CN108805892A - Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma - Google Patents
Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma Download PDFInfo
- Publication number
- CN108805892A CN108805892A CN201810558897.8A CN201810558897A CN108805892A CN 108805892 A CN108805892 A CN 108805892A CN 201810558897 A CN201810558897 A CN 201810558897A CN 108805892 A CN108805892 A CN 108805892A
- Authority
- CN
- China
- Prior art keywords
- heterogeneous
- distance
- swimming
- matrix
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- 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
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
-
- 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
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to medical image analysis field, heterogeneous quantitative depicting method in specially a kind of PET image nasopharynx carcinoma.The present invention includes：The PET/CT scan datas for obtaining patient, are converted into standard uptake value (SUV) image；Three-dimensional segmentation is carried out to Primary lesions of NPC；Sliding-model control is carried out to lesion, lesion builds 4 kinds of image group matrixes to treated, and considers 5 kinds of symmetry, Average Strategy, distance, neighborhood number and window width size parameter settings in matrix building process comprehensively；415 textural characteristics are extracted from the matrix built, in conjunction with 5 tradition SUV features to realize that heterogeneous the quantifying of intratumoral metabolism is portrayed.The information that the present invention is contained by fully excavating noninvasive PET image, the intratumoral metabolism for featuring nasopharyngeal carcinoma more fully hereinafter is heterogeneous, realizes that accurate nasopharyngeal carcinoma discriminating provides quantitative evaluation index for doctor.
Description
Technical field
The present invention relates to medical image analysis technical fields, and in particular to heterogeneous in a kind of PET image nasopharynx carcinoma
Quantitative depicting method.
Background technology
Nasopharyngeal carcinoma is that a kind of local characteristic being mainly in South China is sick, in every 100,000 people, up to 30 human hairs disease
。18F-FDG PET/CT have been widely used for the inspection of nasopharyngeal carcinoma, however, since benign inflammatory disorders tissue generally also can table
Reveal higher glucose uptake, image department doctor observes by the naked eye the metabolism degree of lesion and anatomic form cannot achieve tumor
Interior quantifying for image-region of heterogeneity is portrayed.Therefore the discriminating of early stage nasopharyngeal carcinoma is easily interfered by inflammatory disorders, leads to specificity
Less than 85%, the histopathology biopsy for still relying on wound is made a definite diagnosis.Textural characteristics are in the diagnosis of kinds cancer, curative effect
Assessment and prognosis prediction etc. show its potential value.However, the prior art for texture matrix construction method compared with
To be single, not comprehensively in view of the ginsengs such as symmetry, Average Strategy, distance, neighborhood number, window width size in matrix building process
The influences of number setting significantly limit in tumor the heterogeneous adequacy quantitatively portrayed and comprehensive.Therefore, it fully excavates noninvasive
The information that is contained of PET image, realize that heterogeneous quantifying is portrayed very necessary in nasopharynx carcinoma.
Invention content
In view of above-mentioned technical background, determine the purpose of the present invention is to provide heterogeneous in a kind of PET image nasopharynx carcinoma
Measure depicting method.
The above-mentioned purpose of the present invention is realized by following technological means：
Quantitative depicting method heterogeneous in a kind of PET image nasopharynx carcinoma is provided, including：
(1) original image of the DICOM format of target object PET/CT scannings is obtained, and the original image is carried out pre-
Processing；
(2) Primary lesions of NPC three-dimensional segmentation is carried out on standard uptake value image after treatment；
(3) SUV grayscale sliding-model controls are carried out to the three-dimensional lesion divided；And using many kinds of parameters setting structure texture
Matrix；
(4) it from texture matrix texture feature extraction, carries out heterogeneity in tumor and quantitatively portrays.
Preferably, the original image of the DICOM format for obtaining target object PET/CT scannings, and to described original
Image carries out pretreatment：It to PET image into row interpolation, is merged with CT image registrations, and uses formula (1) by PET activity figures
Be converted to standard uptake value image：
Preferably, step (2) is specifically to use SUV>2.5 threshold value carries out initial segmentation, is then adjusted so as to again to the end
3D divide lesion.
Preferably, the SUV grayscale sliding-model controls in step (3), using formula (2)：
Wherein, SUV (x) indicates that the standard uptake value of voxel x, B indicate discretization spacing, take 0.1, SUVDis(x) body is indicated
Standard uptake value after plain x discretizations.
