next standard deviation v is calculated:
Finally standard deviation is standardized, each tissue points in regional area i.e. dotted line frame handled well Vector Pk.Formula is as follows:
Finally calculate the weighted value at tissue points (x, x):
Wherein P'kIndicate vector of the input picture after regional area standardization.σ indicates constant, is generally assigned a value of 0.01.
It may be seen that above method has two, one is that we are not aware that the corresponding fixation of grayscale image Region must be best suitable for segmentation result, and for local weighted ballot method on the basis of majority voting method, more consideration is given to maps The information of corresponding position field block in image assigns different weighted values to different maps by obtaining the information of part, but The shortcomings that being local weighted ballot method is exactly that not more preferable consideration point to be treated corresponds to the position on map, is Most suitable position calculates weight simply by the same coordinate, and the precision of segmentation can be made to become less accurate.And We lack to its necessary pretreatment before gray level image processing.
Step of the present invention is implemented as follows:
Step 1: the calculating (area-of-interest) of mask；
The brain magnetic resonance imaging figure specification for being primarily due to width three-dimensional is 124 × 88 × 72, and required tissue points to be processed are very It is more, if calculated one by one, many times can be wasted, so the concept of a Mask is proposed here, due to the label of map Be that prior mark is good, wherein it is many it is unwanted put us and just pass through introduce Mask function, these are not needed into the point calculated It dispenses.In experiment, use the label data of three maps as sample, when carrying out tag fusion, we will be handled Tissue points be actually the labeled all tissue points of this three width map, greatly reduce in this way calculate the time.It is emerging due to feeling Interesting region be it is ready-portioned in advance, ready-portioned area-of-interest in advance can be found using mask function.Reduce calculation amount. As soon as that is classified as the tissue points as long as the mask in experiment is that have a sample that label is marked in certain tissue points in three samples mask。
Step 2: a region of search is set in advance, and the similarity for calculating block in region of search is the core of the algorithm, and formula is such as Under:
Wherein x indicates that input picture, y indicate map.Indicate region unit Pl(x) mean value,Indicate region unit Pl (x) variance.Indicate corresponding region blockMean value,Indicate corresponding region blockVariance.
By above formula it is recognised that if variance and mean value are closer, the value of s (x, y) is closer to 1, otherwise Just close to 0.To obtaining more like region unit, next just and local weighted voting algorithm is quite similar, only vector mark The process of standardization and the calculation method different from of weight.Wherein the standardized formula of vector is as follows:
Wherein, N indicate regional area to be calculated all tissue points obtain number, IkIndicate that each tissue points obtains gray scale Value.
The formula for finally obtaining weighted value is as follows:
Wherein, P'kIndicate vector of the input picture after regional area standardization.σ indicates constant.
Step 3: carrying out the ballot of label value using the weighted value obtained, obtain the result of final tag fusion；
Step 4: utilizing degree of overlapping Dice (Sa,Sb) calculation formula calculate segmentation precision, as shown in Figures 2 and 3:
Wherein Dice (Sa,Sb) indicate block degree of overlapping (Dice Overlap), wherein Sa∩SbIndicate the coincidence area of two width figures Domain.|Sa| indicate the label of input picture, | Sb| indicate the label of image after tag fusion.By formula it follows that if the two Image is completely coincident, then Dice (Sa,Sb) value will level off to 1, if the two picture registration is all very low, Dice (Sa,Sb) Value will be intended to 0, and therefore, which can be very good the accuracy for evaluating tag fusion algorithm.
Embodiment 1:
Step (1): the four big brain magnetic resonance imaging map that three-dimensional specification is 124 × 88 × 72 is obtained first, selects it In three as test maps, be used for tag fusion.That remaining map is as input map.
Step (2): initialization OutputImage_Label and Ratio_, the two respectively indicate output label image and Segmentation precision.
Step (3): traversal tri-dimensional picture need to only calculate the fusion tissue points labeled with Mask, and carry out to picture Enhancing pretreatment.
Step (4): it is registrated using formula (7), finds out and meet the several of most matching threshold in corresponding position region of search Tissue points.Because region of search is arranged to be the region of 3*3*3, several reasonable similar areas are found in the region of this 3*3*3 Domain (tissue points).It is used to calculate similarity first with formula 7, then sets a threshold value 0.05, then recognize more than this threshold value It is more similar.Then it is just calculated over the region of the threshold value, is equivalent to and does a pretreatment.Not so 27 regions will calculate, This step is mainly for reduction calculation amount.
Step (5): three samples, the weight of most matched tissue points corresponding label value are calculated separately out using formula (9).
Example: I, j, k indicate that voxel location, LabelHisto indicate the weight and be 34.xx that the voxel getting label value is 0； The weight and be 36.xx that label value is 1；Because 36 are greater than 34, the label of output is exactly 2.
Step (6): the ballot of label value is carried out using the weight obtained, obtains the result of final tag fusion OutputImage_Label, and utilize degree of overlapping Dice (Sa,Sb) calculation formula calculate segmentation precision Ratio_.
Interpretation of result:
2 tag fusion arithmetic result of table
As shown in Fig. 2, respectively indicating from left to right that three kinds of algorithms are finally split as a result, this three figures are cerebral nucleus One layer therein of magnetic resonance image, due to image be it is three-dimensional, for the ease of observation, I has intercepted one layer therein.
MV, LWV are respectively indicated from left to right, the final result that our improved methods are split.

