CN110136100A - The automatic classification method and device of CT sectioning image - Google Patents

The automatic classification method and device of CT sectioning image Download PDF

Info

Publication number
CN110136100A
CN110136100A CN201910302815.8A CN201910302815A CN110136100A CN 110136100 A CN110136100 A CN 110136100A CN 201910302815 A CN201910302815 A CN 201910302815A CN 110136100 A CN110136100 A CN 110136100A
Authority
CN
China
Prior art keywords
dcm
dimensional array
box
lbp
operator
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910302815.8A
Other languages
Chinese (zh)
Other versions
CN110136100B (en
Inventor
布社辉
胡金龙
麦善锐
梁晓阳
黄松
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.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
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 South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201910302815.8A priority Critical patent/CN110136100B/en
Publication of CN110136100A publication Critical patent/CN110136100A/en
Application granted granted Critical
Publication of CN110136100B publication Critical patent/CN110136100B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • 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
    • G06T2207/10081—Computed x-ray tomography [CT]
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses the automatic classification methods and device of a kind of CT sectioning image, body contour box is extracted among CT sectioning image, it is divided into different sub-boxs again, integrate the global characteristics vector that gradient and LBP characteristic statistics information in different sub-boxs generate corresponding CT sectioning image, by comparing calculating with template set to obtain final classification results, to realize with lesser calculation amount, lower computation complexity, quickly realize the classification to CT sectioning image, and it is able to satisfy the requirement handled in real time, and reduce the requirement to the performance of software and hardware, it can be with save the cost, reduce the difficulty of exploitation, meet the requirement to high speed large-scale data tupe.