Preferably, texture matrix is built in step (3), specifically includes four kinds of matrixes, respectively gray level co-occurrence matrixes, gray scale
Run-length matrix, gray areas dimension matrix and neighborhood difference matrix.
Preferably, three symmetry, Average Strategy and distance parameters, gray scale trip are included in gray level co-occurrence matrixes building process
Include Average Strategy parameter in journey matrix building process, include neighborhood number parameter in gray areas dimension matrix building process,
Include window width dimensional parameters in neighborhood difference matrix building process.
Preferably, many kinds of parameters setting that structure texture matrix uses, wherein symmetry are symmetrical and non-including constructing
Symmetrical GLCM；Average Strategy is that the structure of GLCM or GLRLM includes being built respectively from 13 directions, then takes 13 sides again
Further include considering that 13 directions only builds and obtain 1 GLCM or GLRLM simultaneously to being averaged for feature, when extraction feature without
It is average；10 pixels that distance is set as 1,2,3 ... build GLCM in turn；Neighborhood number is set as 6,18,26 and then builds
GLSZM；Window width is sized to 3,5,7,9,11 pixels and then builds NGTDM.
Preferably, texture feature extraction is 415 total, wherein comprising 338 GLCM features, 26 GLRLM features, 26
GLSZM features and 25 NGTDM features.
Preferably, the textural characteristics of extraction, specifically include：
26 GLCM features, including energy, entropy, poor entropy and entropy, variance and the difference of two squares and variance, maximum likelihood are right
Than degree, non-similarity, homogeney, inverse difference moment, correlation, difference variance, autocorrelation, cluster conspicuousness, cluster shade, cluster
Trend, information correlation amount 1, information correlation amount 2, inverse variance, standard inverse difference moment, standard unfavourable balance, and average 1, and average 2,
Consistency；
13 GLRLM features, short distance of swimming enhancing, long distance of swimming enhancing, gray scale inhomogeneities, distance of swimming inhomogeneities, the distance of swimming hundred
Point ratio, the enhancing of the low gray scale distance of swimming, high gray scale distance of swimming enhancing, the low grey level enhancement of the short distance of swimming, the short high grey level enhancement of the distance of swimming, the long distance of swimming are low
Grey level enhancement, the long high grey level enhancement of the distance of swimming, gray variance, distance of swimming variance；
13 GLSZM features, zonule enhancing, big region enhancing, gray scale inhomogeneities, zone nonuniformity, region hundred
Ratio, low gray level areas is divided to enhance, high gray areas enhances, the low grey level enhancement in zonule, the high grey level enhancement in zonule, and big region is low
Grey level enhancement, the big high grey level enhancement in region, gray variance, Local Deviation；
5 NGTDM features, roughness, contrast, lengthy and tedious degree, complexity, intensity.
Present invention simultaneously provides method heterogeneous in nasopharynx carcinoma is quantitatively portrayed based on PET image in nasopharyngeal carcinoma judgement side
The purposes in face.
Heterogeneous quantitative depicting method, can effectively realize in nasopharynx carcinoma in the PET image nasopharynx carcinoma of the present invention
Heterogeneous quantifying is portrayed.The present invention considers symmetry in matrix building process comprehensively, Average Strategy, distance, neighborhood number,
The influence of the parameter settings such as window width size enriches adequacy that PET image intratumoral metabolism heterogeneity is quantitatively portrayed and comprehensive.
Realize that accurate nasopharyngeal carcinoma discriminating provides quantitative evaluation index for auxiliary doctor.
Description of the drawings
Using attached drawing, the present invention is further illustrated, but the content in attached drawing does not constitute any limit to the present invention
System.
Fig. 1 is the schematic diagram of quantitative depicting method heterogeneous in PET image nasopharynx carcinoma of the present invention.
Fig. 2 is the schematic diagram of heterogeneous quantitative depicting method in embodiment 1PET image nasopharynx carcinomas.
Specific implementation mode
The invention will be further described with the following Examples.
Embodiment 1.
Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma, as shown in Figure 1, including：
First step obtains the original image of the DICOM format of target object PET/CT scannings, and to the original image
It is pre-processed.
