CN110874843B - Organ image segmentation method and device - Google Patents
Organ image segmentation method and device Download PDFInfo
- Publication number
- CN110874843B CN110874843B CN201911029097.8A CN201911029097A CN110874843B CN 110874843 B CN110874843 B CN 110874843B CN 201911029097 A CN201911029097 A CN 201911029097A CN 110874843 B CN110874843 B CN 110874843B
- Authority
- CN
- China
- Prior art keywords
- image
- distribution
- sampling
- target organ
- segmented
- 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.)
- Active
Links
Images
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
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a method and a device for segmenting an organ image, comprising the following steps: nonlinear registration is carried out to obtain registration coefficients; after the binary image is transformed, overlapping the binary image to obtain prior distribution; space sampling is carried out based on Markov chain Monte Carlo to obtain sampling points; obtaining corresponding subareas from the image to be segmented; obtaining and superposing sub-region segmentation results to obtain likelihood functions; setting and superposing sub-region oscillography functions to obtain prior distribution; and combining the prior distribution and the likelihood function to obtain posterior distribution, and binarizing the posterior distribution to obtain a final segmentation result. According to the embodiment of the invention, the prior distribution of the target organ is obtained through the registration of the sample image, the sampling is carried out through a Markov chain Monte Carlo algorithm, the space sampling point is obtained, the likelihood function of the target organ is obtained according to the segmentation result of the corresponding subarea, the indication function of the subarea is constructed and overlapped, the prior distribution is obtained, the posterior distribution is obtained through the likelihood function and the prior distribution, and the final segmentation image is obtained after the binarization processing.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a method and a device for segmenting an organ image.
Background
In the current technology, the segmentation of many organs has achieved a good effect, but for some organs with large individual difference, small size, insignificant boundaries and easy deformation, the segmentation results are improved to a certain extent, but not to a satisfactory extent, for example: pancreas. For such organ segmentation, more representative ideas are: the rough-to-fine segmentation is performed, namely rough segmentation is performed to obtain the approximate position of the organ, so that a bounding box of the organ is obtained, and the organ is segmented in the bounding box by using a deep neural network. However, for organs of small size, blurred boundaries, and irregular morphology like the pancreas, there are two problems with roughly segmented bounding boxes:
1) The bounding box does not cover the target organ well;
2) Too many background areas are in the bounding box, so that the training data set is unbalanced, and the deep neural network cannot learn the interior of the target organ and the outline characteristics of the target organ well.
Disclosure of Invention
Embodiments of the present invention aim to solve at least one of the technical problems in the related art to some extent. To this end, an object of an embodiment of the present invention is to provide an organ image segmentation method.
To this end, a second object of an embodiment of the present invention is to provide an organ image segmentation apparatus.
The technical scheme adopted by the invention is as follows:
in a first aspect, an embodiment of the present invention provides an organ image segmentation method, including: carrying out nonlinear registration on a plurality of sample images and images to be segmented to obtain registration coefficients of the sample images; transforming binary images of the target organ in the sample image based on the registration coefficient, and superposing the binary images to obtain prior distribution of the target organ; based on Markov chain Monte Carlo, spatial sampling is carried out by utilizing prior distribution of a target organ, and a precise sampling point is obtained; extracting a sub-region from the image to be segmented according to the precise sampling points; processing the subareas and outputting segmentation results of the subareas; overlapping the sub-region segmentation results in the image space to be segmented to obtain a likelihood function of a target organ in the image to be segmented; setting the readiness function of the subarea in the image space to be segmented, and superposing the readiness function to obtain prior distribution; and combining the prior distribution and likelihood function of the target organ in the image to be segmented to obtain posterior distribution of the target organ, and binarizing the posterior distribution to obtain a final segmentation result.
Preferably, the method comprises the steps of processing the subareas based on a learning model which is completed by training, outputting segmentation results of the subareas, and setting a corresponding learning model, wherein the steps comprise: performing spatial sampling by using prior distribution of a target organ to obtain sampling points; extracting corresponding sub-regions from the registration images of the sample images according to the sampling points; and training by taking the subareas as training samples of the learning model.
Preferably, the learning model is based on a convolutional neural network or a support vector machine or random forest.