Description

The automatic classification method and device of CT sectioning image
Technical field
The present invention relates to the technical fields of CT sectioning image processing, refer in particular to a kind of automatic classification side of CT sectioning image Method and device.
Background technique
Along with the development of Medical Imaging Technology, all kinds of digital medical image equipments are obtained in medical institutions at different levels It is widely applied.Especially CT (Computed Tomography, computed tomography) has imaging clearly, resolution ratio The advantages that height, speed is fast, moderate cost has been used as in clinical field to brain, chest, and the physical feelings such as abdomen carry out visible diagnosis A standard configuration scheme.For a patient, single pass probably generates few then tens, more then up to a hundred CT slice maps As data.Manual operation is still relied primarily on to the differentiations of these image datas processing at present, thus there are accuracys rate can not be true The problems such as guarantor, low efficiency and shortage of manpower, seriously restricts the processing to CT slice image data.Therefore artificial intelligence (Artificial Intelligence, abbreviation AI) technology is considered as the key to solve the above problems, and as computer aided manufacturing Help a core technology of diagnosis (Computer aided diagnosis, abbreviation CAD).It is sliced in artificial intelligence application in CT It when the processing of image data, needs to classify to CT sectioning image, that is, judges that the CT sectioning image belongs to the specific portion of body Position, and then differentiation processing is carried out using the model of corresponding physical feeling.Existing classification method mainly has two major classes: Yi Leifang Method be by match with the template constructed in advance compare realize differentiate；In addition one kind is by " Feature extraction~+ classification This traditional mode identification method of device " carries out, especially such as convolutional neural networks (Convolutional Neural Networks, abbreviation CNN) as deep learning (deep learning) method have very high performance.Using the first mould The matched method of plate needs to construct standard form, and the accuracy of this method relies on the quality of template, and this method Due to not accounting for individual difference, and the otherness of physical condition when being detected, therefore practical application effect is undesirable. Classifier among second method generally uses supervised learning mode, this means that must have one has mark letter The training sample data collection of breath.Since mark training sample data collection is a time-consuming and laborious job, and this method Performance relies on the quality and quantity of mark training sample data collection, therefore the extensive instruction of the artificial markup information with high quality Practicing sample data set is less easy acquisition.
Summary of the invention
The purpose of the present invention is to overcome the shortcomings of the existing technology and deficiency, proposes a kind of dividing automatically for CT sectioning image Class method and apparatus solve among existing CT sectioning image classification method, this technical solution compared using template matching The quality of existing matching accuracy heavy dependence matching template, and matching template can not be repaired for the otherness of individual Just, the computation complexity of matching algorithm is higher, and calculation amount is huger；And it is big to use the sorting algorithm of traditional mode identification to rely on These technical problems of the artificial mark sample data set of amount.
To achieve the above object, technical solution provided by the present invention is as follows:
The automatic classification method of CT sectioning image, comprising the following steps:
Step 1 is calculated using maximum between-cluster variance algorithm by the CT sectioning image two-dimensional array DCM of inputIIt is carried on the back Scene area and targeted body region carry out the threshold value V of optimal segmentationTH, according to the threshold value VTHBy the CT sectioning image two dimension Array DCMIIt carries out binary conversion treatment and obtains bianry image two-dimensional array DCMB, to the bianry image two-dimensional array DCMBIt carries out The processing of two-value opening operation obtains contour images two-dimensional array DCMC, from the contour images two-dimensional array DCMCIt extracts comprising described The box C of targeted body regionBox=[Xleft,Ytop,Xright,Ybottom], the XleftFor the box CBOXIn X-direction Left margin, the YtopFor the box CBOXCoboundary, the X in Y directionrightFor the box CBOXIn the X-axis side To right margin, the YbottomFor the box CBOXIn the lower boundary of the Y direction；
Step 2, by the box CBOXIt is divided into the identical sub-box C of M*N sizesubBOX(m,n)=[Xleft+(m-1)* W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H], the M is by the box CBOXIn the number of the X-direction equal part, The N is by box CBOXIn the number of the Y direction equal part, the m is the sub-box CsubBOX(m,n)In the X-axis side To serial number, the value range of the m is [1, M], and the n is the sub-box CsubBOX(m,n)In the serial number of the Y direction, The value range of the n is [1, N], and the W is the sub-box CsubBOX(m,n)In the width of the X-direction, the W= (Xright-Xleft)/M, the H are the sub-box CsubBOX(m,n)In the height of the Y direction, the H=(Ybottom- Ytop)/N；
Step 3 calculates the CT sectioning image two-dimensional array DCM using operator Sobel_XIGradient distribution two-dimensional array DCMG_X, the operator Sobel_X=[- 101, -2 02, -1 0 1] uses operator Sobel_Y to calculate the CT slice map As two-dimensional array DCMIGradient distribution two-dimensional array DCMG_Y, the operator Sobel_Y=[- 1-2-1,000,12 1], the CT sectioning image two-dimensional array DCM is calculated using operator Sobel_XYIGradient distribution two-dimensional array DCMG_XY, described Operator Sobel_XY=[0 12 ,-1 01 ,-2-1 0] calculates the CT sectioning image two-dimemsional number using operator Sobel_YX Group DCMIGradient distribution two-dimensional array DCMG_YX, the operator Sobel_YX=[- 2-1 0 ,-1 01,01 2], calculating institute State CT sectioning image two-dimensional array DCMILocal binary patterns be distributed two-dimensional array DCMLBP；
Step 4, from the gradient distribution two-dimensional array DCMG_XCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_X(m,n), from the ladder Degree distribution two-dimensional array DCMG_YCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)* H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_Y(m,n), from the gradient distribution two-dimensional array DCMG_XYIt calculates The corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] ladder Spend histogram distribution vector HG_XY(m,n), from the gradient distribution two-dimensional array DCMG_YXCalculate the corresponding sub-box CsubBOX(m,n) Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_YX(m,n), Two-dimensional array DCM is distributed from local binary patternsLBPCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)* W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] local binary patterns histogram distribution vector HLBP(m,n)；
Step 5 passes through formula HG_LBP=concat ([HG_X(m,n),HG_Y(m,n),HG_XY(m,n),HG_YX(m,n),HLBP(m,n)]) |m:1->M,n:1->NBy the gradient histogram distribution vector HG_X(m,n), the gradient histogram distribution vector HG_Y(m,n), the gradient it is straight Square distribution vector HG_XY(m,n), the gradient histogram distribution vector HG_YX(m,n), the local binary patterns histogram distribution vector HLBP(m,n)Carry out vector merging, the HG_LBPFor the CT sectioning image two-dimensional array DCMIGlobal characteristics vector, it is described Concat is vector pooled function；
Step 6 passes through formulaCalculate with the global characteristics to Measure HG_LBPApart from shortest CiCorresponding classification sequence number ioptAs the CT sectioning image two-dimensional array DCMIAffiliated class Not, the CiFor the class template collection { C1,C2,…,CKInterior classification i global characteristics vector, the K is the classification Sum, the global characteristics vector HG_LBPWith the global characteristics vector C of the classification iiDistance calculation formula dist (Ci, HG_LBP)=| | Ci-HG_LBP||2。
Further, in step 3, the CT sectioning image two-dimensional array DCM is calculated using operator Prewitt_XILadder Spend distribution map DCMG_X, the operator Prewitt_X=[- 101, -1 01, -1 0 1] calculated using operator Prewitt_Y The CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_Y, the operator Prewitt_Y=[- 1-1-1,00 0,11 1], the CT sectioning image two-dimensional array DCM is calculated using operator Prewitt_XYIGradient distribution figure DCMG_XY, institute Operator Prewitt_XY=[0 11 ,-1 01 ,-1-1 0] is stated, calculates the CT sectioning image using operator Prewitt_YX Two-dimensional array DCMIGradient distribution figure DCMG_YX, the operator Prewitt_YX=[- 1-1 0 ,-1 01,01 1], calculating The CT sectioning image two-dimensional array DCMILocal binary patterns distribution map DCMLBP。