Since PET image resolution ratio is relatively low, picture size is smaller, usually 128*128, and therefore, it is necessary to by DICOM format
PET image into row interpolation, so that it is registrated fusion with CT images (512*512), cubic spline interpolation may be used.In order to obtain
The standard uptake value respectively organized, it is also desirable to by what is rebuild, standard uptake value SUV figures are converted to by the activity figure of correction for attenuation
Picture, the present invention, which adopts to original image pre-process, includes：To PET image into row interpolation, merged with CT image registrations, and use
PET activity figures are converted to standard uptake value image by formula (1)：
Second step carries out Primary lesions of NPC three-dimensional segmentation on standard uptake value image after treatment, specifically
Using SUV>2.5 threshold value carries out initial segmentation, is then adjusted so as to 3D segmentation lesions to the end again.
Third step, build matrix before, in order to lifting matrixes structure speed, and also to so that different cases it
Between be comparable, it is also necessary to unified grayscale sliding-model control is carried out to the lesion divided, to the three-dimensional lesion divided
SUV grayscale sliding-model controls are carried out, then using many kinds of parameters setting structure texture matrix.
SUV grayscale sliding-model controls in step (3), using formula (2)：
Wherein, SUV (x) indicates that the standard uptake value of voxel x, B indicate discretization spacing, take 0.1, SUVDis(x) body is indicated
Standard uptake value after plain x discretizations.
It should be strongly noted that the building process of matrix considers symmetry, Average Strategy, distance, neighborhood comprehensively
Number, the influence of the parameter settings such as window width size.
1. the structure of gray level co-occurrence matrixes (GLCM)
The element P (i, j) of gray level co-occurrence matrixes indicates that two gray scales are respectively i and j, and the voxel that distance is D is in direction θ
The number of upper appearance.The structure of the matrix is related to the setting of three symmetry, Average Strategy and distance parameters.
1.1 symmetry
When the situation of the voxel pair on the only directions statistics θ, what is built is unsymmetrical matrix (being denoted as A), when simultaneously
When counting the situation of voxel pair on the negative direction in the directions θ and the directions θ, what is built is symmetrical matrix (being denoted as S).
1.2 Average Strategy
If counting the voxel in three dimensions under 13 directions respectively to there is situation, it will obtain 13 matrixes, often
A matrix carries out feature extraction respectively, and the feature that 13 directions obtain is averaged again later, to obtain final characteristic results
(being denoted as 13).If counting the appearance situation of the voxel pair in three dimensions under 13 directions simultaneously, can only build to obtain
One matrix directly extracts feature from the matrix, it is no longer necessary to carry out average (being denoted as 1).Two in comprehensive 1.1 and 1.2
Following 4 kinds of strategies can be obtained in a factor：1S, 1A, 13S and 13A, fixed range is 1 at this time.
1.3 distance
Different distance settings can obtain the texture information under different scale, and therefore, the present invention has chosen D=1 respectively,
Distance setting when 2,3 ... 10 voxels are built as matrix, fixed policy is 1S at this time.
2. the structure of gray scale run-length matrix (GLRLM)
Element P (i, j) in gray scale run-length matrix indicates that on the θ of direction, the j voxel arranged adjacent that gray scale is i occurs
Number.
2.1 Average Strategy
It is similar with the Average Strategy of gray level co-occurrence matrixes, can 13 gray scale run-length matrix be built from 13 directions respectively, then
The feature extracted from this 13 matrixes is carried out average (being denoted as M13)；The voxel arrangement feelings in 13 directions can also be considered simultaneously
Condition obtains a matrix, therefrom extracts feature (being denoted as M1).
3. the structure of gray areas dimension matrix (GLSZM)
The element P (i, j) of gray areas dimension matrix indicates that in three dimensions, j gray scale is the voxel of i in N neighborhoods
Under the conditions of the case where interconnecting the number that occurs.
3.1 neighborhood numbers
In two-dimensional space, in three dimensions, then there are 6 neighborhoods, 18 neighborhoods, 26 neighborhoods in 4 neighborhood of generally use or 8 neighborhoods.
Therefore, the gray areas dimension matrix that the present invention is built considers neighborhood N=6,18,26 these three situations.
4. the structure of neighborhood difference matrix (NGTDM)
Neighborhood difference matrix is a column vector, and element P (i) is indicated in three dimensions, and fixed window width size is W,
All gray scales are the summation of the difference of other all voxel mean values in the center voxel and window of i.
4.1 window width sizes
The selection of different window width sizes can embody difference of the voxel on different scale, therefore, what the present invention was built
Gray areas dimension matrix be arranged window width size be W=3,5,7,9,11.
