CN108364294A - Abdominal CT images multiple organ dividing method based on super-pixel - Google Patents
Abdominal CT images multiple organ dividing method based on super-pixel Download PDFInfo
- Publication number
- CN108364294A CN108364294A CN201810112755.9A CN201810112755A CN108364294A CN 108364294 A CN108364294 A CN 108364294A CN 201810112755 A CN201810112755 A CN 201810112755A CN 108364294 A CN108364294 A CN 108364294A
- Authority
- CN
- China
- Prior art keywords
- abdominal
- images
- super
- pixel
- pixel block
- 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
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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20028—Bilateral filtering
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a kind of abdominal CT images multiple organ dividing method based on super-pixel, it include the pretreatment to abdominal CT images, wherein pretreatment includes filtering and the grey scale mapping to abdominal CT images, utilize existing correlation between levels in the middle image of abdominal CT images collection, and abdominal CT images are divided by super-pixel block using the K means clustering methods of Euclidean distance and Gray homogeneity as metric range after fusion, after feature extraction being carried out to super-pixel block, classified to super-pixel block using extreme learning machine binding site and gray-scale statistical probabilistic model, sorted super-pixel block is merged again, obtain abdominal CT images multiple organ segmentation result；Method provided by the invention can accurately carry out abdominal CT images the segmentation of multiple organ, and calculation amount is reduced to block of pixels grade from pixel scale, greatly reduces calculation amount, accelerates the speed of abdominal CT images multiple organ segmentation.
Description
Technical field
The present invention relates to a kind of image partition methods, and in particular to a kind of abdominal CT images multiple organ based on super-pixel point
Segmentation method.
Background technology
With the development of medical imaging technology, medical image has indispensable status in medical diagnosis, counts simultaneously
The fast development of calculation machine auxiliary diagnosis (Computer Aided Diagnosis, CAD) technology can utilize at area of computer aided
Reason, analysis medical image, help doctor preferably to make correct judgement to the state of an illness.Medical image segmentation as medical image at
The first stage of reason suffers from important meaning to medical image analysis and visualization, and medical image segmentation is that lesion region carries
It takes, organ-tissue measures and the basis of three-dimensional reconstruction, registration between image-guided surgery, medical image etc. also function to
Irreplaceable role.
Human abdomen's organ complex distribution includes mainly liver, kidney, gall-bladder, spleen, stomach, intestines, inferior caval vein, active
Other important internal organs such as arteries and veins.Computed tomography (Computed Tomography, abbreviation CT) is to abdomen, chest
High-resolution anatomic image can be provided with pelvic cavity, frequently as doctor's first choice detection methods, but due to the limitation of device for image
Property and tissue wriggle, abdominal CT images often show shade, noise, organ overlapping and the non-uniform characteristic of gray scale.In addition, mould
The organization edge of paste is also that organ segmentation brings sizable difficulty.Abdominal multivisceral organ is divided, the position of organ and shape
Shape, higher density similitude, histokinesis and local bulk effect are all the factors for influencing segmentation precision.
For medical image segmentation especially abdominal CT images organ segmentation, due to the low comparison between abdomen organ
Degree, organ lesion presence and individual between organ shape difference so that traditional dividing method has certain limitation
Property.It is carried out in addition, current medical image segmentation is directed to single organ mostly, abdominal multivisceral organ segmentation is also not implemented.
Invention content
The purpose of the present invention is to provide a kind of abdominal CT multiple organ dividing method based on super-pixel, it is existing to solve
The problem of one organ carries out is directed to mostly to abdominal organ segmentation in technology.
In order to realize that above-mentioned task, the present invention use following technical scheme：
A kind of abdominal CT images multiple organ dividing method based on super-pixel, includes the following steps：
Step 1, multiple abdominal CT images are acquired, abdominal CT images collection is obtained；
Step 2, each abdominal CT images concentrated to the abdominal CT images pre-process, after being pre-processed
Abdominal CT images collection, the pretreatment includes image filtering and grey scale mapping；
Step 3, each abdominal CT images that the pretreated abdominal CT images are concentrated are used with the side of cluster
Method carries out super-pixel segmentation, obtains multiple super-pixel block of abdominal CT images collection；
Step 4, the gray feature and textural characteristics of the super-pixel block are extracted, feature complete or collected works are obtained；
Step 5, the gray scale, textural characteristics complete or collected works are selected, obtains feature set；
Step 6, according to the gray feature and texture feature set, the super-pixel block is divided using grader
Class obtains the organ classification of super-pixel block；
Step 7, the super-pixel similar super-pixel block in the block is merged, obtains the abdomen after multiple organ segmentation
CT images.
Further, the step 2, each abdominal CT images concentrated to the abdominal CT images are located in advance
Reason, obtains pretreated abdominal CT images collection, the pretreatment includes image filtering and grey scale mapping, including following step
Suddenly：
Bilateral filtering is carried out to each abdominal CT images that the abdominal CT is concentrated, obtains filtered abdominal CT figure
Image set；
Grey scale mapping is carried out to each abdominal CT images that the filtered abdominal CT images are concentrated, by abdominal CT
The tonal range of image maps to the tonal range of abdomen organ, obtains pretreated abdominal CT images collection.
Further, the step 3, each abdominal CT figure that the pretreated abdominal CT images are concentrated
As carrying out super-pixel segmentation using the method for cluster, multiple super-pixel block of abdominal CT images collection are obtained, are included the following steps：
The each abdominal CT images that the abdominal CT images are concentrated are divided into multiple rectangular blocks；
For first abdominal CT images that the abdominal CT images are concentrated, on each rectangular block after its division
It randomly generates a seed point to be clustered as first cluster centre point, the abdominal CT images is obtained by iteration cluster
Final cluster centre；
For other abdominal CT images that the abdominal CT images are concentrated, the above abdominal CT images it is described most
Whole cluster centre point is clustered as the first cluster centre point of this abdominal CT images, obtains the abdominal CT images collection
Multiple super-pixel block.
Further, in the step 3, the abdominal CT images are concentrated using K-means clustering methods each
It opens abdominal CT images to be clustered, obtains multiple super-pixel block of the abdominal CT images collection.