Further, in step 3, the CT sectioning image two-dimensional array DCM is calculated using operator Roberts_XILadder Spend distribution map DCMG_X, the operator Roberts_X=[1-1,1-1] uses operator Roberts_Y to calculate CT slice Two-dimensional image array DCMIGradient distribution figure DCMG_Y, the operator Roberts_Y=[- 1-1,1 1] uses operator Roberts_XY calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_XY, the operator Roberts_XY =[1 0,0-1] calculates the CT sectioning image two-dimensional array DCM using operator Roberts_YXIGradient distribution figure DCMG_YX, the operator Roberts_YX=[0 1, -1 0] calculates the CT sectioning image two-dimensional array DCMILocal binary Mode distribution map DCMLBP。
Further, in step 6, pass through formula argmin (distKL(Ci,HG_LBP))|Ci∈{C1,C2,…,CK}Calculate with it is described Global characteristics vector HG_LBPApart from shortest CiCorresponding classification sequence number ioptAs the CT sectioning image DCMIAffiliated class Not, the CiFor the class template collection { C1,C2,…,CKInterior classification i global characteristics vector, the K is the classification Sum, the global characteristics vector HG_LBPWith the global characteristics vector C of the classification iiDistance calculation formula distKL(Ci, HG_LBP)=(ΣJ=1:L(Ci[j]*log(Ci[j]/HG_LBP[j]))+ΣJ=1:L(HG_LBP[j]*log(HG_LBP[j]/Ci[j])))/ 2, the j are the global characteristics vector HG_LBPAnd the global characteristics vector CiThe serial number of inner element, the L are described Global characteristics vector HG_LBPAnd the global characteristics vector CiLength.
Further, in step 1, the CT sectioning image two-dimensional array DCMIAll array elements value range quilt Canonical turns to [0,1024].
The automatic classification of CT sectioning image is set, comprising:
Targeted body region module, for being calculated using maximum between-cluster variance algorithm by the CT sectioning image two dimension of input Array DCMIIt carries out background area and targeted body region carries out the threshold value V of optimal segmentationTH, according to the threshold value VTHBy the CT Sectioning image two-dimensional array DCMIIt carries out binary conversion treatment and obtains bianry image two-dimensional array DCMB, to the bianry image two dimension Array DCMBIt carries out the processing of two-value opening operation and obtains contour images two-dimensional array DCMC, from the contour images two-dimensional array DCMC Extract the box C comprising the targeted body regionBox=[Xleft,Ytop,Xright,Ybottom], the XleftFor the box CBOXLeft margin, the Y in X-directiontopFor the box CBOXCoboundary, the X in Y directionrightFor the box CBOXRight margin, the Y in the X-directionbottomFor the box CBOXIn the lower boundary of the Y direction；
Sub-box divides module, is used for the box CBOXIt is divided into the identical sub-box C of M*N sizesubBOX(m,n)= [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H], the M is by the box CBOXIn the X-direction The number of equal part, the N are by box CBOXIn the number of the Y direction equal part, the m is the sub-box CsubBOX(m,n) In the serial number of the X-direction, the value range of the m is [1, M], and the n is the sub-box CsubBOX(m,n)In the Y-axis The serial number in direction, the value range of the n are [1, N], and the W is the sub-box CsubBOX(m,n)In the width of the X-direction Degree, the W=(Xright-Xleft)/M, the H are the sub-box CsubBOX(m,n)In the height of the Y direction, the H= (Ybottom-Ytop)/N；
Characteristic extracting module, for calculating the CT sectioning image two-dimensional array DCM using operator Sobel_XIGradient It is distributed two-dimensional array DCMG_X, the operator Sobel_X=[- 101, -2 02, -1 0 1] calculated using operator Sobel_Y The CT sectioning image two-dimensional array DCMIGradient distribution two-dimensional array DCMG_Y, the operator Sobel_Y=[- 1-2-1, 000,12 1], the CT sectioning image two-dimensional array DCM is calculated using operator Sobel_XYIGradient distribution two-dimensional array DCMG_XY, the operator Sobel_XY=[0 12 ,-1 01 ,-2-1 0] calculates the CT using operator Sobel_YX and cuts Picture two-dimensional array DCMIGradient distribution two-dimensional array DCMG_YX, the operator Sobel_YX=[- 2-1 0 ,-1 01,0 1 2], the CT sectioning image two-dimensional array DCM is calculatedILocal binary patterns be distributed two-dimensional array DCMLBP；
Sub-box characteristic module is used for from the gradient distribution two-dimensional array DCMG_XCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_X(m,n), from the gradient distribution two-dimensional array DCMG_YCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m- 1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_Y(m,n), from the gradient distribution two dimension Array DCMG_XYCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m* W,Ytop+ n*H] gradient histogram distribution vector HG_XY(m,n), from the gradient distribution two-dimensional array DCMG_YXIt calculates described in corresponding to Sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram point Cloth vector HG_YX(m,n), two-dimensional array DCM is distributed from local binary patternsLBPCalculate the corresponding sub-box CsubBOX(m,n)Institute is in place Set [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] local binary patterns histogram distribution vector HLBP(m,n)；
Global characteristics module, for passing through formula HG_LBP=concat ([HG_X(m,n),HG_Y(m,n),HG_XY(m,n), HG_YX(m,n),HLBP(m,n)])|m:1->M,n:1->NBy the gradient histogram distribution vector HG_X(m,n), the gradient histogram distribution vector HG_Y(m,n), the gradient histogram distribution vector HG_XY(m,n), the gradient histogram distribution vector HG_YX(m,n), the local binary mould Formula histogram distribution vector HLBP(m,n)Carry out vector merging, the HG_LBPFor the CT sectioning image two-dimensional array DCMIThe overall situation it is special Vector is levied, the concat is vector pooled function；
Classification discrimination module, for passing through formulaCalculating and institute State global characteristics vector HG_LBPApart from shortest CiCorresponding classification sequence number ioptAs the CT sectioning image two-dimensional array DCMIAffiliated classification, the CiFor the class template collection { C1,C2,…,CKInterior classification i global characteristics vector, the K is The sum of the classification, the global characteristics vector HG_LBPWith the global characteristics vector C of the classification iiDistance calculation formula dist(Ci,HG_LBP)=| | Ci-HG_LBP||2。
Further, in the characteristic extracting module, the CT sectioning image two-dimemsional number is calculated using operator Prewitt_X Group DCMIGradient distribution figure DCMG_X, the operator Prewitt_X=[- 101, -1 01, -1 0 1] uses operator Prewitt_Y calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_Y, the operator Prewitt_Y= [- 1-1-1,000,11 1] calculate the CT sectioning image two-dimensional array DCM using operator Prewitt_XYIGradient Distribution map DCMG_XY, the operator Prewitt_XY=[0 11 ,-1 01 ,-1-1 0] counted using operator Prewitt_YX Calculate the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_YX, the operator Prewitt_YX=[- 1-1 0 ,-1 01,01 1], the CT sectioning image two-dimensional array DCM is calculatedILocal binary patterns distribution map DCMLBP。
Further, in the characteristic extracting module, the CT sectioning image two-dimemsional number is calculated using operator Roberts_X Group DCMIGradient distribution figure DCMG_X, the operator Roberts_X=[1-1,1-1] calculated using operator Roberts_Y The CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_Y, the operator Roberts_Y=[- 1-1,1 1] makes The CT sectioning image two-dimensional array DCM is calculated with operator Roberts_XYIGradient distribution figure DCMG_XY, the operator Roberts_XY=[1 0,0-1] calculates the CT sectioning image two-dimensional array DCM using operator Roberts_YXIGradient Distribution map DCMG_YX, the operator Roberts_YX=[0 1, -1 0] calculates the CT sectioning image two-dimensional array DCMIOffice Portion binary pattern distribution map DCMLBP。
Further, in the classification discrimination module, pass through formula It calculates and the global characteristics vector HG_LBPApart from shortest CiCorresponding classification sequence number ioptAs the CT Sectioning image DCMIAffiliated classification, the CiFor the class template collection { C1,C2,…,CKInterior classification i global characteristics to Amount, the K are the sum of the classification, the global characteristics vector HG_LBPWith the global characteristics vector C of the classification iiAway from From calculation formula distKL(Ci,HG_LBP)=(ΣJ=1:L(Ci[j]*log(Ci[j]/HG_LBP[j]))+ΣJ=1:L(HG_LBP[j]* log(HG_LBP[j]/Ci[j])))/2, the j is the global characteristics vector HG_LBPAnd global characteristics vector CiInner element Serial number, the L be the global characteristics vector HG_LBPAnd global characteristics vector CiLength.
Further, in the targeted body region module, the CT sectioning image two-dimensional array DCMIAll arrays The value range of element turns to [0,1024] by canonical.
Compared with prior art, the present invention have the following advantages that with the utility model has the advantages that