Texture matrix is built in step (3), specifically includes four kinds of matrixes, respectively gray level co-occurrence matrixes, gray scale distance of swimming square
Battle array, gray areas dimension matrix and neighborhood difference matrix.In gray level co-occurrence matrixes building process comprising symmetry, Average Strategy and
Three parameters of distance include Average Strategy parameter, gray areas dimension matrix building process in gray scale run-length matrix building process
In include neighborhood number parameter, include window width dimensional parameters in neighborhood difference matrix building process.
In order to which the intratumoral metabolism for portraying lesion comprehensively is heterogeneous, the present invention is extracted 26* (4+9)=338 GLCM feature
(4 kinds of strategies, 9 kinds of distances), 13*2=26 GLRLM feature (2 kinds of Average Strategies), 13*3=26 GLSZM features (3 kinds of neighbours
Domain number) and 5*5=25 NGTDM feature (5 kinds of window width sizes), totally 415 features, specific as follows：
GLCM features (26)：Energy, entropy, poor entropy and entropy, variance and the difference of two squares and variance, maximum likelihood, comparison
Degree, non-similarity, homogeney, inverse difference moment, correlation, difference variance, autocorrelation cluster conspicuousness, cluster shade, and cluster becomes
Gesture, information correlation amount 1, information correlation amount 2, inverse variance, standard inverse difference moment, standard unfavourable balance, and average 1, and average 2, one
Cause property.
GLRLM features (13)：Short distance of swimming enhancing, long distance of swimming enhancing, gray scale inhomogeneities, distance of swimming inhomogeneities, the distance of swimming
Percentage, the enhancing of the low gray scale distance of swimming, high gray scale distance of swimming enhancing, the low grey level enhancement of the short distance of swimming, the short high grey level enhancement of the distance of swimming, the long distance of swimming
Low grey level enhancement, the long high grey level enhancement of the distance of swimming, gray variance, distance of swimming variance.
GLSZM features (13)：Zonule enhances, big region enhancing, gray scale inhomogeneities, zone nonuniformity, region
Percentage, low gray level areas enhancing, high gray areas enhancing, the low grey level enhancement in zonule, the high grey level enhancement in zonule, big region
Low grey level enhancement, the big high grey level enhancement in region, gray variance, Local Deviation.
NGTDM features (5)：Roughness, contrast, lengthy and tedious degree, complexity, intensity.
Finally, it into four steps, from texture matrix texture feature extraction, carries out heterogeneity in tumor and quantitatively portrays.
Present invention simultaneously provides method heterogeneous in nasopharynx carcinoma is quantitatively portrayed based on PET image in nasopharyngeal carcinoma judgement side
The purposes in face.
Heterogeneous quantitative depicting method, can effectively realize in nasopharynx carcinoma in the PET image nasopharynx carcinoma of the present invention
Heterogeneous quantifying is portrayed.The present invention solves the effect of technical problem：Symmetry in matrix building process is considered comprehensively, it is average
Strategy, distance, neighborhood number, the influence of the parameter settings such as window width size enrich PET image intratumoral metabolism heterogeneity and quantitatively carve
The adequacy of picture and comprehensive.Realize that accurate nasopharyngeal carcinoma discriminating provides quantitative evaluation index for auxiliary doctor.
Embodiment 2.
Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma, as shown in Fig. 2, including the following steps.
The DICOM format PET image acquired is subjected to cubic spline interpolation first, and itself and CT image registrations are melted
It closes, it is 512*512, voxel size 0.98*0.98*3 to obtain image size.PET activity figures are converted to using following formula
Standard uptake value SUV images：
Then SUV is utilized>2.5 106 cases of threshold segmentation method combination manual adjustment method pair (nasopharyngeal carcinoma 69,
Chronic nasopharyngitis 37) carry out lesion segmentation.
SUV grayscale sliding-model control (discrete interval takes SUV=0.1) is carried out as follows to the lesion divided：
Wherein, SUV (x) indicates that the standard uptake value of voxel x, B indicate that discretization spacing, value are set as 0.1, SUVDis(x)
Indicate the standard uptake value after voxel x discretizations.
Then in the case of structure different parameters setting (symmetry, Average Strategy, distance, neighborhood number, window width size)
Four kinds of texture matrixes.It is specific as follows：
1. the structure of gray level co-occurrence matrixes (GLCM)
The element P (i, j) of gray level co-occurrence matrixes indicates that two gray scales are respectively i and j, and the voxel that distance is D is in direction θ
The number of upper appearance.The structure of the matrix is related to the setting of three symmetry, Average Strategy and distance parameters.