Further, each abdominal CT images I that the abdomen images are concentrated is carried out using K-means clustering methods
Cluster obtains multiple super-pixel block of the abdominal CT images collection, includes the following steps：
A, centered on the first cluster centre point, using the length of side for L rectangular area as initial clustering neighborhood, will
Euclidean distance is used as metric range D at a distance from after Gray homogeneity fusionsA K-means cluster is carried out, is calculated using formula (1)
Go out metric range Ds：
Wherein, dxyFor two pixel Euclidean distances, dgFor two pixel gray level distances, S is the length of side of rectangular block after segmentation,
μ parameters in order to control, 20≤μ≤100；
B, update cluster centre point C after the completion of a K-means clustert(x, y)=I (x, y),N is the number of pixel in clustering cluster after once clustering, and t is cluster iterations, t >=1；
C, the cluster centre point C after the updatet(x, y) and last cluster centre point Ct-1(x, y) is overlapped, and is obtained
Obtain the final cluster centre point Ct(x, y) completes the super-pixel segmentation to this abdominal CT images；
Above-mentioned steps A-C is repeated, until completing point for last abdominal CT images concentrated to abdominal CT images
It cuts, obtains multiple super-pixel block of the abdominal CT images collection.
Further, the step 4 extracts the gray feature and textural characteristics of the super-pixel block, obtains feature
Complete or collected works specifically include：
Gray feature and texture feature extraction are carried out for multiple super-pixel block of the abdominal CT images obtained in step 3,
The gray feature is mainly intensity histogram feature, and textural characteristics include that Gabor textural characteristics, LBP features and gray scale are total
Raw matrix character, obtains the feature complete or collected works.
Further, the feature complete or collected works are selected using Principal Component Analysis in the step 5, is obtained special
Collection.
Further, in the step 6, using the classification of extreme learning machine binding site and gray-scale statistical probabilistic model
Device classifies to the super-pixel block, obtains the organ classification of super-pixel block.
Further, using the grader of extreme learning machine binding site and gray-scale statistical probabilistic model to the super picture
Plain block is classified, and is obtained the organ classification of super-pixel block, is included the following steps：
A, the feature set obtained in step 5 is input to extreme learning machine to be trained and classify, obtains super-pixel block
Class probability, establish the super-pixel block class probability table based on extreme learning machine, wherein row represent extreme learning machine super-pixel
Block class probability；
B, the gray scale and location information of the organ are counted, established respectively based on gray scale gaussian probability model and is based on position
Gaussian probability model, establish the two-dimentional mixed Gaussian super-pixel block class probability table based on position and gray scale, wherein row represent
Gauss super-pixel block class probability；
C, the super-pixel block class probability table by described based on extreme learning machine with described based on position and gray scale
Corresponding cell is multiplied in two-dimentional mixed Gaussian super-pixel block class probability table, obtains super-pixel block organ class probability table,
Wherein row represent the organ class probability of super-pixel block；
D, in the super-pixel block organ class probability table, the organ classification conduct of maximum probability is chosen in every a line
The organ classification of the super-pixel block of the row obtains the organ classification of super-pixel block.
The present invention has following technical characterstic compared with prior art：
1, grey scale mapping is being added to abdominal CT images pretreatment stage, the gray-scale level of abdominal CT images is mapped into abdomen
The profile and structure letter of each organ of abdomen can preferably be presented the step for relative to traditional method on image for portion
Breath, is conducive to later stage super-pixel segmentation and feature extraction；
2, in the super-pixel segmentation stage, for first abdominal CT images of abdominal CT images collection, initial clustering seed point is adopted
With the method generated at random, since abdominal CT images are continuous between levels there are certain correlation and similitude, because
This in the present invention, using the final cluster centre of upper layer abdominal CT images as the initial cluster center of lower layer's abdominal CT images,
It can avoid being in the presence of encountering organ edge just when initial cluster center selects, can allow the result of super-pixel segmentation
The edge of more identical organ；
3, it when carrying out super-pixel block segmentation to abdominal CT images, uses and is made with Gray homogeneity with the Euclidean distance after fusion
Super-pixel segmentation is carried out for the K-means of the metric range methods clustered, the super-pixel segmentation of abdominal CT images is obtained, avoids
Because between organ edge gray difference it is small bring the problem of being difficult to divide so that the result after segmentation is more accurate；
4, traditional image, semantic segmentation problem is converted to the classification problem of super-pixel block, to by calculation amount from pixel
Grade drops greatly reduce calculation amount as low as block of pixels grade, accelerate the speed of abdominal CT images multiple organ segmentation.
Description of the drawings
Fig. 1 is the abdominal CT images multiple organ dividing method flow chart provided by the invention based on super-pixel；
Fig. 2 is the original abdominal CT images of acquisition of the embodiment of the present invention；
Fig. 3 is the abdominal CT images obtained after the embodiment of the present invention pre-processes original abdominal CT images；
Fig. 4 is the abdominal CT figure that the embodiment of the present invention obtains pretreated abdominal CT images after rectangular block divides
Picture；
Fig. 5 is the abdominal CT figure that the embodiment of the present invention obtains pretreated abdominal CT images after super-pixel segmentation
Picture；
Fig. 6 is the schematic diagram in feature extraction region in the embodiment of the present invention；
Fig. 7 is the liver intensity Gauss model that the embodiment of the present invention is established；
Fig. 8 is that the embodiment of the present invention carries out abdominal CT images the abdominal CT images after multiple organ segmentation.
Specific implementation mode
In compliance with the above technical solution, as shown in Figures 1 to 8, the invention discloses a kind of abdominal CT figure based on super-pixel
As multiple organ dividing method, as shown in Figure 1, including the following steps：
Step 1, multiple abdominal CT images are acquired, abdominal CT images collection is obtained；
CT images are by CT scan equipment utilization X-ray beam to the certain thickness level in human body portion
It is scanned the image collection of acquisition.
In the present embodiment, spiral shell CT machines are arranged by Toshiba Aquilion64 and acquires abdominal CT images totally 161, wherein every
The size of abdominal CT images is 512 × 512, thickness 1mm, as shown in Fig. 2, Fig. 2 is one in the present embodiment midriff CT image sets
The abdominal CT images opened.
Step 2, each abdominal CT images concentrated to the abdominal CT images pre-process, after being pre-processed
Abdominal CT images collection, the pretreatment includes image filtering and grey scale mapping.