The present invention can need not rely on a large amount of artificial mark sample data set, so that it may which realization quick and precisely cuts CT Picture is classified, while can also greatly reduce calculation amount, reduces the complexity of calculating, accelerates processing speed, is reduced The calculating time, meet the requirement handled in real time, and reduce the requirement to the performance of software and hardware, can with save the cost, The difficulty for reducing exploitation meets the requirement to high speed large-scale data tupe.
Detailed description of the invention
Fig. 1 is the flow diagram of the automatic classification method of CT sectioning image of the present invention.
Fig. 2 is the functional block diagram of the apparatus for automatically sorting of CT sectioning image of the present invention.
Specific embodiment
The present invention is further explained in the light of specific embodiments.
The mobile terminal of each embodiment of the present invention is realized in description with reference to the drawings.In subsequent description, use For indicate element such as " module ", " component " or " unit " suffix only for being conducive to explanation of the invention, itself There is no specific meanings.Therefore, " module " can be used mixedly with " component ".
As shown in Figure 1, the automatic classification method of CT sectioning image provided by the present embodiment, comprising the following steps:
Step S10, the region of physical target is obtained.
Calculated using maximum between-cluster variance algorithm by the CT sectioning image two-dimensional array DCM of inputICarry out background area Domain and targeted body region carry out the threshold value V of optimal segmentationTH, according to the threshold value VTHBy the CT sectioning image two-dimensional array DCMIIt carries out binary conversion treatment and obtains bianry image two-dimensional array DCMB, to the bianry image two-dimensional array DCMBCarry out two-value Opening operation processing obtains contour images two-dimensional array DCMC, from the contour images two-dimensional array DCMCExtracting includes the body The box C of target areaBox=[Xleft,Ytop,Xright,Ybottom], the XleftFor the box CBOXOn the left side of X-direction Boundary, the YtopFor the box CBOXCoboundary, the X in Y directionrightFor the box CBOXIn the X-direction Right margin, the YbottomFor the box CBOXIn the lower boundary of the Y direction.
Step S20, sub-box is divided.
I.e. by the box CBOXIt is divided into the identical sub-box C of M*N sizesubBOX(m,n)=[Xleft+(m-1)*W,Ytop +(n-1)*H,Xleft+m*W,Ytop+ n*H], the M is by the box CBOXIn the number of the X-direction equal part, the N For by box CBOXIn the number of the Y direction equal part, the m is the sub-box CsubBOX(m,n)In the sequence of the X-direction Number, the value range of the m is [1, M], and the n is the sub-box CsubBOX(m,n)In the serial number of the Y direction, the n Value range be [1, N], the W be the sub-box CsubBOX(m,n)In the width of the X-direction, the W=(Xright- Xleft)/M, the H are the sub-box CsubBOX(m,n)In the height of the Y direction, the H=(Ybottom-Ytop)/N。
Step S30, feature extraction.
The CT sectioning image two-dimensional array DCM is calculated using operator Sobel_XIGradient distribution two-dimensional array DCMG_X, the operator Sobel_X=[- 101, -2 02, -1 0 1] uses operator Sobel_Y to calculate the CT slice map As two-dimensional array DCMIGradient distribution two-dimensional array DCMG_Y, the operator Sobel_Y=[- 1-2-1,000,12 1], the CT sectioning image two-dimensional array DCM is calculated using operator Sobel_XYIGradient distribution two-dimensional array DCMG_XY, described Operator Sobel_XY=[0 12 ,-1 01 ,-2-1 0] calculates the CT sectioning image two-dimemsional number using operator Sobel_YX Group DCMIGradient distribution two-dimensional array DCMG_YX, the operator Sobel_YX=[- 2-1 0 ,-1 01,01 2], calculating institute State CT sectioning image two-dimensional array DCMILocal binary patterns be distributed two-dimensional array DCMLBP。
Step S40, sub-box characteristic statistics.
I.e. from the gradient distribution two-dimensional array DCMG_XCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+ (m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_X(m,n), from the gradient distribution Two-dimensional array DCMG_YCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft +m*W,Ytop+ n*H] gradient histogram distribution vector HG_Y(m,n), from the gradient distribution two-dimensional array DCMG_XYCalculate corresponding institute State sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram Distribution vector HG_XY(m,n), from the gradient distribution two-dimensional array DCMG_YXCalculate the corresponding sub-box CsubBOX(m,n)Institute is in place Set [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_YX(m,n), from part Binary pattern is distributed two-dimensional array DCMLBPCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+ (n-1)*H,Xleft+m*W,Ytop+ n*H] local binary patterns histogram distribution vector HLBP(m,n)。
Step S50, global characteristics calculate.
Pass through formula HG_LBP=concat ([HG_X(m,n),HG_Y(m,n),HG_XY(m,n),HG_YX(m,n),HLBP(m,n)]) |m:1->M,n:1->NBy the gradient histogram distribution vector HG_X(m,n), the gradient histogram distribution vector HG_Y(m,n), the gradient it is straight Square distribution vector HG_XY(m,n), the gradient histogram distribution vector HG_YX(m,n), the local binary patterns histogram distribution vector HLBP(m,n)Carry out vector merging, the HG_LBPFor the CT sectioning image two-dimensional array DCMIGlobal characteristics vector, it is described Concat is vector pooled function.
Step S60, classification differentiates.
Pass through formula argmin (dist (Ci, HG_LBP))|Ci∈{C1,C2,…,CK}It calculates and the global characteristics vector HG_LBP The classification sequence number i corresponding to the shortest CioptAs the CT sectioning image two-dimensional array DCMIAffiliated classification, it is described Ci is the global characteristics vector of the class template collection { C1, C2 ..., CK } interior classification i, and the K is the sum of the classification, institute State global characteristics vector HG_LBPWith distance calculation formula dist (Ci, the H of the global characteristics vector Ci of the classification iG_LBP)=| | Ci-HG_LBP||2。
The actual physical meaning of the pixel value (referred to as CT value) of CT sectioning image is the density of tissue or organ, meter Amount unit is HU (hounsfield unit), and the CT value of general air is about -1000, the CT of the highest bone of body density Value close+1000.Probably there are two main wave crests, a wave crest institutes for the histogram distribution of the pixel value of one CT sectioning image Regional Representative's air of distribution, and Regional Representative's bodily tissue that another wave crest is distributed.Therefore maximum between-cluster variance is used Algorithm (OTSU algorithm) can search out the threshold value V of optimal segmentationTHCT sectioning image is subjected to binary conversion treatment, by DCMIAmong Pixel value is lower than threshold value VTHCorresponding bianry image DCMBThe pixel value of same coordinate position be set as 0, that is, determine that the pixel is Air；By DCMIAmong pixel value be lower than threshold value VTHCorresponding bianry image DCMBThe pixel value of same coordinate position be set as 1, Determine the pixel for bodily tissue.Again by DCMBTwo-value opening operation is carried out to handle to obtain DCMC, i.e., to DCMBRow first carries out The processing of two-value erosion algorithm, then carries out two-value expansion algorithm processing, to filter out DCM againBInterior isolated noise spot, and And it is smooth to get to body contour DCM to obtain the profile of physical feelingC.It can be obtained by by contours extract algorithm in DCMC Among targeted body region box CBox=[Xleft,Ytop,Xright,Ybottom].Targeted body region acquired in the above method Box has accurate and reliable, strong robustness, and calculates simple feature.
According to the box C of acquired targeted body regionBox=[Xleft,Ytop,Xright,Ybottom], it can be by body mesh Mark region segmentation is the identical sub-box C of M*N sizesubBOX(m,n).Count each sub-box C respectively againsubBOX(m,n)Corresponding position The gradient information DCM for the four direction setG_X,DCMG_Y,DCMG_XY,DCMG_YX, and DCMLBPIt is counted respectively, to generate Feature histogram distribution vector HG_X(m,n),HG_Y(m,n),HG_XY(m,n),HG_YX(m,n),HLBP(m,n).By above-mentioned all feature histograms point Cloth vector merges according to the following equation, to generate corresponding DCMIGlobal characteristics vector HG_LBP。
HG_LBP=concat ([HG_X(m,n),HG_Y(m,n),HG_XY(m,n),HG_YX(m,n),HLBP(m,n)])|m:1->M,n:1->N
Due to obtaining the position of body profile by processing step, from the global characteristics vector of CT slice extraction HG_LBP, different body contour regions are eliminated in the position difference of different CT sectioning images.Although the same area of different people There are certain individual differences for body make-up, but the CT sectioning image of the same area of different people still has phase on the whole Like property, therefore calculate resulting global characteristics vector HG_LBPThis different people can be accurately described to cut in the CT of same area Entirety similitude possessed by picture.