Claims (1)

1. a kind of brain image dividing method based on multichannel chromatogram, it is characterised in that include the following steps:
Step 1: the calculating of mask:
1-1. introduces Mask function, for the label of the map marked in advance, does not need to calculate body by the omission of Mask function Vegetarian refreshments；
1-2. uses the label data of L maps as sample, and when carrying out tag fusion, actual treatment is that L maps are labeled All tissue points；
Step 2: a region of search is set in advance；
2-1. calculates the similarity of block in region of search, and calculating formula of similarity is as follows:
Wherein, x indicates that input picture, y indicate map；Indicate region unit Pl(x) mean value,Indicate region unit Pl(x) Variance；Indicate corresponding region blockMean value,Indicate corresponding region blockVariance；
By formula (1) it is found that if variance and mean value are closer, the value of s (x, y) is closer to 1, otherwise just close to 0, thus Obtain more like region unit；
The standardization of 2-2. vector, formula are as follows:
Wherein, N indicate regional area to be calculated all tissue points obtain number, IkIndicate that each tissue points obtains gray value；
The formula that 2-3. finally obtains weighted value is as follows:
Wherein, P 'kIndicate that vector of the input picture after regional area standardization, σ indicate constant；
Step 3: carrying out the ballot of label value using the weighted value obtained, obtain final tag fusion result；
Step 4: utilizing degree of overlapping Dice (Sa,Sb) calculation formula calculate segmentation precision:
Wherein Dice (Sa,Sb) indicate block degree of overlapping, wherein Sa∩SbIndicate the overlapping region of two width figures；|Sa| indicate input picture Label, | Sb| indicate the label of image after tag fusion；
It is concluded that from formula (4) if the two image is completely coincident, Dice (Sa,Sb) value will level off to 1, if The two picture registration is all very low, then Dice (Sa,Sb) value will be intended to 0.
CN201910619691.6A 2019-07-10 2019-07-10 A kind of brain image dividing method based on multichannel chromatogram Pending CN110517222A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910619691.6A CN110517222A (en) 2019-07-10 2019-07-10 A kind of brain image dividing method based on multichannel chromatogram

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910619691.6A CN110517222A (en) 2019-07-10 2019-07-10 A kind of brain image dividing method based on multichannel chromatogram

Publications (1)

Publication Number Publication Date
CN110517222A true CN110517222A (en) 2019-11-29

Family

ID=68622311

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910619691.6A Pending CN110517222A (en) 2019-07-10 2019-07-10 A kind of brain image dividing method based on multichannel chromatogram

Country Status (1)

Country Link
CN (1) CN110517222A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140226882A1 (en) * 2011-09-16 2014-08-14 Mcgill University Simultaneous segmentation and grading of structures for state determination
CN109523512A (en) * 2018-10-17 2019-03-26 哈尔滨理工大学 A kind of Automatic medical image segmentation method based on multichannel chromatogram tag fusion
CN109886944A (en) * 2019-02-02 2019-06-14 浙江大学 A kind of white matter high signal intensity detection and localization method based on multichannel chromatogram

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140226882A1 (en) * 2011-09-16 2014-08-14 Mcgill University Simultaneous segmentation and grading of structures for state determination
CN109523512A (en) * 2018-10-17 2019-03-26 哈尔滨理工大学 A kind of Automatic medical image segmentation method based on multichannel chromatogram tag fusion
CN109886944A (en) * 2019-02-02 2019-06-14 浙江大学 A kind of white matter high signal intensity detection and localization method based on multichannel chromatogram

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
MONTSERRAT ROBLES等: "Patch-based segmentation using expert priors Application to hippocampus and ventricle segmentation", 《NEUROIMAGE》 *
程有娥: "基于多图谱最优标签融合的脑MR图像分割", 《浙江工贸职业技术学院学报》 *

Similar Documents

Publication Publication Date Title
CN109035263B (en) Automatic brain tumor image segmentation method based on convolutional neural network
Avrunin et al. Using a priori data for segmentation anatomical structures of the brain
CN101809619B (en) Physical property value estimating system, age estimating system, gender estimating system, age/gender estimating system, and physical property value estimating method
García-Lorenzo et al. Review of automatic segmentation methods of multiple sclerosis white matter lesions on conventional magnetic resonance imaging
EP2481349B1 (en) Method, device, and program for processing images obtained from asl mri
CN110555835B (en) Brain slice image region division method and device
CN111932559A (en) New coronary pneumonia lung focus region segmentation system based on deep learning
CN106096636A (en) A kind of Advancement Type mild cognition impairment recognition methods based on neuroimaging
CN107316334B (en) Personalized precise magnetic resonance imaging method
CN105205810B (en) MR image hippocampus dividing method based on distance field fusion
CN111415324B (en) Classification and identification method for brain disease focus image space distribution characteristics based on magnetic resonance imaging
CN112348785B (en) Epileptic focus positioning method and system
CN110796661A (en) Fungal microscopic image segmentation detection method and system based on convolutional neural network
CN111968130A (en) Brain angiography image processing method, apparatus, medium, and electronic device
CN114519719A (en) Brain tumor MR image segmentation method
Liu et al. Robust cortical thickness morphometry of neonatal brain and systematic evaluation using multi-site MRI datasets
Hrdlicka et al. Diffusion tensor imaging and tractography in autistic, dysphasic, and healthy control children
Kumar et al. E-fuzzy feature fusion and thresholding for morphology segmentation of brain MRI modalities
CN108805181A (en) A kind of image classification device and sorting technique based on more disaggregated models
CN115984266B (en) Target positioning method and system in brain region
CN110517222A (en) A kind of brain image dividing method based on multichannel chromatogram
CN110751664B (en) Brain tissue segmentation method based on hyper-voxel matching
CN113487579B (en) Multi-mode migration method for automatically sketching model
Deng et al. Democratizing pathological image segmentation with lay annotators via molecular-empowered learning
Kong et al. Adversarial learning based structural brain-network generative model for analyzing mild cognitive impairment

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: 20191129