Preferably, based on markov chain monte carlo, spatially sampling with a priori distribution of the target organ comprises: setting a stable distribution P (x) and a symmetrical state transition matrix q; setting the threshold value of the state transition times as n 1 The number of samples required is n 2 The method comprises the steps of carrying out a first treatment on the surface of the Optionally generating an initial state x 0 The method comprises the steps of carrying out a first treatment on the surface of the From a conditional probability distribution q (x|x t ) Mid-sampling to obtain sample x * ，0≤t≤n 1 +n 2 -1; sampling from Uniform distribution form (0, 1) to obtain a random number u; acceptance rate alpha (x) t ,x * )＝min{p(x * )/p(x t ) 1, when u < alpha (x) t ,x * ) Let x t+1 ＝x * Extracting x t+1 And obtaining a space sampling result by the corresponding three-dimensional subareas.
Preferably, the likelihood function comprises a sum C of classification results of the pixel points i The prior distribution includes the number M of times the pixel is classified i The posterior distribution includes the probability V of the pixel point as the target organ t ＝C i /M i 。
In a second aspect, an embodiment of the present invention provides an organ image segmentation apparatus, including: the registration module is used for carrying out nonlinear registration on the sample image and the image to be segmented to obtain a registration coefficient of the sample image; the sample processing module is used for transforming the binary images of the target organ in the sample image based on the registration coefficient, and superposing the binary images to obtain prior distribution of the target organ; the prior sampling module is used for performing spatial sampling by utilizing prior distribution of a target organ based on a Markov chain Monte Carlo method to obtain a precise sampling point; the prior region module is used for extracting a sub-region from the image to be segmented according to the precise sampling points; the priori segmentation module is used for processing the subareas and outputting segmentation results of the subareas; the likelihood module is used for overlaying the segmentation result of the subarea in the air of the image to be segmented to obtain a likelihood function of a target organ in the image to be segmented; the oscillography module is used for setting corresponding oscillography functions for the subareas, superposing the oscillography functions and obtaining prior distribution; and the posterior module is used for combining the prior distribution and likelihood function of the target organ in the image to be segmented to obtain posterior distribution of the target organ, and binarizing the posterior distribution to obtain a final segmentation result.
Preferably, the prior segmentation module is configured to process the sub-region based on a learning model that is completed by training, output a segmentation result of the sub-region, and the setting of the learning model includes: based on the prior distribution model, performing spatial sampling to obtain sampling points; extracting corresponding sub-regions from the registration images of the sample images according to the sampling points; and training by taking the subareas as training samples of the learning model.
Preferably, the learning model is based on a convolutional neural network or a support vector machine or random forest.
Preferably, based on markov chain monte carlo, spatially sampling with a priori distribution of the target organ comprises: setting a stable distribution P (x) and a symmetrical state transition matrix q; setting the threshold value of the state transition times as n 1 The number of samples required is n 2 The method comprises the steps of carrying out a first treatment on the surface of the Optionally generating an initial state x 0 The method comprises the steps of carrying out a first treatment on the surface of the From a conditional probability distribution q (x|x t ) Mid-sampling to obtain sample x * ，0≤t≤n 1 +n 2 -1; sampling from Uniform distribution form (0, 1) to obtain a random number u; acceptance rate alpha (x) t ,x * )＝min{p(x * )/p(x t ) 1, when u < alpha (x) t ,x * ) Let x t+1 ＝x * Extracting x t+1 And obtaining a space sampling result by the corresponding three-dimensional subareas.
Preferably, the likelihood function comprises a sum C of classification results of the pixel points i The prior distribution includes the number M of times the pixel is classified i The posterior distribution includes the probability V of the pixel point as the target organ t ＝C i /M i 。
The embodiment of the invention has the beneficial effects that:
according to the embodiment of the invention, the prior distribution of a possible target organ is acquired through the registration of sample images, sampling is carried out through a Markov chain Monte Carlo algorithm, the sampling points in space can be reasonably obtained, the sub-regions can be extracted through the sampling points, the likelihood function of the target organ in the image to be segmented is obtained through superposition of the segmentation result of the sub-regions in the image space to be segmented, the prior distribution is obtained through superposition of the indication function of the sub-regions in the image space to be segmented, the posterior distribution can be obtained according to the likelihood function and the prior distribution, and the final segmented image can be obtained through binarization processing.