Finally by formula argmin (dist (Ci, the H between vectorG_LBP))|Ci∈{C1,C2,…,CK}It calculates and the overall situation Feature vector HG_LBPThe classification sequence number i corresponding to the shortest CioptAs the CT sectioning image two-dimensional array DCMIIt is affiliated Classification, the Ci be the class template collection { C1, C2 ..., CK } classification i global characteristics vector.The class template Collection { C1, C2 ..., CK } generates by way of can marking clustering algorithm or manually.
Therefore use above-mentioned processing step that can quickly extract CT with lesser calculation amount, lower computation complexity Sectioning image two-dimensional array DCMICorresponding global characteristics vector HG_LBP, and then looked among the class template collection constructed in advance It finds out and HG_LBPClassification corresponding to immediate Ci is final classification result.And it reduces and the performance of software and hardware is wanted It asks, it can be with save the cost.In addition, the processing step also reduces the difficulty of exploitation, meet to the processing of high speed large-scale data The requirement of mode.
Further, the step S30, the embodiment based on above-mentioned Fig. 1 is cut using the operator Prewitt_X calculating CT Picture two-dimensional array DCMIGradient distribution figure DCMG_X, the operator Prewitt_X=[- 101, -1 01, -1 0 1], the CT sectioning image two-dimensional array DCM is calculated using operator Prewitt_YIGradient distribution figure DCMG_Y, the operator Prewitt_Y=[- 1-1-1,000,11 1] calculates the CT sectioning image two-dimemsional number using operator Prewitt_XY Group DCMIGradient distribution figure DCMG_XY, the operator Prewitt_XY=[0 11 ,-1 01 ,-1-1 0] uses operator Prewitt_YX calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_YX, the operator Prewitt_YX =[- 1-1 0 ,-1 01,01 1], calculates the CT sectioning image two-dimensional array DCMILocal binary patterns distribution map DCMLBP。
The image border of the opposite weighted average filtering of Prewitt operator and detection that belong to average filter is likely larger than 2 The Sobel operator of pixel, which has, to be calculated simply, and calculating speed is fast, and is more not susceptible to the benefit of noise jamming.
Further, the step S30, the embodiment based on above-mentioned Fig. 1 is cut using the operator Roberts_X calculating CT Picture two-dimensional array DCMIGradient distribution figure DCMG_X, the operator Roberts_X=[1-1,1-1] uses operator Roberts_Y calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_Y, the operator Roberts_Y= [- 1-1,1 1] calculate the CT sectioning image two-dimensional array DCM using operator Roberts_XYIGradient distribution figure DCMG_XY, the operator Roberts_XY=[1 0,0-1] uses operator Roberts_YX to calculate the CT sectioning image two Dimension group DCMIGradient distribution figure DCMG_YX, the operator Roberts_YX=[0 1, -1 0] calculates the CT sectioning image Two-dimensional array DCMILocal binary patterns distribution map DCMLBP。
Roberts operator gradient calculated is more accurate with respect to Prewitt operator and Sobel operator, and gradient value It positions more acurrate.Additionally due to Roberts operator is the operator of 2*2, thus Prewitt operator of the calculating speed with respect to 3*3 and Sobel operator is faster.
Further, the step S60, the embodiment based on above-mentioned Fig. 1 passes through formula argmin (distKL(Ci, HG_LBP))|Ci∈{C1,C2,…,CK}It calculates and the global characteristics vector HG_LBPThe classification sequence number i corresponding to the shortest CioptMake For the CT sectioning image DCMIAffiliated classification, the Ci are the complete of the class template collection { C1, C2 ..., CK } interior classification i Office's feature vector, the K are the sum of the classification, the global characteristics vector HG_LBPWith the global characteristics of the classification i to Measure the distance calculation formula dist of CiKL(Ci,HG_LBP)=(ΣJ=1:L(Ci[j]*log(Ci[j]/HG_LBP[j]))+ΣJ=1:L (HG_LBP[j]*log(HG_LBP[j]/Ci [j])))/2, the j is the global characteristics vector HG_LBPAnd global characteristics vector The serial number of Ci inner element, the L are the global characteristics vector HG_LBPAnd the length of global characteristics vector Ci.
Due to global characteristics vector HG_LBPBelong to a kind of statistical information, also can be regarded as a kind of distribution.Therefore relatively simple Single Euclidean distance, distance calculation formula distKL(Ci,HG_LBP) used by KL distance between two distribution vectors can be more Add the accurate similarity degree between two vectors of reliable evaluation.
Further, the embodiment based on above-mentioned Fig. 1, in the step S10, the CT sectioning image two-dimensional array DCMI's The value range of all array elements turns to [0,1024] by canonical.
Different CT sectioning image two-dimensional array DCM can be more effectively eliminated by regularizationIDifference, improve classification Accuracy.
The automatic classification method of the above-mentioned CT sectioning image of the present embodiment can be by the automatic classification of following CT sectioning image Device is realized.
As shown in Fig. 2, the apparatus for automatically sorting 100 of the CT sectioning image includes:
Targeted body region module 10, for being calculated using maximum between-cluster variance algorithm by the CT sectioning image two of input Dimension group DCMIIt carries out background area and targeted body region carries out the threshold value V of optimal segmentationTH, according to the threshold value VTHIt will be described CT sectioning image two-dimensional array DCMIIt carries out binary conversion treatment and obtains bianry image two-dimensional array DCMB, to the bianry image two Dimension group DCMBIt carries out the processing of two-value opening operation and obtains contour images two-dimensional array DCMC, from the contour images two-dimensional array DCMCExtract the box C comprising the targeted body regionBox=[Xleft,Ytop,Xright,Ybottom], the XleftFor the side Frame CBOXLeft margin, the Y in X-directiontopFor the box CBOXCoboundary, the X in Y directionrightFor the side Frame CBOXRight margin, the Y in the X-directionbottomFor the box CBOXIn the lower boundary of the Y direction；
Sub-box divides module 20, is used for the box CBOXIt is divided into the identical sub-box C of M*N sizesubBOX(m,n) =[Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H], the M is by the box CBOXIn the X-axis side To the number of equal part, the N is by box CBOXIn the number of the Y direction equal part, the m is the sub-box CsubBOX(m,n)In the serial number of the X-direction, the value range of the m is [1, M], and the n is the sub-box CsubBOX(m,n) In the serial number of the Y direction, the value range of the n is [1, N], and the W is the sub-box CsubBOX(m,n)In the X-axis The width in direction, the W=(Xright-Xleft)/M, the H are the sub-box CsubBOX(m,n)In the height of the Y direction, H=(the Ybottom-Ytop)/N；
Characteristic extracting module 30, for calculating the CT sectioning image two-dimensional array DCM using operator Sobel_XILadder Degree distribution two-dimensional array DCMG_X, the operator Sobel_X=[- 101, -2 02, -1 0 1] counted using operator Sobel_Y Calculate the CT sectioning image two-dimensional array DCMIGradient distribution two-dimensional array DCMG_Y, the operator Sobel_Y=[- 1-2- 1,000,12 1], the CT sectioning image two-dimensional array DCM is calculated using operator Sobel_XYIGradient distribution two-dimemsional number Group DCMG_XY, the operator Sobel_XY=[0 12 ,-1 01 ,-2-1 0] uses operator Sobel_YX to calculate the CT Sectioning image two-dimensional array DCMIGradient distribution two-dimensional array DCMG_YX, the operator Sobel_YX=[- 2-1 0 ,-1 0 1,01 2], the CT sectioning image two-dimensional array DCM is calculatedILocal binary patterns be distributed two-dimensional array DCMLBP；
Sub-box characteristic module 40 is used for from the gradient distribution two-dimensional array DCMG_XCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_X(m,n), from the gradient distribution two-dimensional array DCMG_YCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m- 1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_Y(m,n), from the gradient distribution two dimension Array DCMG_XYCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m* W,Ytop+ n*H] gradient histogram distribution vector HG_XY(m,n), from the gradient distribution two-dimensional array DCMG_YXIt calculates described in corresponding to Sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram point Cloth vector HG_YX(m,n), two-dimensional array DCM is distributed from local binary patternsLBPCalculate the corresponding sub-box CsubBOX(m,n)Institute is in place Set [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] local binary patterns histogram distribution vector HLBP(m,n)；
Global characteristics module 50, for passing through formula HG_LBP=concat ([HG_X(m,n),HG_Y(m,n),HG_XY(m,n), HG_YX(m,n),HLBP(m,n)])|m:1->M,n:1->NBy the gradient histogram distribution vector HG_X(m,n), the gradient histogram distribution vector HG_Y(m,n), the gradient histogram distribution vector HG_XY(m,n), the gradient histogram distribution vector HG_YX(m,n), the local binary mould Formula histogram distribution vector HLBP(m,n)Carry out vector merging, the HG_LBPFor the CT sectioning image two-dimensional array DCMIThe overall situation it is special Vector is levied, the concat is vector pooled function；