Specifically, bilateral filtering is carried out to each abdominal CT images that the abdominal CT is concentrated, obtained filtered
Abdominal CT images collection；Grey scale mapping is carried out to each abdominal CT images that the filtered abdominal CT images are concentrated, it will
The tonal range of abdominal CT images maps to the tonal range of abdomen organ, obtains pretreated abdominal CT images collection.
Grey scale mapping is carried out again to filtered abdominal CT images, the tonal range of original abdominal CT images is mapped to
In [0,255], so that the abdominal CT images after mapping can preferably react the structural information and texture information of abdomen organ.
Specifically, tonal range of the abdomen organ in abdominal CT images is calculated by formula (2)：
Wherein Intercept and slope is the DICOM Tag information of abdominal CT images, can be in the category of abdominal CT images
Property in read, by formula (2) obtain abdomen organ minimal gray threshold value be amin, maximum gray threshold is amax, then substitute into formula (3)
Grey scale mapping is carried out to whole abdominal CT images, so that the tonal range of abdominal CT images is mapped to [0,255], after being pre-processed
Abdominal CT images, each image concentrated to abdominal CT images be filtered and grey scale mapping, obtain pretreated
Abdominal CT images collection.
In the present embodiment, to the abdominal CT images of acquisition, as shown in Fig. 2, filtering parameter is set as：Criterion distance difference σd
=1, gray value standard difference σr=15, bilateral filtering is carried out, filtered abdominal CT images are obtained, to remove in abdominal CT images
Noise information, and the edge details information of organ can be retained.
In the present embodiment, since tissue is different to the absorptivity of X-ray and the image-forming principle of CT images, abdomen
There is each organ specific CT values, the CT value Hu ranges of normal abdominal organ to be generally [- 50,140], read abdomen in CT images
Intercept=0, slope=1 of portion's CT images then obtain minimal gray threshold value a according to formula (2)min, maximum gray threshold
amaxRespectively -50,140, by minimal gray threshold value aminWith maximum gray threshold amaxIt is filtered to this in substitution formula (3)
Abdominal CT images carry out grey scale mapping, obtain pretreated abdominal CT images as shown in Figure 3.
Step 3, each abdominal CT images that the pretreated abdominal CT images are concentrated are used with the side of cluster
Method carries out super-pixel segmentation, obtains multiple super-pixel block of abdominal CT images collection.
It is different from general pattern, there is very strong correlation between abdominal CT images levels, carrying out super-pixel segmentation
When last layer image information to next tomographic image information have certain directive significance, if I0First concentrated for abdominal CT images
Abdominal CT images are opened, then randomly generate an initial seed point in each rectangular tiles as initial cluster center point；If I0
It is not that the final cluster centre of a upper abdominal CT images is then mapped downwards work by first abdominal CT images that abdominal CT is concentrated
For the initial cluster center point of this image.
Optionally, each abdominal CT images that the abdominal CT images are concentrated are divided into multiple rectangular blocks；
The abdominal CT images that size is M*N are divided into K rectangular block, the length of side of each rectangular block isI.e. K values determine the granularity of the super-pixel block of generation.
For first abdominal CT images that the abdominal CT images are concentrated, on each rectangular block after its division
It randomly generates a seed point to be clustered as first cluster centre point, in the final cluster for obtaining the abdominal CT images
The heart；
For other abdominal CT images that the abdominal CT images are concentrated, the above abdominal CT images it is described most
Whole cluster centre point is clustered as the first cluster centre point of this abdominal CT images, obtains the abdominal CT images collection
Multiple super-pixel block.
Therefore, each abdominal CT images concentrated to abdominal CT images all carry out super-pixel segmentation so that next layer of figure
Seem to be corrected on the basis of last layer image, using the correlation before and after abdominal CT images, can not only improve point
The accuracy cut, while calculation amount can also be reduced, improve segmentation efficiency.
Optionally, in the step (3), the abdominal CT images are concentrated using K-means clustering methods each
It opens abdominal CT images to be clustered, obtains multiple super-pixel block of the abdominal CT images collection.
The clustering method can be using K-means clustering methods, spectral clustering etc., to obtain the multiple of abdominal CT images collection
Super-pixel block.
Optionally, each abdominal CT images that the abdomen images are concentrated are gathered using K-means clustering methods
Class obtains multiple super-pixel block of the abdominal CT images collection, includes the following steps：
A, centered on the first cluster centre point, using the length of side for L rectangular area as initial clustering neighborhood, will
Euclidean distance is used as metric range D at a distance from after Gray homogeneity fusionsA K-means cluster is carried out, is calculated using formula (4)
Go out metric range Ds：
Wherein, dxyFor two pixel Euclidean distances, dgFor two pixel gray level distances, S is the length of side of rectangular block after segmentation,
μ parameters in order to control, 20≤μ≤100；
B, update cluster centre point C after the completion of a K-means clustert(x, y)=I (x, y),N is the number of pixel in clustering cluster after once clustering, and t is cluster iterations, t >=1；
C, the cluster centre point C after the updatet(x, y) and last cluster centre point Ct-1(x, y) is overlapped, and is obtained
Obtain the final cluster centre point Ct(x, y) completes the super-pixel segmentation to this abdominal CT images；
Above-mentioned steps A-C is repeated, until completing point for last abdominal CT images concentrated to abdominal CT images
It cuts, obtains multiple super-pixel block of the abdominal CT images collection.
Abdominal CT images are different from general nature image, and with single color channel, gray difference is small between organ adjacent organs
The characteristics of, thus the application calculate cluster neighborhood in pixel to cluster centre distance metric when, not merely use
Euclidean distance, but the Gray homogeneity and Euclidean distance of pixel are considered simultaneously, after Euclidean distance is merged with Gray homogeneity
Distance is used as metric range DsSo that the super-pixel block after segmentation can more be close to the boundary of organ.