Drawings
FIG. 1 is a flow chart of one embodiment of a method of organ image segmentation;
FIG. 2 is a segmentation effect diagram of an organ image segmentation method;
FIG. 3 is a connection diagram of one embodiment of an organ image segmentation apparatus.
Detailed Description
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other.
The technical names involved are as follows:
markov Chains (MC) are random processes (stochastic process) in probabilistic and statistical statistics that have Markov properties and exist within discrete index sets (index sets) and state space (state space). Markov chains suitable for Continuous index sets are known as Markov processes (Markov processes), but are sometimes also regarded as a subset of Markov chains, i.e. Continuous-Time chains (CTMC), corresponding to Discrete-Time Markov chains (DTMC), and thus Markov chains are a broader concept.
Markov chains may be defined by transition matrices and transition diagrams, and in addition to Markov, markov chains may have irreducibility, reproducibility, periodicity, and traversability. An irreducible and recurring Markov chain is a strictly smooth Markov chain with a unique smooth distribution. The limit distribution of the traverse markov chain (ergodic MC) converges to its plateau distribution.
Example 1.
The embodiment of the invention provides an organ image segmentation method as shown in fig. 1, which comprises the following steps:
s1, carrying out nonlinear registration on a plurality of sample images and an image to be detected to obtain registration coefficients of the sample images;
s2, transforming binary images of the target organ in the sample image based on the registration coefficient, and superposing the binary images to obtain prior distribution of the target organ;
s3, based on Markov chain Monte Carlo, performing spatial sampling by using prior distribution of a target organ to obtain a precise sampling point;
s4, extracting a sub-region from the image to be segmented according to the precise sampling points;
s5, processing the subareas and outputting segmentation results of the subareas;
s6, superposing the segmentation results of the subareas in the image space to be segmented to obtain a likelihood function of a target organ in the image to be segmented;
s7, setting an indication function corresponding to the image space to be segmented for the subareas, and superposing the indication function to obtain prior distribution;
s8, combining prior distribution and likelihood function of the target organ in the graph to be segmented to obtain posterior distribution of the target organ, and binarizing the posterior distribution to obtain a final segmentation result.
The processing thought of the scheme comprises the following steps: firstly, a rough segmentation result (i.e. prior distribution) of a target organ is obtained by a multi-atlas segmentation method, a foreground region can basically cover the target organ region in an image to be segmented, the foreground region is equivalent to an irregular bounding box, MCMC (Markov chain Monte Carlo) sampling is used for guiding CNN (Convolutional Neural Networks, convolutional neural network) to predict the irregular bounding box, and finally, the prediction result is optimized, so that the organ segmentation effect is achieved.
The specific processing flow comprises the following steps: the image to be segmented is marked as I, in order to segment a certain target organ in the image I. N images (i.e. sample images) of the same organ which have been segmented are noted as I i I=1,.. wherein the segmented binary image of the target organ is denoted J i 。
Image I i Nonlinear registration is carried out on the image I to be segmented, and a space transformation function of registration transformation is obtained as followsT i I.e. the registration factor, step S1. With each T i Pair J i Transforming/mapping to obtain transformed binary images, and superposing the transformed binary images, which can be summarized into a composite map J p ＝∑J i (T i (I i ))，J p Is a prior distribution model of the target organ, step S2, the specific registration takes the existing registration tools, e.g. ANTs (Advanced Normalized Tools), ITK,3D Slicer,deedsBCV.
According to J p The expressed probability distribution acquires a series of spatial sampling points, and extracts a sub-region of the image I (i.e., a pixel point region with the sampling point as a base point) at the sampling points. Because of J p Not a special probability density function, such as a uniform distribution, a normal distribution, etc., so sampling from it requires a more complex markov chain monte carlo algorithm.