Classification discrimination module 60, for passing through formula argmin (dist (Ci, HG_LBP))|Ci∈{C1,C2,…,CK}Calculating and institute State global characteristics vector HG_LBPThe classification sequence number i corresponding to the shortest CioptAs the CT sectioning image two-dimensional array DCMIAffiliated classification, the Ci are the global characteristics vector of the class template collection { C1, C2 ..., CK } interior classification i, the K For the sum of the classification, the global characteristics vector HG_LBPIt is public with being calculated at a distance from the global characteristics vector Ci of the classification i Formula dist (Ci, HG_LBP)=| | Ci-HG_LBP||2。
The actual physical meaning of the pixel value (referred to as CT value) of CT sectioning image is the density of tissue or organ, meter Amount unit is HU (hounsfield unit), and the CT value of general air is about -1000, the CT of the highest bone of body density Value close+1000.Probably there are two main wave crests, a wave crest institutes for the histogram distribution of the pixel value of one CT sectioning image Regional Representative's air of distribution, and Regional Representative's bodily tissue that another wave crest is distributed.Therefore maximum between-cluster variance is used Algorithm (OTSU algorithm) can search out the threshold value V of optimal segmentationTHCT sectioning image is subjected to binary conversion treatment, by DCMIAmong Pixel value is lower than threshold value VTHCorresponding bianry image DCMBThe pixel value of same coordinate position be set as 0, that is, determine that the pixel is Air；By DCMIAmong pixel value be lower than threshold value VTHCorresponding bianry image DCMBThe pixel value of same coordinate position be set as 1, Determine the pixel for bodily tissue.Again by DCMBTwo-value opening operation is carried out to handle to obtain DCMC, i.e., to DCMBRow first carries out The processing of two-value erosion algorithm, then carries out two-value expansion algorithm processing, to filter out DCM againBInterior isolated noise spot, and And it is smooth to get to body contour DCM to obtain the profile of physical feelingC.It can be obtained by by contours extract algorithm in DCMC Among targeted body region box CBox=[Xleft,Ytop,Xright,Ybottom].Targeted body region acquired in the above method Box has accurate and reliable, strong robustness, and calculates simple feature.
According to the box C of acquired targeted body regionBox=[Xleft,Ytop,Xright,Ybottom], it can be by body mesh Mark region segmentation is the identical sub-box C of M*N sizesubBOX(m,n).Count each sub-box C respectively againsubBOX(m,n)Corresponding position The gradient information DCM for the four direction setG_X,DCMG_Y,DCMG_XY,DCMG_YX, and DCMLBPIt is counted respectively, to generate Feature histogram distribution vector HG_X(m,n),HG_Y(m,n),HG_XY(m,n),HG_YX(m,n),HLBP(m,n).By above-mentioned all feature histograms point Cloth vector is according to HG_LBPFormula merges, to generate corresponding DCMIGlobal characteristics vector HG_LBP.Due to passing through body Target area module 10 obtains the position of body profile, therefore the global characteristics vector H extracted from CT sliceG_LBP, eliminate not With body contour region different CT sectioning images position difference.Although there are one for the body make-up of the same area of different people Fixed individual difference, but the CT sectioning image of the same area of different people still has similitude on the whole, therefore calculates Resulting global characteristics vector HG_LBPThis different people can accurately be described to be had in the CT sectioning image of same area Whole similitude.
Finally by formula argmin (dist (Ci, the H between vectorG_LBP))|Ci∈{C1,C2,…,CK}It calculates and the overall situation Feature vector HG_LBPThe classification sequence number i corresponding to the shortest CioptAs the CT sectioning image two-dimensional array DCMIIt is affiliated Classification, the Ci be the class template collection { C1, C2 ..., CK } classification i global characteristics vector.The class template Collection { C1, C2 ..., CK } generates by way of can marking clustering algorithm or manually.
Therefore use above-mentioned processing module that can quickly extract CT with lesser calculation amount, lower computation complexity Sectioning image two-dimensional array DCMICorresponding global characteristics vector HG_LBP, and then looked among the class template collection constructed in advance It finds out and HG_LBPClassification corresponding to immediate Ci is final classification result.And it reduces and the performance of software and hardware is wanted It asks, it can be with save the cost.In addition, the processing step also reduces the difficulty of exploitation, meet to the processing of high speed large-scale data The requirement of mode.
Further, among the characteristic extracting module: calculating the CT sectioning image two-dimemsional number using operator Prewitt_X Group DCMIGradient distribution figure DCMG_X, the operator Prewitt_X=[- 101, -1 01, -1 0 1] uses operator Prewitt_Y calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_Y, the operator Prewitt_Y= [- 1-1-1,000,11 1] calculate the CT sectioning image two-dimensional array DCM using operator Prewitt_XYIGradient Distribution map DCMG_XY, the operator Prewitt_XY=[0 11 ,-1 01 ,-1-1 0] counted using operator Prewitt_YX Calculate the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_YX, the operator Prewitt_YX=[- 1-1 0 ,-1 01,01 1], the CT sectioning image two-dimensional array DCM is calculatedILocal binary patterns distribution map DCMLBP。
The image border of the opposite weighted average filtering of Prewitt operator and detection that belong to average filter is likely larger than 2 The Sobel operator of pixel, which has, to be calculated simply, and calculating speed is fast, and is more not susceptible to the benefit of noise jamming.
Further, among the characteristic extracting module 30: calculating the CT sectioning image two dimension using operator Roberts_X Array DCMIGradient distribution figure DCMG_X, the operator Roberts_X=[1-1,1-1] counted using operator Roberts_Y Calculate the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_Y, the operator Roberts_Y=[- 1-1,1 1], The CT sectioning image two-dimensional array DCM is calculated using operator Roberts_XYIGradient distribution figure DCMG_XY, the operator Roberts_XY=[1 0,0-1] calculates the CT sectioning image two-dimensional array DCM using operator Roberts_YXIGradient Distribution map DCMG_YX, the operator Roberts_YX=[0 1, -1 0] calculates the CT sectioning image two-dimensional array DCMIOffice Portion binary pattern distribution map DCMLBP。
Roberts operator gradient calculated is more accurate with respect to Prewitt operator and Sobel operator, and gradient value It positions more acurrate.Additionally due to Roberts operator is the operator of 2*2, thus Prewitt operator of the calculating speed with respect to 3*3 and Sobel operator is faster.
Further, among the classification discrimination module 60: passing through formula argmin (distKL(Ci,HG_LBP)) |Ci∈{C1,C2,…,CK}It calculates and the global characteristics vector HG_LBPThe classification sequence number i corresponding to the shortest CioptAs described CT sectioning image DCMIAffiliated classification, the Ci are the global characteristics of the class template collection { C1, C2 ..., CK } interior classification i Vector, the K are the sum of the classification, the global characteristics vector HG_LBPWith the global characteristics vector Ci's of the classification i Distance calculation formula distKL(Ci,HG_LBP)=(ΣJ=1:L(Ci[j]*log(Ci[j]/HG_LBP[j]))+ΣJ=1:L(HG_LBP [j]*log(HG_LBP[j]/Ci [j])))/2, the j is the global characteristics vector HG_LBPAnd inside global characteristics vector Ci The serial number of element, the L are the global characteristics vector HG_LBPAnd the length of global characteristics vector Ci.
Due to global characteristics vector HG_LBPBelong to a kind of statistical information, also can be regarded as a kind of distribution.Therefore relatively simple Single Euclidean distance, distance calculation formula distKL(Ci,HG_LBP) used by KL distance between two distribution vectors can be more Add the accurate similarity degree between two vectors of reliable evaluation.
Further, among the targeted body region module 10: the CT sectioning image two-dimensional array DCMIAll numbers The value range of group element turns to [0,1024] by canonical.
Different CT sectioning image two-dimensional array DCM can be more effectively eliminated by regularizationIDifference, improve classification Accuracy.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art The part contributed out can be embodied in the form of software products, which is stored in a storage medium In (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that a terminal device (can be mobile phone, computer, clothes Business device, air conditioner or the network equipment etc.) execute method described in each embodiment of the present invention.
Embodiment described above is only the preferred embodiments of the invention, and but not intended to limit the scope of the present invention, therefore All shapes according to the present invention change made by principle, should all be included within the scope of protection of the present invention.