1.1 symmetry
When the situation of the voxel pair on the only directions statistics θ, what is built is unsymmetrical matrix (being denoted as A), when simultaneously
When counting the situation of voxel pair on the negative direction in the directions θ and the directions θ, what is built is symmetrical matrix (being denoted as S).
1.2 Average Strategy
If counting the voxel in three dimensions under 13 directions respectively to there is situation, it will obtain 13 matrixes, often
A matrix carries out feature extraction respectively, and the feature that 13 directions obtain is averaged again later, to obtain final characteristic results
(being denoted as 13).If counting the appearance situation of the voxel pair in three dimensions under 13 directions simultaneously, can only build to obtain
One matrix directly extracts feature from the matrix, it is no longer necessary to carry out average (being denoted as 1).Two in comprehensive 1.1 and 1.2
Following 4 kinds of strategies can be obtained in a factor：1S, 1A, 13S and 13A, fixed range is 1 at this time.
1.3 distance
Different distance settings can obtain the texture information under different scale, and therefore, the present invention has chosen D=1 respectively,
Distance setting when 2,3 ..., 10 voxels are built as matrix, fixed policy is 1S at this time.
2. the structure of gray scale run-length matrix (GLRLM)
Element P (i, j) in gray scale run-length matrix indicates that on the θ of direction, the j voxel arranged adjacent that gray scale is i occurs
Number.
2.1 Average Strategy
It is similar with the Average Strategy of gray level co-occurrence matrixes, can 13 gray scale run-length matrix be built from 13 directions respectively, then
The feature extracted from this 13 matrixes is carried out average (being denoted as M13)；The voxel arrangement feelings in 13 directions can also be considered simultaneously
Condition obtains a matrix, therefrom extracts feature (being denoted as M1).
3. the structure of gray areas dimension matrix (GLSZM)
The element P (i, j) of gray areas dimension matrix indicates that in three dimensions, j gray scale is the voxel of i in N neighborhoods
Under the conditions of the case where interconnecting the number that occurs.
3.1 neighborhood numbers
In two-dimensional space, in three dimensions, then there are 6 neighborhoods, 18 neighborhoods, 26 neighborhoods in 4 neighborhood of generally use or 8 neighborhoods.
Therefore, the gray areas dimension matrix that the present invention is built considers neighborhood N=6,18,26 these three situations.
4. the structure of neighborhood difference matrix (NGTDM)
Neighborhood difference matrix is a column vector, and element P (i) is indicated in three dimensions, and fixed window width size is W,
All gray scales are the summation of the difference of other all voxel mean values in the center voxel and window of i.
4.1 window width sizes
The selection of different window width sizes can embody difference of the voxel on different scale, therefore, what the present invention was built
Gray areas dimension matrix considers window width size W=3,5,7,9,11 this five kinds of situations.
After having built matrix, then 5 traditional SUV features are extracted in the lesion before carrying out sliding-model control：SUV is most
Big value, mean value, peak value (SUVmax, SUVmean, SUVpeak), volume (MATV), total glycolysis (TLG), from the square built
415 textural characteristics are extracted in battle array.Including 26* (4+9)=338 GLCM feature (4 kinds of strategies, 9 kinds of distances), 13*2=26 is a
GLRLM features (2 kinds of Average Strategies), 13*3=26 GLSZM feature (3 kinds of neighborhood numbers) and 5*5=25 NGTDM feature
(5 kinds of window width sizes), it is specific as follows：
GLCM features (26)：Energy, entropy, poor entropy and entropy, variance and the difference of two squares and variance, maximum likelihood, comparison
Degree, non-similarity, homogeney, inverse difference moment, correlation, difference variance, autocorrelation cluster conspicuousness, cluster shade, and cluster becomes
Gesture, information correlation amount 1, information correlation amount 2, inverse variance, standard inverse difference moment, standard unfavourable balance, and average 1, and average 2, one
Cause property.
GLRLM features (13)：Short distance of swimming enhancing, long distance of swimming enhancing, gray scale inhomogeneities, distance of swimming inhomogeneities, the distance of swimming
Percentage, the enhancing of the low gray scale distance of swimming, high gray scale distance of swimming enhancing, the low grey level enhancement of the short distance of swimming, the short high grey level enhancement of the distance of swimming, the long distance of swimming
Low grey level enhancement, the long high grey level enhancement of the distance of swimming, gray variance, distance of swimming variance.