In the present embodiment, abdominal CT images are concentrated with 161 abdominal CT images, and the size of original abdominal CT images is
512 × 512, K=800, each rectangular block length of side S=18, the image after being divided using rectangular block to pretreated image is such as
Shown in Fig. 4, for first abdominal CT images of abdominal CT images collection, randomly generated on each rectangular block after its segmentation
One seed point is as cluster centre point, centered on the cluster centre point, length of side L=3S, that is, length of side be 54 rectangle region
Domain calculates metric range D as initial clustering neighborhood, by formula (4)sA K-means cluster, wherein μ=40 are carried out, is calculated
The central point of all pixels is as this cluster centre C in the clustering clustert(x, y), i.e.,xi
For the abscissa of the pixel in clustering cluster, yiFor the ordinate of the pixel in clustering cluster, n is of clustering cluster pixel
Number；Until when this cluster centre point is overlapped with last time cluster centre point, end cluster centre point is obtained, is completed to first abdomen
The super-pixel segmentation of portion's CT images；
For other abdominal CT images of abdominal CT atlas, the end cluster centre point of the above abdominal CT images successively
As the first cluster centre point of this abdominal CT images, i.e., second abdominal CT images concentrated to abdominal CT images surpass
When pixel is divided, in first cluster when using the end cluster centre point of first abdominal CT images as this super-pixel segmentation
Heart point repeats step A-C, until complete the segmentation of the 161st abdominal CT images concentrated to abdominal CT images, obtain described in
Multiple super-pixel block of abdominal CT images, as shown in figure 5, true edge of the segmentation result of super-pixel close to organ.
Step 4, the gray feature and textural characteristics of the super-pixel block are extracted, feature complete or collected works are obtained；
Optionally, gray feature is carried out for multiple super-pixel block of the abdominal CT images obtained in step 3 and texture is special
Sign extraction, the gray feature are intensity histogram feature, textural characteristics specifically include, Gabor textural characteristics, LBP features with
And gray level co-occurrence matrixes feature, the gray feature and textural characteristics are incorporated into a set, the feature is obtained
Complete or collected works.
It is in irregular shape due to the super-pixel block after segmentation, the mode pair in setting feature extraction region may be used
The gray feature and textural characteristics of super-pixel block extract, for example, rectangle is inscribed in the maximum of setting super-pixel block, maximum is inscribed
Circles etc. are extraction region, in the present embodiment, for each super-pixel block after segmentation, take the inscribed rectangle conduct of its maximum
Feature extraction region, as shown in Figure 6.
Extract intensity histogram feature, Gabor textural characteristics, LBP features, gray scale respectively in the feature extraction region
Co-occurrence matrix feature.
Wherein, grey level histogram feature is calculated by formula (5),
In formula (5), rkIt is the gray level of pixel, nkIt is with gray scale rkPixel number, MN is whole abdominal CT images
In total number of pixels, that is, count the percentage that number of pixels under each gray level accounts for total number of pixels in whole abdominal CT images
Than.
In the present embodiment, due in step 2, the tonal range of abdominal CT images being mapped to [0,255], therefore right
In each feature extraction region, tonal range is also [0,255], the grey level histogram feature in extraction feature extraction region
Amount obtains the matrix of one 1 × 256 dimension, and 256 percentage numbers in matrix, which are added, is equal to 1.
The extraction process of Gabor textural characteristics is as follows：
All super-pixel feature extraction region in the block is normalized, can be by feature extraction region specifically
It adjusts to unified size, the feature extraction region after being normalized.
Gabor filter is established, specifically, the direction θ of Gabor function parallel stripes can be set, value range is
0 ° to 360 °, phase offset ψ is set, value range is -180 ° to+180 °, and the aspect ratio γ of installation space is determined
The ellipticity of Gabor function shapes.As γ=1, shape is round.Work as γ<When 1, shape is drawn with parallel stripes direction
It is long.
Feature extraction region after being normalized to super-pixel block using Gabor filter carries out n times convolution, by each secondary volume
Long-pending result is converted into one-dimensional vector, then the transformed one-dimensional vector series connection of this n times is merged and generates a matrix, the matrix
It is then Gabor textural characteristics.
In the present embodiment, the size in feature extraction region is uniformly adjusted to 21 × 21, establishes the Gabor filters in 8 directions
Wave device is 0 respectively,Frequency range is set as [0,4], and use is established
Gabor filter to after adjustment feature extraction region carry out 40 convolution, by convolution results each time be converted into it is one-dimensional to
Amount, then this 40 times transformed one-dimensional vectors are merged to the matrix for generating one 1 × 4400 dimension, i.e. Gabor in the present embodiment
Texture characteristic amount is the matrix of one 1 × 4400 dimension.
LBP feature extractions：Due to LBP features be it is a kind of be used for describing the operator of image local feature, be defined on pixel 3 ×
In 3 neighborhood, using centre of neighbourhood pixel as threshold value, adjacent 8 gray values of pixel and the pixel value of the centre of neighbourhood are compared
Compared with if surrounding pixel is more than center pixel value, the position of the pixel is marked as 1, is otherwise 0, therefore feature is carried
It takes for region, LBP values share 28 kinds of possibility, therefore the LBP extracted is characterized as the matrix of 1 × 256 dimension.
Gray level co-occurrence matrixes characteristic quantity：Gray level co-occurrence matrixes are repeatedly appearance formation figure of the pixel grey scale on spatial position
The texture of picture describes the Joint Distribution with certain two pixel grey scale of spatial relation.It is calculated by gray level co-occurrence matrixes
14 statistics such as entropy, contrast, related coefficient, cluster notable, angular second moment, inverse difference moment can be obtained.
In the present embodiment, in feature extraction region, gray level co-occurrence matrixes characteristic quantity is extracted, chooses 0 °,
45 °, 90 °, 135 °, 180 °, 225 °, 270 °, 315 °, 360 ° of 8 directions, number of greyscale levels 16 calculates contrast, phase relation
Number, entropy and entropy, poor entropy, cluster shade, inverse difference moment this 7 statistics are indicated characteristic block.The square of available 1 × 56 dimension
Battle array, the feature vector as gray level co-occurrence matrixes.
After above four kinds of eigenmatrixes are merged, feature complete or collected works are obtained.
In the present embodiment, that the Gabor characteristic of the grey level histogram feature vector of 1 × 256 dimension, 1 × 4400 dimension is vectorial,
The LBP feature vectors of 1 × 256 dimension, the gray level co-occurrence matrixes feature vector of 1 × 56 dimension merge, and obtain one 1 × 4968 dimension
Feature vector as feature complete or collected works.
Step 5, the feature complete or collected works are selected, obtains feature set.
Due to obtaining the higher feature complete or collected works of a dimension in step 4, complete according to the gray feature and textural characteristics
When set pair super-pixel block is classified, it may appear that calculating speed is slow, the problems such as bringing error into, therefore the method that dimensionality reduction may be used
The gray feature and textural characteristics complete or collected works are selected, the method for dimensionality reduction can be independent component analysis method, principal component
Analytic approach etc..