Claims (10)

  1. The automatic classification method of 1.CT sectioning image, which comprises the following steps:
    Step 1 is calculated using maximum between-cluster variance algorithm by the CT sectioning image two-dimensional array DCM of inputICarry out background area The threshold value V of optimal segmentation is carried out with targeted body regionTH, according to the threshold value VTHBy the CT sectioning image two-dimensional array DCMI It carries out binary conversion treatment and obtains bianry image two-dimensional array DCMB, to the bianry image two-dimensional array DCMBIt carries out two-value and opens fortune It calculates processing and obtains contour images two-dimensional array DCMC, from the contour images two-dimensional array DCMCExtracting includes the physical target The box C in regionBox=[Xleft,Ytop,Xright,Ybottom], the XleftFor the box CBOXX-direction left margin, The YtopFor the box CBOXCoboundary, the X in Y directionrightFor the box CBOXOn the right side of the X-direction Boundary, the YbottomFor the box CBOXIn the lower boundary of the Y direction；
    Step 2, by the box CBOXIt is divided into the identical sub-box C of M*N sizesubBOX(m,n)=[Xleft+(m-1)*W,Ytop +(n-1)*H,Xleft+m*W,Ytop+ n*H], the M is by the box CBOXIn the number of the X-direction equal part, the N For by box CBOXIn the number of the Y direction equal part, the m is the sub-box CsubBOX(m,n)In the sequence of the X-direction Number, the value range of the m is [1, M], and the n is the sub-box CsubBOX(m,n)In the serial number of the Y direction, the n Value range be [1, N], the W be the sub-box CsubBOX(m,n)In the width of the X-direction, the W=(Xright- Xleft)/M, the H are the sub-box CsubBOX(m,n)In the height of the Y direction, the H=(Ybottom-Ytop)/N；
    Step 3 calculates the CT sectioning image two-dimensional array DCM using operator Sobel_XIGradient distribution two-dimensional array DCMG_X, the operator Sobel_X=[- 101, -2 02, -1 0 1] uses operator Sobel_Y to calculate the CT slice map As two-dimensional array DCMIGradient distribution two-dimensional array DCMG_Y, the operator Sobel_Y=[- 1-2-1,000,12 1], the CT sectioning image two-dimensional array DCM is calculated using operator Sobel_XYIGradient distribution two-dimensional array DCMG_XY, described Operator Sobel_XY=[0 12 ,-1 01 ,-2-1 0] calculates the CT sectioning image two-dimemsional number using operator Sobel_YX Group DCMIGradient distribution two-dimensional array DCMG_YX, the operator Sobel_YX=[- 2-1 0 ,-1 01,01 2], calculating institute State CT sectioning image two-dimensional array DCMILocal binary patterns be distributed two-dimensional array DCMLBP；
    Step 4, from the gradient distribution two-dimensional array DCMG_XCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+ (m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_X(m,n), from the gradient distribution Two-dimensional array DCMG_YCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft +m*W,Ytop+ n*H] gradient histogram distribution vector HG_Y(m,n), from the gradient distribution two-dimensional array DCMG_XYCalculate corresponding institute State sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram Distribution vector HG_XY(m,n), from the gradient distribution two-dimensional array DCMG_YXCalculate the corresponding sub-box CsubBOX(m,n)Institute is in place Set [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_YX(m,n), from part Binary pattern is distributed two-dimensional array DCMLBPCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+ (n-1)*H,Xleft+m*W,Ytop+ n*H] local binary patterns histogram distribution vector HLBP(m,n)；
    Step 5 passes through formula HG_LBP=concat ([HG_X(m,n),HG_Y(m,n),HG_XY(m,n),HG_YX(m,n),HLBP(m,n)]) |m:1->M,n:1->NBy the gradient histogram distribution vector HG_X(m,n), the gradient histogram distribution vector HG_Y(m,n), the gradient it is straight Square distribution vector HG_XY(m,n), the gradient histogram distribution vector HG_YX(m,n), the local binary patterns histogram distribution vector HLBP(m,n)Carry out vector merging, the HG_LBPFor the CT sectioning image two-dimensional array DCMIGlobal characteristics vector, it is described Concat is vector pooled function；
    Step 6 passes through formulaIt calculates and the global characteristics vector HG_LBPApart from shortest CiCorresponding classification sequence number ioptAs the CT sectioning image two-dimensional array DCMIAffiliated classification, The CiFor the class template collection { C1,C2,…,CKInterior classification i global characteristics vector, the K is the total of the classification Number, the global characteristics vector HG_LBPWith the global characteristics vector C of the classification iiDistance calculation formula dist (Ci,HG_LBP) =| | Ci-HG_LBP||2。
  2. 2. the automatic classification method of CT sectioning image according to claim 1, it is characterised in that: in step 3, use calculation Sub- Prewitt_X calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_X, the operator Prewitt_X =[- 101, -1 01, -1 0 1] calculates the CT sectioning image two-dimensional array DCM using operator Prewitt_YILadder Spend distribution map DCMG_Y, the operator Prewitt_Y=[- 1-1-1,000,11 1] counted using operator Prewitt_XY Calculate the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_XY, the operator Prewitt_XY=[0 11, -1 0 1 ,-1-1 0], the CT sectioning image two-dimensional array DCM is calculated using operator Prewitt_YXIGradient distribution figure DCMG_YX, The operator Prewitt_YX=[- 1-1 0 ,-1 01,01 1], calculates the CT sectioning image two-dimensional array DCMIOffice Portion binary pattern distribution map DCMLBP。
  3. 3. the automatic classification method of CT sectioning image according to claim 1, it is characterised in that: in step 3, use calculation Sub- Roberts_X calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_X, the operator Roberts_X =[1-1,1-1] calculates the CT sectioning image two-dimensional array DCM using operator Roberts_YIGradient distribution figure DCMG_Y, the operator Roberts_Y=[- 1-1,1 1] uses operator Roberts_XY to calculate CT sectioning image two dimension Array DCMIGradient distribution figure DCMG_XY, the operator Roberts_XY=[1 0,0-1] counted using operator Roberts_YX Calculate the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_YX, the operator Roberts_YX=[0 1, -1 0], Calculate the CT sectioning image two-dimensional array DCMILocal binary patterns distribution map DCMLBP。
  4. 4. the automatic classification method of CT sectioning image according to any one of claims 1 to 3, it is characterised in that: in step 6 In, pass through formulaIt calculates and the global characteristics vector HG_LBPAway from From shortest CiCorresponding classification sequence number ioptAs the CT sectioning image DCMIAffiliated classification, the CiFor the classification Template set { C1,C2,…,CKInterior classification i global characteristics vector, the K is the sum of the classification, the global characteristics to Measure HG_LBPWith the global characteristics vector C of the classification iiDistance calculation formula distKL(Ci,HG_LBP)=(ΣJ=1:L(Ci[j]* log(Ci[j]/HG_LBP[j]))+ΣJ=1:L(HG_LBP[j]*log(HG_LBP[j]/Ci[j])))/2, the j is the global characteristics Vector HG_LBPAnd the global characteristics vector CiThe serial number of inner element, the L are the global characteristics vector HG_LBPAnd The global characteristics vector CiLength.
  5. 5. the automatic classification method of CT sectioning image according to claim 1, it is characterised in that: in step 1, the CT Sectioning image two-dimensional array DCMIThe value ranges of all array elements [0,1024] is turned to by canonical.
  6. The automatic classification of 6.CT sectioning image is set characterized by comprising
    Targeted body region module, for being calculated using maximum between-cluster variance algorithm by the CT sectioning image two-dimensional array of input DCMIIt carries out background area and targeted body region carries out the threshold value V of optimal segmentationTH, according to the threshold value VTHThe CT is sliced Two-dimensional image array DCMIIt carries out binary conversion treatment and obtains bianry image two-dimensional array DCMB, to the bianry image two-dimensional array DCMBIt carries out the processing of two-value opening operation and obtains contour images two-dimensional array DCMC, from the contour images two-dimensional array DCMCIt extracts Box C comprising the targeted body regionBox=[Xleft,Ytop,Xright,Ybottom], the XleftFor the box CBOXIn X The left margin of axis direction, the YtopFor the box CBOXCoboundary, the X in Y directionrightFor the box CBOX? The right margin of the X-direction, the YbottomFor the box CBOXIn the lower boundary of the Y direction；
    Sub-box divides module, is used for the box CBOXIt is divided into the identical sub-box C of M*N sizesubBOX(m,n)= [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H], the M is by the box CBOXIn the X-direction The number of equal part, the N are by box CBOXIn the number of the Y direction equal part, the m is the sub-box CsubBOX(m,n) In the serial number of the X-direction, the value range of the m is [1, M], and the n is the sub-box CsubBOX(m,n)In the Y-axis The serial number in direction, the value range of the n are [1, N], and the W is the sub-box CsubBOX(m,n)In the width of the X-direction Degree, the W=(Xright-Xleft)/M, the H are the sub-box CsubBOX(m,n)In the height of the Y direction, the H= (Ybottom-Ytop)/N；
    Characteristic extracting module, for calculating the CT sectioning image two-dimensional array DCM using operator Sobel_XIGradient distribution two Dimension group DCMG_X, the operator Sobel_X=[- 101, -2 02, -1 0 1] uses operator Sobel_Y to calculate the CT Sectioning image two-dimensional array DCMIGradient distribution two-dimensional array DCMG_Y, the operator Sobel_Y=[- 1-2-1,000, 12 1], the CT sectioning image two-dimensional array DCM is calculated using operator Sobel_XYIGradient distribution two-dimensional array DCMG_XY, The operator Sobel_XY=[0 12 ,-1 01 ,-2-1 0] calculates the CT sectioning image two using operator Sobel_YX Dimension group DCMIGradient distribution two-dimensional array DCMG_YX, the operator Sobel_YX=[- 2-1 0 ,-1 01,01 2], meter Calculate the CT sectioning image two-dimensional array DCMILocal binary patterns be distributed two-dimensional array DCMLBP；
    Sub-box characteristic module is used for from the gradient distribution two-dimensional array DCMG_XCalculate the corresponding sub-box CsubBOX(m,n) Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_X(m,n), From the gradient distribution two-dimensional array DCMG_YCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W, Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_Y(m,n), from the gradient distribution two-dimensional array DCMG_XYCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop + n*H] gradient histogram distribution vector HG_XY(m,n), from the gradient distribution two-dimensional array DCMG_YXCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] gradient histogram distribution vector HG_YX(m,n), two-dimensional array DCM is distributed from local binary patternsLBPCalculate the corresponding sub-box CsubBOX(m,n)Position [Xleft+(m-1)*W,Ytop+(n-1)*H,Xleft+m*W,Ytop+ n*H] local binary patterns histogram distribution vector HLBP(m,n)；
    Global characteristics module, for passing through formula HG_LBP=concat ([HG_X(m,n),HG_Y(m,n),HG_XY(m,n),HG_YX(m,n), HLBP(m,n)])|m:1->M,n:1->NBy the gradient histogram distribution vector HG_X(m,n), the gradient histogram distribution vector HG_Y(m,n), institute State gradient histogram distribution vector HG_XY(m,n), the gradient histogram distribution vector HG_YX(m,n), the local binary patterns histogram point Cloth vector HLBP(m,n)Carry out vector merging, the HG_LBPFor the CT sectioning image two-dimensional array DCMIGlobal characteristics vector, The concat is vector pooled function；
    Classification discrimination module, for passing through formulaIt calculates and described complete Office feature vector HG_LBPApart from shortest CiCorresponding classification sequence number ioptAs the CT sectioning image two-dimensional array DCMIInstitute The classification of category, the CiFor the class template collection { C1,C2,…,CKInterior classification i global characteristics vector, the K is described The sum of classification, the global characteristics vector HG_LBPWith the global characteristics vector C of the classification iiDistance calculation formula dist (Ci,HG_LBP)=| | Ci-HG_LBP||2。
  7. 7. the automatic classification of CT sectioning image according to claim 6 is set, it is characterised in that: in the characteristic extracting module In, the CT sectioning image two-dimensional array DCM is calculated using operator Prewitt_XIGradient distribution figure DCMG_X, the operator Prewitt_X=[- 101, -1 01, -1 0 1] calculates the CT sectioning image two-dimensional array using operator Prewitt_Y DCMIGradient distribution figure DCMG_Y, the operator Prewitt_Y=[- 1-1-1,000,11 1] uses operator Prewitt_XY calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_XY, the operator Prewitt_XY =[0 11 ,-1 01 ,-1-1 0] calculates the CT sectioning image two-dimensional array DCM using operator Prewitt_YXILadder Spend distribution map DCMG_YX, the operator Prewitt_YX=[- 1-1 0 ,-1 01,01 1] calculates the CT sectioning image Two-dimensional array DCMILocal binary patterns distribution map DCMLBP。
  8. 8. the automatic classification of CT sectioning image according to claim 6 is set, it is characterised in that: in the characteristic extracting module In, the CT sectioning image two-dimensional array DCM is calculated using operator Roberts_XIGradient distribution figure DCMG_X, the operator Roberts_X=[1-1,1-1] calculates the CT sectioning image two-dimensional array DCM using operator Roberts_YIGradient Distribution map DCMG_Y, the operator Roberts_Y=[- 1-1,1 1] uses operator Roberts_XY to calculate the CT slice map As two-dimensional array DCMIGradient distribution figure DCMG_XY, the operator Roberts_XY=[1 0,0-1] uses operator Roberts_YX calculates the CT sectioning image two-dimensional array DCMIGradient distribution figure DCMG_YX, the operator Roberts_YX =[0 1, -1 0], calculates the CT sectioning image two-dimensional array DCMILocal binary patterns distribution map DCMLBP。
  9. 9. the automatic classification according to the described in any item CT sectioning images of claim 6 to 8 is set, it is characterised in that: in the class In other discrimination module, pass through formulaIt calculates and the global characteristics Vector HG_LBPApart from shortest CiCorresponding classification sequence number ioptAs the CT sectioning image DCMIAffiliated classification, the Ci For the class template collection { C1,C2,…,CKInterior classification i global characteristics vector, the K is the sum of the classification, described Global characteristics vector HG_LBPWith the global characteristics vector C of the classification iiDistance calculation formula distKL(Ci,HG_LBP)= (ΣJ=1:L(Ci[j]*log(Ci[j]/HG_LBP[j]))+ΣJ=1:L(HG_LBP[j]*log(HG_LBP[j]/Ci[j])))/2, the j For the global characteristics vector HG_LBPAnd the global characteristics vector CiThe serial number of inner element, the L are described global special Levy vector HG_LBPAnd the global characteristics vector CiLength.
  10. 10. the automatic classification according to the described in any item CT sectioning images of claim 6 to 8 is set, it is characterised in that: in the body In the module of body target area, the CT sectioning image two-dimensional array DCMIThe value ranges of all array elements turned to by canonical [0,1024]。
CN201910302815.8A 2019-04-16 2019-04-16 Automatic classification method and device for CT slice images Expired - Fee Related CN110136100B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910302815.8A CN110136100B (en) 2019-04-16 2019-04-16 Automatic classification method and device for CT slice images