GLSZM features (13)：Zonule enhances, big region enhancing, gray scale inhomogeneities, zone nonuniformity, region
Percentage, low gray level areas enhancing, high gray areas enhancing, the low grey level enhancement in zonule, the high grey level enhancement in zonule, big region
Low grey level enhancement, the big high grey level enhancement in region, gray variance, Local Deviation.
NGTDM features (5)：Roughness, contrast, lengthy and tedious degree, complexity, intensity.
The feature extracted is imported into single factor test logistic one by one and returns grader, using leave one cross validation and
AUC assesses classification results.AUC>0.80 thinks that this feature has preferable carving effect to heterogeneity in nasopharynx carcinoma.
Comparison of 1. textural characteristics of table with tradition SUV features to nasopharyngeal carcinoma and chronic nasopharyngitis's classification AUC
The I-IV phases：Include the nasopharyngeal carcinoma case (N=69) of all I-IV phases
The I-II phases：Only include the nasopharyngeal carcinoma case (N=22) of I-II phases.
Table 1 gives the comparison of the preferably several textural characteristics of classification performance and tradition SUV features, it is seen that textural characteristics pair
The classification performance of I-II phase nasopharyngeal carcinoma is better than tradition SUV features (AUC=0.88-0.91vs0.72-0.88), can effectively realize
Heterogeneous quantifying is portrayed in nasopharynx carcinoma.
Finally it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention rather than is protected to the present invention
The limitation of range, although being explained in detail to the present invention with reference to preferred embodiment, those skilled in the art should manage
Solution, technical scheme of the present invention can be modified or replaced equivalently, without departing from technical solution of the present invention essence and
Range.
Claims (10)
1. heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma, which is characterized in that including：
(1) original image of the DICOM format of target object PET/CT scannings is obtained, and the original image is located in advance
Reason；
(2) Primary lesions of NPC three-dimensional segmentation is carried out on standard uptake value image after treatment；
(3) SUV grayscale sliding-model controls are carried out to the three-dimensional lesion divided；And using many kinds of parameters setting structure texture moments
Battle array；
(4) it from texture matrix texture feature extraction, carries out heterogeneity in tumor and quantitatively portrays.
2. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 1
In, the original image of the DICOM format for obtaining target object PET/CT scannings, and the original image is located in advance
Reason includes：It to PET image into row interpolation, is merged with CT image registrations, and PET activity figures is converted to by standard using formula (1) and are taken the photograph
Value image：
3. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 2
In step (2) is specifically to use SUV>2.5 threshold value carries out initial segmentation, is then adjusted so as to 3D segmentation lesions to the end again.
4. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 3
In：SUV grayscale sliding-model controls in step (3), using formula (2)：
Wherein, SUV (x) indicates that the standard uptake value of voxel x, B indicate discretization spacing, take 0.1, SUVDis(x) voxel x is indicated
Standard uptake value after discretization.
5. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 4
In：Texture matrix is built in step (3), specifically includes four kinds of matrixes, respectively gray level co-occurrence matrixes, gray scale run-length matrix, ash
Spend area size's matrix and neighborhood difference matrix.
6. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 5
In：It was built comprising three symmetry, Average Strategy and distance parameters, gray scale run-length matrix in gray level co-occurrence matrixes building process
Include Average Strategy parameter in journey, includes neighborhood number parameter, neighborhood difference matrix in gray areas dimension matrix building process
Include window width dimensional parameters in building process.
7. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 6
In many kinds of parameters setting that structure texture matrix uses, wherein symmetry include constructing symmetrical and asymmetrical GLCM；It is flat
Equal strategy is that the structure of GLCM or GLRLM includes being built respectively from 13 directions, then takes being averaged for 13 direction characters again,
Further include considering that 13 directions are only built simultaneously to obtain 1 GLCM or GLRLM, without average when extracting feature；Distance setting
Being 1,2,3 ..., 10 pixels build GLCM in turn；Neighborhood number is set as 6,18,26 and then builds GLSZM；Window width size is set
It is set to 3,5,7,9,11 pixels and then builds NGTDM.
8. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 6
It is 415 total in, texture feature extraction, wherein include 338 GLCM features, 26 GLRLM features, 26 GLSZM features with
And 25 NGTDM features.