Optionally, the feature complete or collected works are selected using Principal Component Analysis, obtains feature set.
In the present embodiment, feature selecting and dimensionality reduction are carried out using Principal Component Analysis PCA, finally obtains one 1 × 600
The feature vector of dimension, as feature set.
Step 6, according to the feature set, classified to the super-pixel block using grader, obtain super-pixel
The organ classification of block.
According to feature set, it can be classified to super-pixel block using grader, which can be neural network, branch
Hold vector machine, extreme learning machine etc..
Optionally, in the step 6, using the grader of extreme learning machine binding site and gray-scale statistical probabilistic model
Classify to the super-pixel block, obtains the organ classification of super-pixel block.
Block of pixels after super-pixel segmentation is labeled, it is classified according to organ, using extreme learning machine ELM
Classify to super-pixel block, extreme learning machine parameter is set, grader is trained and is tested.
A, the feature set obtained in step 5 is input to extreme learning machine to be trained and classify, obtains super-pixel block
Class probability, establish the super-pixel block class probability table based on extreme learning machine, wherein row represent extreme learning machine super-pixel
Block class probability；
B, the gray scale and location information of organ are counted, establishes be based on gray scale gaussian probability model and location-based height respectively
This probabilistic model establishes the two-dimentional mixed Gaussian super-pixel block class probability table based on position and gray scale, wherein row represent Gauss
Super-pixel block class probability；
C, the super-pixel block class probability table by described based on extreme learning machine with described based on position and gray scale
Corresponding cell is multiplied in two-dimentional mixed Gaussian super-pixel block class probability table, obtains super-pixel block organ class probability table,
Wherein row represent the organ class probability of super-pixel block；
D, in the super-pixel block organ class probability table, the organ classification conduct of maximum probability is chosen in every a line
The organ classification of the super-pixel block of the row obtains the organ classification of super-pixel block.
In the present embodiment, the super-pixel block after segmentation is divided into 6 classes, is liver, left kidney, right kidney, gall-bladder, pancreas respectively
With other, when classifying to super-pixel block using extreme learning machine ELM, using leaving-one method, i.e., 160 images are chosen each time
Feature set after step 5 screening chooses feature set of 1 image after step 5 screening as test as training set
Collection.Extreme learning machine the results are shown in Table 1 to what super-pixel block was classified, and table 1 has only intercepted partial results and has been used as explanation, specifically, right
For super-pixel block 1, after being classified using extreme learning machine, the probability of liver is 0.5, and the probability of left kidney is 0.04, right
The probability of kidney is 0.04, and the probability of gall-bladder is 0.01, and the probability of pancreas is 0.4, and other probability are 0.01.
Super-pixel block class probability table of the table 1 based on extreme learning machine
| Organ classification | Liver | Left kidney | Right kidney | Gall-bladder | Pancreas | Other |
| 1 probability of extreme learning machine super-pixel block | 0.5 | 0.04 | 0.04 | 0.01 | 0.4 | 0.01 |
| 2 probability of extreme learning machine super-pixel block | 0.2 | 0.05 | 0.35 | 0.1 | 0.25 | 0.05 |
| …… | …… | …… | …… | …… | …… | …… |
| 762 probability of extreme learning machine super-pixel block | 0.05 | 0.35 | 0.5 | 0.04 | 0.02 | 0.04 |
The gaussian probability model based on gray scale for establishing each organ, establishes the location-based gaussian probability of each organ
Model, wherein the gray scale gaussian probability model of liver are as shown in fig. 7, establish the super picture of two-dimentional mixed Gaussian based on position and gray scale
Plain block class probability table is shown in Table 2, and table 2 has only intercepted partial results as explanation, specifically, for super-pixel block 1, uses
After two-dimentional mixed Gauss model based on position and gray scale classifies to super-pixel block, the probability of liver is 0.1, and the probability of left kidney is
0.01, the probability of right kidney is 0.01, and the probability of gall-bladder is 0.04, and the probability of pancreas is 0.8, and other probability are 0.02.
Two-dimentional mixed Gaussian super-pixel block class probability table of the table 2 based on position and gray scale
| Organ classification | Liver | Left kidney | Right kidney | Gall-bladder | Pancreas | Other |
| 1 probability of Gauss super-pixel block | 0.1 | 0.01 | 0.01 | 0.04 | 0.8 | 0.02 |
| 2 probability of Gauss super-pixel block | 0.08 | 0.45 | 0.2 | 0.1 | 0.02 | 0.15 |
| …… | …… | …… | …… | …… | …… | …… |
| 762 probability of Gauss super-pixel block | 0.15 | 0.05 | 0.3 | 0.05 | 0.25 | 0.2 |
Table 1 is multiplied with table 2, i.e., it is the same with matrix multiple, the position corresponding with table 2 of table 1 is multiplied, super-pixel is obtained
Block organ class probability table is shown in Table 3, and table 3 has only intercepted partial results and has been used as explanation.
3 super-pixel block organ class probability table of table
In the super-pixel block organ class probability table, being used as in the organ classification of every a line selection maximum probability should
The organ classification of capable super-pixel block obtains the organ classification of super-pixel block.
As can be seen that the first behavior super-pixel block 1 is the probability of 6 kinds of organ classes' results, wherein super-pixel block in table 3
1 for the probability of liver is 0.05, and the probability of left kidney is 0.0004, and the probability of right kidney is 0.0004, and the probability of gall-bladder is 0.0004,
The probability of pancreas is 0.32, and other probability are 0.0004, wherein the maximum probability of pancreas, therefore, it is considered that super-pixel block 1 is pancreas
Gland；Similarly, for the maximum probability that super-pixel block 2 is right kidney, then it is assumed that super-pixel block 2 is right kidney, for super-pixel block 762
For the maximum probability of right kidney, then it is assumed that super-pixel block 762 is right kidney.
Step 7, the super-pixel similar super-pixel block in the block is merged, obtains the abdomen after multiple organ segmentation
CT images.
Due to the division Jing Guo step 6, the super-pixel block of Different Organs classification has many, by these similar super-pixel
Block merges, so that it may to obtain the abdominal CT images after multiple organ segmentation.