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910302815.8A CN110136100B (en) 2019-04-16 2019-04-16 Automatic classification method and device for CT slice images

Publications (2)

Publication Number Publication Date
CN110136100A true CN110136100A (en) 2019-08-16
CN110136100B CN110136100B (en) 2021-02-19

Family

ID=67570183

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910302815.8A Expired - Fee Related CN110136100B (en) 2019-04-16 2019-04-16 Automatic classification method and device for CT slice images

Country Status (1)

Country Link
CN (1) CN110136100B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102782708A (en) * 2009-12-02 2012-11-14 高通股份有限公司 Fast subspace projection of descriptor patches for image recognition
CN104299005A (en) * 2013-07-15 2015-01-21 深圳市振邦实业有限公司 Head detection method and system
CN105930791A (en) * 2016-04-19 2016-09-07 重庆邮电大学 Road traffic sign identification method with multiple-camera integration based on DS evidence theory
US20170132773A1 (en) * 2014-07-29 2017-05-11 Hicof Inc. Method and apparatus for proving and determining an integrity status of an outer packaging
CN109145947A (en) * 2018-07-17 2019-01-04 昆明理工大学 A kind of Fashionable women dress image fine grit classification method based on component detection and visual signature

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102782708A (en) * 2009-12-02 2012-11-14 高通股份有限公司 Fast subspace projection of descriptor patches for image recognition
CN104299005A (en) * 2013-07-15 2015-01-21 深圳市振邦实业有限公司 Head detection method and system
US20170132773A1 (en) * 2014-07-29 2017-05-11 Hicof Inc. Method and apparatus for proving and determining an integrity status of an outer packaging
CN105930791A (en) * 2016-04-19 2016-09-07 重庆邮电大学 Road traffic sign identification method with multiple-camera integration based on DS evidence theory
CN109145947A (en) * 2018-07-17 2019-01-04 昆明理工大学 A kind of Fashionable women dress image fine grit classification method based on component detection and visual signature

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
RAKESH MEHTA 等: "Dominant Rotated Local Binary Patterns (DRLBP) for Texture Classification", 《PATTERN RECOGNITION LETTERS 》 *
刘晓虹 等: "基于改进多尺度LBP算法的肝脏CT图像特征提取方法", 《计算机科学》 *
布社辉 等: "基于运动目标检测的交互式图片浏览系统", 《华南理工大学学报(自然科学版)》 *

Also Published As

Publication number Publication date
CN110136100B (en) 2021-02-19

Similar Documents

Publication Publication Date Title
CN109409190A (en) Pedestrian detection method based on histogram of gradients and Canny edge detector
Jahanifar et al. Supervised saliency map driven segmentation of lesions in dermoscopic images
CN105740945B (en) A kind of people counting method based on video analysis
Viswanathan Fuzzy C means detection of leukemia based on morphological contour segmentation
Dahab et al. Automated brain tumor detection and identification using image processing and probabilistic neural network techniques
CN104715238B (en) A kind of pedestrian detection method based on multi-feature fusion
CN103632132B (en) Face detection and recognition method based on skin color segmentation and template matching
CN109034045A (en) A kind of leucocyte automatic identifying method based on convolutional neural networks
CN105574527B (en) A kind of quick object detecting method based on local feature learning
CN110120056B (en) Blood leukocyte segmentation method based on adaptive histogram threshold and contour detection
CN108537751B (en) Thyroid ultrasound image automatic segmentation method based on radial basis function neural network
CN104123554B (en) SIFT image characteristic extracting methods based on MMTD
CN107886507B (en) A kind of salient region detecting method based on image background and spatial position
CN106682641A (en) Pedestrian identification method based on image with FHOG- LBPH feature
Cao et al. A robust parameter-free thresholding method for image segmentation
CN109461163A (en) A kind of edge detection extraction algorithm for magnetic resonance standard water mould
CN108734200A (en) Human body target visible detection method and device based on BING features
CN111815640B (en) Memristor-based RBF neural network medical image segmentation algorithm
CN107067037A (en) A kind of method that use LLC criterions position display foreground
CN104680189B (en) Based on the bad image detecting method for improving bag of words
CN109002850A (en) The method and device of fuel value of food in a kind of calculating image
Chen et al. Automated mammographic risk classification based on breast density estimation
CN103366183B (en) Nonparametric automatic detection method of focal niduses
CN110910497B (en) Method and system for realizing augmented reality map
Ahmad et al. Brain tumor detection & features extraction from MR images using segmentation, image optimization & classification techniques

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210219