9. heterogeneous quantitative depicting method, feature exist in a kind of PET image nasopharynx carcinoma according to claim 8
In the textural characteristics of extraction specifically include：
26 GLCM features, including energy, entropy, poor entropy and entropy, variance and the difference of two squares and variance, maximum likelihood, comparison
Degree, non-similarity, homogeney, inverse difference moment, correlation, difference variance, autocorrelation cluster conspicuousness, cluster shade, and cluster becomes
Gesture, information correlation amount 1, information correlation amount 2, inverse variance, standard inverse difference moment, standard unfavourable balance, and average 1, and average 2, one
Cause property；
13 GLRLM features, short distance of swimming enhancing, long distance of swimming enhancing, gray scale inhomogeneities, distance of swimming inhomogeneities, distance of swimming percentage,
The enhancing of the low gray scale distance of swimming, high gray scale distance of swimming enhancing, the low grey level enhancement of the short distance of swimming, the short high grey level enhancement of the distance of swimming, the low gray scale of the long distance of swimming
Enhancing, the long high grey level enhancement of the distance of swimming, gray variance, distance of swimming variance；
13 GLSZM features, zonule enhancing, big region enhancing, gray scale inhomogeneities, zone nonuniformity, area percentage,
Low gray level areas enhances, the enhancing of high gray areas, the low grey level enhancement in zonule, the high grey level enhancement in zonule, the low gray scale in big region
Enhancing, the big high grey level enhancement in region, gray variance, Local Deviation；
5 NGTDM features, roughness, contrast, lengthy and tedious degree, complexity, intensity.
10. quantitatively portraying purposes of the method heterogeneous in nasopharynx carcinoma in terms of nasopharyngeal carcinoma judgement, feature based on PET image
Be, using quantitative depicting method heterogeneous in the PET image nasopharynx carcinoma as described in claim 1 to 9 any one into
Row.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810558897.8A CN108805892A (en) | 2018-06-01 | 2018-06-01 | Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810558897.8A CN108805892A (en) | 2018-06-01 | 2018-06-01 | Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108805892A true CN108805892A (en) | 2018-11-13 |
Family
ID=64090337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810558897.8A Pending CN108805892A (en) | 2018-06-01 | 2018-06-01 | Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108805892A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109035208A (en) * | 2018-06-29 | 2018-12-18 | 上海联影医疗科技有限公司 | Recognition methods, device and the PET system in hypermetabolism region |
| CN111583217A (en) * | 2020-04-30 | 2020-08-25 | 深圳开立生物医疗科技股份有限公司 | Tumor ablation curative effect prediction method, device, equipment and computer medium |
| CN112750528A (en) * | 2019-10-30 | 2021-05-04 | 中国医药大学附设医院 | Computer-aided prediction system, method and computer program product for predicting characteristic parameters of a tumor |
| WO2022126581A1 (en) * | 2020-12-18 | 2022-06-23 | 深圳先进技术研究院 | Pet image reconstruction method and apparatus, and device |
| CN115661150A (en) * | 2022-12-26 | 2023-01-31 | 武汉楚精灵医疗科技有限公司 | Method and device for identifying nasopharyngeal cavity endoscope image abnormality |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101669828A (en) * | 2009-09-24 | 2010-03-17 | 复旦大学 | System for detecting pulmonary malignant tumour and benign protuberance based on PET/CT image texture characteristics |
| CN105653858A (en) * | 2015-12-31 | 2016-06-08 | 中国科学院自动化研究所 | Image omics based lesion tissue auxiliary prognosis system and method |
| CN106355023A (en) * | 2016-08-31 | 2017-01-25 | 北京数字精准医疗科技有限公司 | Open quantitative analysis method and system based on medical image |
| CN106530296A (en) * | 2016-11-07 | 2017-03-22 | 首都医科大学 | Lung detection method and device based on PET/CT image features |
-
2018
- 2018-06-01 CN CN201810558897.8A patent/CN108805892A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101669828A (en) * | 2009-09-24 | 2010-03-17 | 复旦大学 | System for detecting pulmonary malignant tumour and benign protuberance based on PET/CT image texture characteristics |
| CN105653858A (en) * | 2015-12-31 | 2016-06-08 | 中国科学院自动化研究所 | Image omics based lesion tissue auxiliary prognosis system and method |
| CN106355023A (en) * | 2016-08-31 | 2017-01-25 | 北京数字精准医疗科技有限公司 | Open quantitative analysis method and system based on medical image |
| CN106530296A (en) * | 2016-11-07 | 2017-03-22 | 首都医科大学 | Lung detection method and device based on PET/CT image features |