The Markov chain Monte Carlo algorithm comprises the following steps:
setting a symmetrical state transition matrix q (for example, a value at a position (i, J) in one state transition matrix is a conditional probability q (j|i), namely, a probability of transition from the state i to the state J.) and a stable distribution p (x) (the stable distribution is J) p ) The method comprises the steps of carrying out a first treatment on the surface of the Setting the threshold value of the state transition times as n 1 The number of samples required is n 2 . Starting circulation, starting circulation from t=0, wherein t is more than or equal to 0 and less than or equal to n 1 +n 2 -1; optionally generating an initial state x 0 The method comprises the steps of carrying out a first treatment on the surface of the From a conditional probability distribution q (x|x t ) Mid-sampling to obtain sample x * The method comprises the steps of carrying out a first treatment on the surface of the Sampling from Uniform distribution form (0, 1) to obtain a random number u; calculating the acceptance rate alpha (x) t ,x * )＝min{p(x * )/p(x t ) 1, where when u < alpha (x) t ,x * ) Let x t+1 ＝x * The method comprises the steps of carrying out a first treatment on the surface of the No make x t+1 ＝x t The method comprises the steps of carrying out a first treatment on the surface of the Extraction of x t+1 The corresponding three-dimensional subarea is extracted by the following steps: the method can be used for central extraction (taking a sampling point as a center to extract a cube region), or left rear lower part (taking the sampling point as a left rear lower corner of a subarea), left front upper part, left rear upper part, left front lower part, right rear lower part, right front upper part, and the like of the sampling point,Three-dimensional subregions are extracted at these positions, lower front right and upper rear right. After the circulation is finished, n of the images to be segmented are obtained 2 The sub-areas, steps S3 and S4.
The sub-region is processed, the segmentation result of the sub-region is output, the sub-region is extracted for segmentation, and the purpose of the sub-region segmentation is to independently separate small clusters of pixels at positions corresponding to sampling points and conduct pixel-level fine classification on an image to be segmented. I.e. step S5.
Some sampling points are repeated or are closely spaced, so that sub-regions are overlapped, and therefore some pixel points are classified for multiple times, and the segmentation results of the sub-regions are overlapped in the image space to be segmented to obtain likelihood functions, wherein the likelihood functions represent the sum of the classification results (the classification results are 0 and 1,0 represents non-target organ pixels, and 1 represents target organ pixels) of each pixel point in the classification operation and can be marked as C i . The larger the sum of the classification results is, the larger the probability that the pixel belongs to the target organ is. I.e. step S6.
And constructing the readiness function of each sub-region in the image to be segmented (the sub-region comprising the sampling point is marked as 1 in detail, the other regions in the image are marked as 0), and superposing the readiness functions corresponding to all the sampling points to obtain the prior distribution. I.e. step S7.
Obtaining posterior probability distribution of a target region in the image to be segmented according to the likelihood function and the prior distribution: the obtained prior distribution is the classified times of each pixel point and can be recorded as M i The method comprises the steps of carrying out a first treatment on the surface of the Likelihood can be understood as the sum of the classification results for each pixel, and can be denoted as C i The method comprises the steps of carrying out a first treatment on the surface of the The posterior distribution is the probability that each pixel is the target organ, which can be expressed as: v (V) i ＝C i /M i . And selecting a proper threshold value, and binarizing the posterior probability distribution to obtain the final segmentation. I.e. step S8.
In step S5, the pixels in the sub-region need to be classified by the learning model. Wherein the learning model is based on convolutional neural networks (e.g., resNet, VGG, and 3D U-Net) or support vector machines or random forests. The specific training process comprises the following steps: in the registered image of the sample image, performing spatial sampling according to prior distribution to obtain sampling points; extracting corresponding sub-regions from the registered images of the sample images according to the sampling points; these subregions are used as training samples for the learning model to perform training. Training a learning model of image processing through a sample image of the divided target organ; compared with manual identification, the learning model has stronger processing capacity and is beneficial to improving the efficiency of image identification and marking.
According to the embodiment of the invention, the position of the target organ can be obtained according to the whole organ structure of the human body in the registration process, and the multi-map registration is utilized, so that the accidental caused by the difference of the internal structures of the individuals is eliminated, and the depth neural network can concentrate on learning the internal and edge characteristics of the target organ due to the determination of the irregular bounding box.