In the present embodiment, super-pixel block 1 is pancreas, and super-pixel block 2 is right kidney, and super-pixel block 3 is liver, super-pixel block
4 be liver, and super-pixel block 5 is left kidney ... ..., and super-pixel block 62 is right kidney, it can be seen that the super-pixel block for liver has：
Super-pixel block 3, super-pixel block 4 etc. have super-pixel block 2, super-pixel block 762 etc. for the super-pixel block of right kidney, therefore by super-pixel
The merging such as block 3, super-pixel block 4 obtain the image of liver, and super-pixel block 2, super-pixel block 762 are merged to the image for obtaining right kidney,
These images are merged, obtain the abdomen images after multiple organ segmentation, as shown in Figure 8.
A kind of abdominal CT images multiple organ dividing method based on super-pixel provided by the invention, pre- to abdominal CT images
Processing stage adds grey scale mapping, and the gray-scale level of abdominal CT images is mapped to abdomen, relative to this step of traditional method
Suddenly the profile and structural information that each organ of abdomen can preferably be presented on image, are conducive to later stage super-pixel segmentation and spy
Sign extraction.
In the super-pixel segmentation stage, for first abdominal CT images of abdominal CT images collection, initial clustering seed point uses
The method generated at random, since abdominal CT images are continuous between levels there are certain correlation and similitude,
In the present invention, using the final cluster centre of upper layer abdominal CT images as the initial cluster center of lower layer's abdominal CT images, energy
Enough avoid being in the presence of encountering organ edge just when initial cluster center selects, can allow super-pixel segmentation result more
Add the edge of identical organ.
To the abdominal CT images after super-pixel segmentation, the feature for extracting each super-pixel block is classified,
The classification results of extreme learning machine are merged with based on position with the Gauss two dimensional model of gray scale when classification, it is final right to obtain
In the classification results of super-pixel block so that classification results are more accurate.Sorted similar super-pixel block is merged, is merged
It can be obtained the abdominal CT images of multiple organ segmentation.
Claims (9)
1. a kind of abdominal CT images multiple organ dividing method based on super-pixel, which is characterized in that include the following steps：
Step 1, multiple abdominal CT images are acquired, abdominal CT images collection is obtained；
Step 2, each abdominal CT images concentrated to the abdominal CT images pre-process, and obtain pretreated abdomen
Portion's CT image sets, the pretreatment include image filtering and grey scale mapping；
Step 3, each abdominal CT images pretreated abdominal CT images concentrated using cluster method into
Row super-pixel segmentation obtains multiple super-pixel block of abdominal CT images collection；
Step 4, the gray feature and textural characteristics of the super-pixel block are extracted, feature complete or collected works are obtained；
Step 5, the feature complete or collected works are selected, obtains feature set；
Step 6, according to the feature set, classified to the super-pixel block using grader, obtain super-pixel block
Organ classification；
Step 7, the super-pixel similar super-pixel block in the block is merged, obtains the abdominal CT figure after multiple organ segmentation
Picture.
2. the abdominal CT images multiple organ dividing method according to claim 1 based on super-pixel, which is characterized in that described
Step 2, each abdominal CT images concentrated to the abdominal CT images pre-process, and obtain pretreated abdominal CT
Image set, the pretreatment include image filtering and grey scale mapping, are included the following steps：
Bilateral filtering is carried out to each abdominal CT images that the abdominal CT is concentrated, obtains filtered abdominal CT images
Collection；
Grey scale mapping is carried out to each abdominal CT images that the filtered abdominal CT images are concentrated, by abdominal CT images
Tonal range map to the tonal range of abdomen organ, obtain pretreated abdominal CT images collection.
3. the abdominal CT images multiple organ dividing method according to claim 2 based on super-pixel, which is characterized in that described
Step 3, each abdominal CT images that the pretreated abdominal CT images are concentrated are carried out using the method for cluster
Super-pixel segmentation obtains multiple super-pixel block of abdominal CT images collection, includes the following steps：
The each abdominal CT images that the abdominal CT images are concentrated are divided into multiple rectangular blocks；
It is random on each rectangular block after its division for first abdominal CT images that the abdominal CT images are concentrated
It generates a seed point to be clustered as first cluster centre point, obtains the final cluster centre of the abdominal CT images；
For other abdominal CT images that the abdominal CT images are concentrated, above the described of an abdominal CT images finally gathers
Class central point is clustered as the first cluster centre point of this abdominal CT images, obtains the more of the abdominal CT images collection
A super-pixel block.
4. the abdominal CT images multiple organ dividing method according to claim 3 based on super-pixel, which is characterized in that described
Step 3 in each abdominal CT images that the abdominal CT images are concentrated are clustered using K-means clustering methods, obtain
Obtain multiple super-pixel block of the abdominal CT images collection.
5. the abdominal CT images multiple organ dividing method according to claim 4 based on super-pixel, which is characterized in that use
K-means clustering methods cluster each abdominal CT images that the abdomen images are concentrated, and obtain the abdominal CT
Multiple super-pixel block of image set, include the following steps：
A, centered on the first cluster centre point, using the length of side for L rectangular area as initial clustering neighborhood, by Euclidean
Distance is used as metric range D at a distance from after Gray homogeneity fusionsA K-means cluster is carried out, out-degree is calculated using formula (1)
Span is from Ds：
Wherein, dxyFor two pixel Euclidean distances, dgFor two pixel gray level distances, S is the length of side of rectangular block after segmentation, and μ is
Control parameter, 20≤μ≤100；
B, update cluster centre point C after the completion of a K-means clustert(x, y)=I (x, y),N is the number of pixel in clustering cluster after once clustering, and t is cluster iterations, t >=1；
C, the cluster centre point C after the updatet(x, y) and last cluster centre point Ct-1(x, y) is overlapped, described in acquisition
Final cluster centre point Ct(x, y) completes the super-pixel segmentation to this abdominal CT images；
Above-mentioned steps A-C is repeated, until completing the segmentation for last abdominal CT images concentrated to abdominal CT images, is obtained
Obtain multiple super-pixel block of the abdominal CT images collection.
6. the abdominal CT images multiple organ dividing method according to claim 1 based on super-pixel, which is characterized in that described
Step 4, extract the gray feature and textural characteristics of the super-pixel block, obtain feature complete or collected works, including：
Gray feature and texture feature extraction are carried out for multiple super-pixel block of the abdominal CT images obtained in step 3, it is described
Gray feature be intensity histogram feature, textural characteristics include that Gabor textural characteristics, LBP features and gray level co-occurrence matrixes are special
Sign obtains feature complete or collected works.