Non-Patent Citations (3)
| Title |
|---|
| LIJUN LU 等: ""Robustness of Radiomic Features in [11C]Choline and [18F]FDG PET/CT imaging of nasopharyngeal carcinoma:impact of segmentation and discretizaion "", 《MOLECULAR IMAGING AND BIOLOGY》 * |
| WENBING LV 等: ""Radiomics analysis of baseline 18F-FDG PET/CT images for improved prognosis in nasopharyngeal carcinoma"", 《2018 IEEE 15TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING (ISBI 2018)》 * |
| WENBING LV 等: ""Robustness versus disease differentiation when varying parameter settings in radiomics features: application to nasopharyngeal PET/CT"", 《EUROPEAN RADIOLOGY》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109035208A (en) * | 2018-06-29 | 2018-12-18 | 上海联影医疗科技有限公司 | Recognition methods, device and the PET system in hypermetabolism region |
| CN112750528A (en) * | 2019-10-30 | 2021-05-04 | 中国医药大学附设医院 | Computer-aided prediction system, method and computer program product for predicting characteristic parameters of a tumor |
| CN111583217A (en) * | 2020-04-30 | 2020-08-25 | 深圳开立生物医疗科技股份有限公司 | Tumor ablation curative effect prediction method, device, equipment and computer medium |
| WO2022126581A1 (en) * | 2020-12-18 | 2022-06-23 | 深圳先进技术研究院 | Pet image reconstruction method and apparatus, and device |
| US12020351B2 (en) | 2020-12-18 | 2024-06-25 | Shenzhen Institutes Of Advanced Technology | Method, device and equipment for reconstructing PET images |
| CN115661150A (en) * | 2022-12-26 | 2023-01-31 | 武汉楚精灵医疗科技有限公司 | Method and device for identifying nasopharyngeal cavity endoscope image abnormality |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805892A (en) | Heterogeneous quantitative depicting method in a kind of PET image nasopharynx carcinoma | |
| CN104380340B (en) | Pixel is scored and adjusted based on neighborhood relationships to disclose the data in image | |
| CN105389811B (en) | A kind of multi-modality medical image processing method split based on multilevel threshold | |
| Chen et al. | Pathological lung segmentation in chest CT images based on improved random walker | |
| CN106355552B (en) | A kind of depth map top sampling method based on virtual viewpoint rendering quality | |
| CN110298804A (en) | One kind is based on generation confrontation network and the decoded medical image denoising method of 3D residual coding | |
| Hasan | A hybrid approach of using particle swarm optimization and volumetric active contour without edge for segmenting brain tumors in MRI scan | |
| CN106991660B (en) | The three dimensional ultrasonic image data methods of sampling decomposed based on modified Octree | |
| Dawood et al. | The importance of contrast enhancement in medical images analysis and diagnosis | |
| Hamad et al. | Segmentation and measurement of lung pathological changes for COVID-19 diagnosis based on computed tomography | |
| CN104517315A (en) | Method and system for reconstructing bilateral ureters based on interactive region growing method | |
| Siddiqi et al. | Investigation of histogram equalization filter for CT scan image enhancement | |
| Talebpour et al. | Automatic lung nodules detection in computed tomography images using nodule filtering and neural networks | |
| Almi'ani et al. | Automatic segmentation algorithm for brain MRA images | |
| Liu et al. | A practical PET/CT data visualization method with dual-threshold PET colorization and image fusion | |
| Shimizu et al. | Extension of automated melanoma screening for non-melanocytic skin lesions | |
| CN109034256A (en) | A kind of the tumor of breast detection system and method for LTP and HOG Fusion Features | |
| Palani et al. | Statistical analysis on impact of image preprocessing of CT texture patterns and its CT radiomic feature stability: a phantom study | |
| Hill et al. | Dynamic breast MRI: image registration and its impact on enhancement curve estimation | |
| Cascio et al. | Automated detection of lung nodules in low-dose computed tomography | |
| CN112634470A (en) | Three-dimensional threshold value stereo graph unfolding method | |
| US20230351557A1 (en) | Method and system for image enhancement | |
| Qiang et al. | Coarse-to-fine lung segmentation in computed tomography images | |
| 大矢めぐみ et al. | Investigation of clinical target volume segmentation for whole breast irradiation using three-dimensional convolutional neural networks with gradient-weighted class activation mapping | |
| Liu et al. | Low-dose liver CT images segmentation and 3D reconstruction |
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 |
Application publication date: 20181113 |
|
| RJ01 | Rejection of invention patent application after publication |