The segmentation effect diagram of the organ image segmentation method shown in fig. 2 includes:
the image to be measured is an image of pancreas and surrounding organs;
a priori distribution diagram is a specific binary image superposition image;
a priori distribution map is an image of a target organ obtained by combining a priori distribution diagram with an image to be detected;
a precise sampling point schematic diagram is an image of a sampling point obtained by performing prior distributed spatial sampling based on Markov chain Monte Carlo;
the precise sampling area image is an image of a target organ obtained by combining the image to be detected based on the precise sampling points;
the regional image of the indication function is a binary distribution image corresponding to the indication function;
the regional image of the likelihood function is an image of a target organ obtained by combining the likelihood function with the image to be detected;
and the final segmentation area of the prior distribution is used for combining the prior distribution with the image to be detected to obtain the image of the target organ.
Example 2.
An embodiment of the present invention provides an organ image segmentation apparatus as shown in fig. 3, including: the registration module 1 is used for carrying out nonlinear registration on a plurality of sample images and the images to be segmented to obtain registration coefficients of the sample images; the sample processing module 2 transforms the binary images of the target organ in the sample image based on the registration coefficient, and superimposes the transformed binary images to obtain prior distribution of the target organ; the prior sampling module 3 is used for carrying out prior distributed spatial sampling on the target organ based on Markov chain Monte Carlo to obtain precise sampling points; the priori region module 4 is used for extracting a sub-region from the image to be segmented according to the precise sampling points; the priori segmentation module 5 processes the subareas and outputs segmentation results of the subareas; the likelihood module 6 is used for superposing the segmentation results of the subareas in the image space to be segmented to obtain likelihood functions; the oscillography module 7 is used for setting the oscillography function of the subarea in the image space to be segmented, and superposing the oscillography function to obtain prior distribution; and the posterior module 8 is used for combining the prior distribution and the likelihood function to obtain posterior distribution of the target organ, and binarizing the posterior distribution to obtain a final segmentation result.
The prior segmentation module is used for processing the subareas based on the trained learning model and outputting segmentation results of the subareas, and the corresponding learning model setting comprises: performing prior distributed spatial sampling to obtain sampling points; extracting corresponding sub-regions from the registration images of the sample images according to the sampling points; these subregions are used as training samples for the learning model to perform training.
The learning model is based on a convolutional neural network or a support vector machine or random forest.
Spatial sampling of the prior distribution of the target organ based on markov chain monte carlo includes: setting a stable distribution P (x) and a symmetrical state transition matrix q; setting the threshold value of the state transition times as n 1 The number of samples required is n 2 . Starting circulation, starting circulation from t=0, wherein t is more than or equal to 0 and less than or equal to n 1 +n 2 -1; optionally generating an initial state x 0 The method comprises the steps of carrying out a first treatment on the surface of the From a conditional probability distribution q (x|x t ) Mid-sampling to obtain sample x * ，0≤t≤n 1 +n 2 -1; sampling from Uniform distribution form (0, 1) to obtain a random number u; acceptance rate alpha (x) t ,x * )＝min{p(x * )/p(x t ) 1, when u < alpha (x) t ,x * ) Let x t+1 ＝x * Extracting x t+1 A corresponding three-dimensional sub-region. After the circulation is finished, n of the images to be segmented are obtained 2 A sub-region.
The likelihood function includes the sum C of the classification results of the pixel points i The prior distribution includes the number M of times the pixel is classified i The posterior distribution includes the probability V of the pixel point as the target organ t ＝C i /M i 。
While the preferred embodiment of the present invention has been described in detail, the present invention is not limited to the embodiments, and those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are included in the scope of the present invention as defined in the appended claims.