7. the abdominal CT images multiple organ dividing method according to claim 1 based on super-pixel, which is characterized in that described
Step 5 in the feature complete or collected works are selected using Principal Component Analysis, obtain feature set.
8. the abdominal CT images multiple organ dividing method according to claim 1 based on super-pixel, which is characterized in that described
Step 6 in, using the grader of extreme learning machine binding site and gray-scale statistical probabilistic model to the super-pixel block into
Row classification, obtains the organ classification of super-pixel block.
9. the abdominal CT images multiple organ dividing method according to claim 8 based on super-pixel, which is characterized in that use
The grader of extreme learning machine binding site and gray-scale statistical probabilistic model classifies to the super-pixel block, obtains super picture
The organ classification of plain block, includes the following steps：
A, the texture feature set obtained in step 5 is input to extreme learning machine to be trained and classify, obtains super-pixel block
Class probability, establish the super-pixel block class probability table based on extreme learning machine, wherein row represent extreme learning machine super-pixel
Block class probability；
B, the gray scale and location information of the organ are counted, establishes be based on gray scale gaussian probability model and location-based height respectively
This probabilistic model establishes the two-dimentional mixed Gaussian super-pixel block class probability table based on position and gray scale, wherein row represent Gauss
Super-pixel block class probability；
C, the super-pixel block class probability table by described based on extreme learning machine and the two dimension based on position and gray scale
Corresponding cell is multiplied in mixed Gaussian super-pixel block class probability table, obtains super-pixel block organ class probability table, wherein
Row represent the organ class probability of super-pixel block；
D, in the super-pixel block organ class probability table, the organ classification of maximum probability is chosen as the row in every a line
Super-pixel block organ classification, obtain the organ classification of the super-pixel block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810112755.9A CN108364294B (en) | 2018-02-05 | 2018-02-05 | Multi-organ segmentation method for abdominal CT image based on superpixels |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810112755.9A CN108364294B (en) | 2018-02-05 | 2018-02-05 | Multi-organ segmentation method for abdominal CT image based on superpixels |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108364294A true CN108364294A (en) | 2018-08-03 |
| CN108364294B CN108364294B (en) | 2021-01-12 |
Family
ID=63004697
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810112755.9A Active CN108364294B (en) | 2018-02-05 | 2018-02-05 | Multi-organ segmentation method for abdominal CT image based on superpixels |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108364294B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109671054A (en) * | 2018-11-26 | 2019-04-23 | 西北工业大学 | The non-formaldehyde finishing method of multi-modal brain tumor MRI |
| CN109685767A (en) * | 2018-11-26 | 2019-04-26 | 西北工业大学 | A kind of bimodal brain tumor MRI dividing method based on Cluster-Fusion algorithm |
| CN110084821A (en) * | 2019-04-17 | 2019-08-02 | 杭州晓图科技有限公司 | A kind of more example interactive image segmentation methods |
| CN110163870A (en) * | 2019-04-24 | 2019-08-23 | 艾瑞迈迪科技石家庄有限公司 | A kind of abdomen body image liver segmentation method and device based on deep learning |
| CN110188788A (en) * | 2019-04-15 | 2019-08-30 | 浙江工业大学 | The classification method of cystic Tumor of Pancreas CT image based on radiation group feature |
| CN110570352A (en) * | 2019-08-26 | 2019-12-13 | 腾讯科技（深圳）有限公司 | image labeling method, device and system and cell labeling method |
| CN110751101A (en) * | 2019-10-22 | 2020-02-04 | 吉林大学 | Fatigue driving judgment method based on multiple clustering algorithm of unsupervised extreme learning machine |
| CN110766693A (en) * | 2018-09-06 | 2020-02-07 | 北京连心医疗科技有限公司 | Method for jointly predicting radiotherapy structure position based on multi-model neural network |
| CN111429460A (en) * | 2020-06-12 | 2020-07-17 | 腾讯科技（深圳）有限公司 | Image segmentation method, image segmentation model training method, device and storage medium |
| CN112598634A (en) * | 2020-12-18 | 2021-04-02 | 燕山大学 | CT image organ positioning method based on 3D CNN and iterative search |
| CN112686898A (en) * | 2021-03-15 | 2021-04-20 | 四川大学 | Automatic radiotherapy target area segmentation method based on self-supervision learning |
| CN112785608A (en) * | 2021-02-09 | 2021-05-11 | 哈尔滨理工大学 | Medical image segmentation method for improving SNIC (single noise integrated circuit) based on adaptive parameters |
| CN115019045A (en) * | 2022-06-24 | 2022-09-06 | 哈尔滨工业大学 | Small data thyroid ultrasound image segmentation method based on multi-component neighborhood |
| CN116797598A (en) * | 2023-08-22 | 2023-09-22 | 山东万牧农业科技有限公司郯城分公司 | Image feature-based cultivation feed quality refinement detection method |
| CN116955681A (en) * | 2023-09-08 | 2023-10-27 | 北京触幻科技有限公司 | Three-dimensional visual medical imaging system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778452A (en) * | 2015-04-02 | 2015-07-15 | 浙江大学 | Feasible region detecting method based on machine learning |
| US20160171707A1 (en) * | 2014-12-10 | 2016-06-16 | Ricoh Co., Ltd. | Realogram Scene Analysis of Images: Superpixel Scene Analysis |
-
2018
- 2018-02-05 CN CN201810112755.9A patent/CN108364294B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160171707A1 (en) * | 2014-12-10 | 2016-06-16 | Ricoh Co., Ltd. | Realogram Scene Analysis of Images: Superpixel Scene Analysis |
| CN104778452A (en) * | 2015-04-02 | 2015-07-15 | 浙江大学 | Feasible region detecting method based on machine learning |
Non-Patent Citations (4)
| Title |
|---|
| 张海涛 等: "基于超像素方法的腹部CT影像多目标器官分割研究", 《中国医疗设备》 * |
| 楚陪陪 等: "基于复合超像素技术的肺部CT图像分割算法", 《合肥工业大学学报（自然科学版）》 * |
| 欧垚江: "基于高斯混合模型的图像分割的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 黄煜峰 等: "基于SMDC连接代价算子的肺血管分割算法研究", 《北京生物医学工程》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766693B (en) * | 2018-09-06 | 2022-06-21 | 北京连心医疗科技有限公司 | Method for jointly predicting radiotherapy structure position based on multi-model neural network |
| CN110766693A (en) * | 2018-09-06 | 2020-02-07 | 北京连心医疗科技有限公司 | Method for jointly predicting radiotherapy structure position based on multi-model neural network |
| CN109685767A (en) * | 2018-11-26 | 2019-04-26 | 西北工业大学 | A kind of bimodal brain tumor MRI dividing method based on Cluster-Fusion algorithm |
| CN109671054A (en) * | 2018-11-26 | 2019-04-23 | 西北工业大学 | The non-formaldehyde finishing method of multi-modal brain tumor MRI |
| CN110188788A (en) * | 2019-04-15 | 2019-08-30 | 浙江工业大学 | The classification method of cystic Tumor of Pancreas CT image based on radiation group feature |
| CN110084821B (en) * | 2019-04-17 | 2021-01-12 | 杭州晓图科技有限公司 | Multi-instance interactive image segmentation method |
| CN110084821A (en) * | 2019-04-17 | 2019-08-02 | 杭州晓图科技有限公司 | A kind of more example interactive image segmentation methods |
| CN110163870A (en) * | 2019-04-24 | 2019-08-23 | 艾瑞迈迪科技石家庄有限公司 | A kind of abdomen body image liver segmentation method and device based on deep learning |
| CN110570352A (en) * | 2019-08-26 | 2019-12-13 | 腾讯科技（深圳）有限公司 | image labeling method, device and system and cell labeling method |
| CN110751101A (en) * | 2019-10-22 | 2020-02-04 | 吉林大学 | Fatigue driving judgment method based on multiple clustering algorithm of unsupervised extreme learning machine |
| CN110751101B (en) * | 2019-10-22 | 2022-05-17 | 吉林大学 | Fatigue driving judgment method based on multiple clustering algorithm of unsupervised extreme learning machine |
| CN111429460A (en) * | 2020-06-12 | 2020-07-17 | 腾讯科技（深圳）有限公司 | Image segmentation method, image segmentation model training method, device and storage medium |
| CN112598634A (en) * | 2020-12-18 | 2021-04-02 | 燕山大学 | CT image organ positioning method based on 3D CNN and iterative search |
| CN112598634B (en) * | 2020-12-18 | 2022-11-25 | 燕山大学 | CT image organ positioning method based on 3D CNN and iterative search |
| CN112785608A (en) * | 2021-02-09 | 2021-05-11 | 哈尔滨理工大学 | Medical image segmentation method for improving SNIC (single noise integrated circuit) based on adaptive parameters |
| CN112785608B (en) * | 2021-02-09 | 2022-06-21 | 哈尔滨理工大学 | Medical image segmentation method for improving SNIC (single noise integrated circuit) based on adaptive parameters |
| CN112686898A (en) * | 2021-03-15 | 2021-04-20 | 四川大学 | Automatic radiotherapy target area segmentation method based on self-supervision learning |
| CN115019045A (en) * | 2022-06-24 | 2022-09-06 | 哈尔滨工业大学 | Small data thyroid ultrasound image segmentation method based on multi-component neighborhood |
| CN116797598A (en) * | 2023-08-22 | 2023-09-22 | 山东万牧农业科技有限公司郯城分公司 | Image feature-based cultivation feed quality refinement detection method |
| CN116797598B (en) * | 2023-08-22 | 2023-11-17 | 山东万牧农业科技有限公司郯城分公司 | Image feature-based cultivation feed quality refinement detection method |
| CN116955681A (en) * | 2023-09-08 | 2023-10-27 | 北京触幻科技有限公司 | Three-dimensional visual medical imaging system |
| CN116955681B (en) * | 2023-09-08 | 2024-04-26 | 北京触幻科技有限公司 | Three-dimensional visual medical imaging system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108364294B (en) | 2021-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108364294A (en) | Abdominal CT images multiple organ dividing method based on super-pixel | |
| CN108053417B (en) | lung segmentation device of 3D U-Net network based on mixed rough segmentation characteristics | |
| CN105957066B (en) | CT image liver segmentation method and system based on automatic context model | |
| CN105894517B (en) | CT image liver segmentation method and system based on feature learning | |
| EP0965104B1 (en) | Autosegmentation/autocontouring methods for use with three-dimensional radiation therapy treatment planning | |
| DE69517524T3 (en) | AUTOMATIC DETECTION OF LESIONS IN COMPUTER TOMOGRAPHY | |
| EP2027566B1 (en) | Automatic recognition of preneoplastic anomalies in anatomic structures based on an improved region-growing segmentation, and computer program therefor | |
| CN101576997B (en) | Abdominal organ segmentation method based on secondary three-dimensional region growth | |
| CN102324109B (en) | Method for three-dimensionally segmenting insubstantial pulmonary nodule based on fuzzy membership model | |
| CN106875409B (en) | A kind of light-type incisional hernia sticking patch three-dimensional ultrasound pattern feature extracting method | |
| CN107977952A (en) | Medical image cutting method and device | |
| CN106778005A (en) | Prostate cancer computer aided detection method and system based on multi-parameter MRI | |
| DE112007002535T5 (en) | Computer diagnosis of malignancies and false positives | |
| US20230005140A1 (en) | Automated detection of tumors based on image processing | |
| CN109753997A (en) | A kind of liver neoplasm automatic and accurate Robust Segmentation method in CT image | |
| Shrivastava et al. | Automatic seeded region growing image segmentation for medical image segmentation: a brief review | |
| CN112820399A (en) | Method and device for automatically diagnosing benign and malignant thyroid nodules | |
| Hassanien et al. | Digital mammogram segmentation algorithm using pulse coupled neural networks | |
| Xu et al. | A pilot study to utilize a deep convolutional network to segment lungs with complex opacities | |
| CN113192036A (en) | Multi-plane breast tumor segmentation method based on three-dimensional MRI (magnetic resonance imaging) image | |
| Öttl et al. | Superpixel pre-segmentation of her2 slides for efficient annotation | |
| Lu et al. | 3d tomosynthesis to detect breast cancer | |
| Yadav et al. | Texture Segmentation and Features of Medical Images | |
| Reddy et al. | Tumor Region detection in MRI Images using Novel Kernalized Fuzzy C-Means Clustering Algorithm | |
| Nedevschi et al. | Kidney CT image segmentation using multi-feature EM algorithm, based on Gabor filters |
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 |