Claims (8)
1. A method of organ image segmentation, comprising:
carrying out nonlinear registration on the plurality of sample images and the image to be segmented to obtain registration coefficients of the plurality of sample images;
transforming the binary image of the target organ of the sample image based on the registration coefficient, and superposing the transformed binary image to obtain prior distribution of the target organ;
performing prior distributed spatial sampling of the target organ based on Markov chain Monte Carlo to obtain a precise sampling point;
extracting corresponding sub-regions from the image to be segmented according to the precise sampling points;
classifying pixels in the subareas, and outputting segmentation results of the subareas;
superposing the segmentation results of the subareas in the image space to be segmented to obtain a likelihood function of a target organ in the image to be segmented;
setting an indication function of the subarea in an image space to be segmented, and superposing the indication function to obtain prior distribution;
combining the prior distribution of the target organ in the image to be segmented and the likelihood function to obtain posterior distribution of the target organ, and binarizing the posterior distribution to obtain a final segmentation result;
wherein, based on Markov chain Monte Carlo, the space sampling of the prior distribution of the target organ is carried out, and the accurate sampling point is obtained, including:
setting a stable distribution P (x) and a symmetrical state transition matrix q;
setting the threshold value of the state transition times as n 1 The number of samples required is n 2 ；
Optionally generating an initial state x 0 ；
From a conditional probability distribution q (x|x t ) Mid-sampling to obtain sample x * ，0≤t≤n 1 + n 2 -1；
Sampling from Uniform distribution form (0, 1) to obtain random number u;
acceptance rate
When u </u->
Let->
Extracting->
And obtaining the prior distributed spatial sampling of the target organ according to the corresponding three-dimensional subareas.
2. The method according to claim 1, comprising processing the subregion based on a trained learning model, outputting a segmentation result of the subregion, and setting the learning model correspondingly includes:
performing prior distributed spatial sampling of the target organ to obtain sampling points;
extracting corresponding sub-regions from the registration images of the sample images according to the sampling points;
and training by taking the subareas as training samples of a learning model.
3. The method of claim 2, wherein the learning model is based on a convolutional neural network or a support vector machine or a random forest.
4. The method of claim 1, wherein the likelihood function comprises a sum C of classification results of pixel points i The prior distribution comprises the number M of times the pixel point is classified i The posterior distribution includes probabilities of pixels as target organs
。
5. An organ image segmentation apparatus, comprising:
the registration module is used for carrying out nonlinear registration on the plurality of sample images and the images to be segmented to obtain registration coefficients of the sample images;
the sample processing module is used for transforming the binary image of the target organ in the sample image based on the registration coefficient, and superposing the transformed binary image to obtain prior distribution of the target organ;
the prior sampling module is used for carrying out prior distributed spatial sampling on the target organ based on Markov chain Monte Carlo to obtain precise sampling points;
the prior region module is used for extracting a corresponding sub-region from the image to be segmented according to the precise sampling points;
the priori segmentation module classifies pixels in the sub-region and outputs segmentation results of the sub-region;
the likelihood module is used for superposing the segmentation results of the subareas in the image space to be segmented to obtain likelihood functions of target organs in the image to be segmented;
the oscillography module is used for setting an oscillography function of the subarea in the image space to be segmented, and superposing the oscillography function to obtain prior distribution of target organs in the image to be segmented;
the posterior module is used for combining the prior distribution of the target organ in the image to be segmented and the likelihood function to obtain posterior distribution of the target organ, and binarizing the posterior distribution to obtain a final segmentation result;
the prior sampling module is configured to perform spatial sampling of prior distribution of the target organ based on markov chain monte carlo, and obtain a precise sampling point, and includes:
setting a stable distribution P (x) and a symmetrical state transition matrix q;
setting the threshold value of the state transition times as n 1 The number of samples required is n 2 ；
Optionally generating an initial state x 0 ；
From a conditional probability distribution q (x|x t ) Mid-sampling to obtain sample x * ，0≤t≤n 1 + n 2 -1；
Sampling from Uniform distribution form (0, 1) to obtain random number u;
acceptance rate
When u </u->
Let->
Extracting->
And obtaining the prior distributed spatial sampling of the target organ according to the corresponding three-dimensional subareas.
6. The apparatus according to claim 5, wherein the prior segmentation module is configured to process the subarea based on a learning model that is completed in training, output a segmentation result of the subarea, and the setting of the learning model includes:
performing spatial sampling of the prior distribution to obtain sampling points;
extracting corresponding sub-regions from the registration images of the sample images according to the sampling points;
and training by taking the subareas as training samples of a learning model.
7. The organ image segmentation device according to claim 6, wherein the learning model is based on a convolutional neural network or a support vector machine or a random forest.
8. The apparatus according to claim 5, wherein the likelihood function includes a sum C of classification results of the pixel points i The prior distribution of the sampling points comprises the classified times M of the pixel points i The posterior distribution includes probabilities of pixels as target organs
。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911029097.8A CN110874843B (en) | 2019-10-28 | 2019-10-28 | Organ image segmentation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911029097.8A CN110874843B (en) | 2019-10-28 | 2019-10-28 | Organ image segmentation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110874843A CN110874843A (en) | 2020-03-10 |
| CN110874843B true CN110874843B (en) | 2023-07-07 |
Family
ID=69717915
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911029097.8A Active CN110874843B (en) | 2019-10-28 | 2019-10-28 | Organ image segmentation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110874843B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111784724B (en) * | 2020-05-28 | 2023-05-09 | 长安大学 | Improved Markov chain Monte Carlo two-dimensional rock slice reconstruction method and system |
| CN113274129A (en) * | 2021-05-11 | 2021-08-20 | 温州医科大学附属第二医院（温州医科大学附属育英儿童医院） | Cardiothoracic surgery auxiliary control system based on virtual reality |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101425184A (en) * | 2008-10-30 | 2009-05-06 | 西安电子科技大学 | Image segmentation method based on second generation Bandelet concealed Markov tree model |
| CN101447080A (en) * | 2008-11-19 | 2009-06-03 | 西安电子科技大学 | Method for segmenting HMT image on the basis of nonsubsampled Contourlet transformation |
| CN105957066A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on automatic context model |
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN106485695A (en) * | 2016-09-21 | 2017-03-08 | 西北大学 | Medical image Graph Cut dividing method based on statistical shape model |
-
2019
- 2019-10-28 CN CN201911029097.8A patent/CN110874843B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101425184A (en) * | 2008-10-30 | 2009-05-06 | 西安电子科技大学 | Image segmentation method based on second generation Bandelet concealed Markov tree model |
| CN101447080A (en) * | 2008-11-19 | 2009-06-03 | 西安电子科技大学 | Method for segmenting HMT image on the basis of nonsubsampled Contourlet transformation |
| CN105957066A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on automatic context model |
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN106485695A (en) * | 2016-09-21 | 2017-03-08 | 西北大学 | Medical image Graph Cut dividing method based on statistical shape model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110874843A (en) | 2020-03-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961303B (en) | Image processing method and device, electronic equipment and computer readable medium | |
| CN109377445B (en) | Model training method, method and device for replacing image background and electronic system | |
| CN107153817B (en) | Pedestrian re-identification data labeling method and device | |
| CN107358258B (en) | SAR image target classification based on NSCT double CNN channels and selective attention mechanism | |
| CN111145209B (en) | Medical image segmentation method, device, equipment and storage medium | |
| CN108492294B (en) | Method and device for evaluating harmony degree of image colors | |
| CN108197644A (en) | A kind of image-recognizing method and device | |
| CN112233129B (en) | Deep learning-based parallel multi-scale attention mechanism semantic segmentation method and device | |
| CN110874843B (en) | Organ image segmentation method and device | |
| CN110909724B (en) | Thumbnail generation method of multi-target image | |
| CN110097616B (en) | Combined drawing method and device, terminal equipment and readable storage medium | |
| CN109712095B (en) | Face beautifying method with rapid edge preservation | |
| CN114663685B (en) | Pedestrian re-recognition model training method, device and equipment | |
| CN110852327A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
| CN114444566A (en) | Image counterfeiting detection method and device and computer storage medium | |
| CN111079626B (en) | Living body fingerprint identification method, electronic equipment and computer readable storage medium | |
| CN110969641A (en) | Image processing method and device | |
| CN113888501B (en) | Attention positioning network-based reference-free image quality evaluation method | |
| CN114724218A (en) | Video detection method, device, equipment and medium | |
| CN108986210B (en) | Method and device for reconstructing three-dimensional scene | |
| CN111985488B (en) | Target detection segmentation method and system based on offline Gaussian model | |
| CN110781936B (en) | Construction method of threshold learnable local binary network based on texture description and deep learning and remote sensing image classification method | |
| CN107704864B (en) | Salient object detection method based on image object semantic detection | |
| CN109949298B (en) | Image segmentation quality evaluation method based on cluster learning | |
| CN115953821A (en) | Virtual face image generation method and device and electronic equipment |
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 |