CN103903255B - A kind of ultrasonic image division method and system - Google Patents
A kind of ultrasonic image division method and system Download PDFInfo
- Publication number
- CN103903255B CN103903255B CN201210590387.1A CN201210590387A CN103903255B CN 103903255 B CN103903255 B CN 103903255B CN 201210590387 A CN201210590387 A CN 201210590387A CN 103903255 B CN103903255 B CN 103903255B
- Authority
- CN
- China
- Prior art keywords
- particle
- segmentation
- module
- point
- training
- 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
Landscapes
- Image Processing (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
The present invention is applicable to technical field of image processing, particularly relates to a kind of ultrasonic image division method and system.Said method comprising the steps of: the three-D ultrasound data of statistical shape model with the designated organ collected is carried out rough registration, obtain initializing coordinate conversion parameter;According to initializing coordinate conversion parameter, utilizing image segmentation algorithm based on particle filter that three-D ultrasound data is iterated segmentation, described statistical shape model is the combination of the mean value by being trained obtaining to the manual segmentation result of multiple fine definition three-dimensional datas and the stack features vector characterizing change mode.So, avoiding manual segmentation and semi-automatic segmentation needs artificial to participate in more problem, compares existing full-automatic partition method, and the present invention solves that image resolution ratio is low and the accuracy problem of segmentation under image blurring state.
Description
Technical field
The invention belongs to technical field of image processing, particularly relate to a kind of ultrasonic image division method and be
System.
Background technology
Image Segmentation Technology has great significance in the many-side such as image-guidance, medical diagnosis on disease.At figure
As, in navigation, the most accurate image segmentation result data could provide good for navigation procedure
Three-dimensional visualization environment, and follow-up path planning function etc. is also based on good segmentation result.
In medical diagnosis on disease, accurate segmentation result can allow doctor preferably observe the shape of internal organs,
And it is provided that some characteristic index of internal organs is to assist diagnosis.Current cutting techniques can divide
For three classes manual, semi-automatic, full-automatic.Manual segmentation relies on doctor personal experience, need to be to graphics
Each frame as data manually sketches the contours the edge of appointment internal organs, wastes time and energy, work load weight,
And due to subjective, it is possible to cause the problem that accuracy is the lowest.Semi-automatic segmentation technology depends on
Rely and preferably initialize, therefore typically require the information that doctor provides extra, be then given via algorithm
Segmentation result.Although need not the substantial amounts of work sketching the contours edge etc, but it relies on and initializes, still
So need man-machine interaction, it is possible to normal surgical procedure is interfered.And full-automatic dividing technology
Then fully rely on algorithm process image and obtain specifying the segmentation result of internal organs, but fully automatic technique is often
The problem that there are the aspects such as Stability and veracity, is extremely difficult to satisfied in the case of poor image quality
Segmentation effect.
Summary of the invention
In order to solve above-mentioned technical problem, it is an object of the invention to provide a kind of Ultrasound Image Segmentation side
Method.
The present invention is achieved in that a kind of ultrasonic image division method, and described method includes following step
Rapid:
The three-D ultrasound data of statistical shape model with the designated organ collected is carried out rough registration,
To initializing coordinate conversion parameter;
According to initializing coordinate conversion parameter, utilize image segmentation algorithm based on particle filter to three-dimensional
Ultrasound data is iterated segmentation,
Described statistical shape model is by tying the manual segmentation of multiple fine definition 3 d image datas
Fruit is trained the combination of mean value and the stack features vector characterizing change mode obtained.
Further, the foundation of described statistical shape model comprises the following steps:
The 3-D view number at the designated organ position of multiple volunteers is gathered by high resolution imaging method
According to, and carried out manual segmentation by professional person;
For each segmentation result, carry out resurfacing, and gridding;
For each grid after gridding, from its summit, respectively choose the summit composition setting quantity
Training point set, described training set need to meet: has the some one_to_one corresponding of same index in training set;
Training point set is carried out principal component analysis, obtains mean value point set u and its association side of training set
The feature space of difference matrix;
The characteristic vector constituting described feature space is pressed the big minispread of characteristic value, selects correspondence in order
Multiple characteristic vectors that characteristic value is big so that its character pair value sum reaches the threshold requirement set,
Described training set mean value point set u and characteristic vector v being selectediCollectively form Statistical Shape mould
Type.
Further, the described three-D ultrasound data by statistical shape model with the designated organ collected
Carrying out rough registration, the step obtaining initializing coordinate conversion parameter includes:
A range estimation is manually chosen closest to organ of interest geometric center in three-D ultrasound data
Point;
Translation relation conduct between geometric center and this point of training set mean value in statistical shape model
Rough registration.
Further, described according to initializing coordinate conversion parameter, utilize image based on particle filter
Partitioning algorithm is iterated the step of segmentation and includes three-D ultrasound data:
Step A, centered by initializing coordinate conversion parameter, parameter space is chosen multiple point,
As particle;
Step B, to each particle, by the coordinate conversion parameter that it is representative, to training point set
Mean value and characteristic vector carry out coordinate transform, are multiplied by each self-corresponding by the characteristic vector after conversion
Weight, is added with the training points ensemble average value after conversion, obtains the prior model corresponding to each particle,
In for the first time circulation, weight corresponding to each characteristic vector is 0, with the priori that each particle is corresponding
Model is seed point set, respectively according to the gray value information of three-dimensional ultrasound pattern, sets up figure, application drawing
Shape partitioning algorithm, by minimizing given object function, splits figure, and obtains segmentation result
Surface point;
Step C, surface point for each particle, registrate with corresponding prior model respectively,
And calculate the similarity of itself and corresponding prior model, all of particle according to sequencing of similarity and is carried out
Resampling, selects an optimal particle, by the surface point corresponding to optimal particle with corresponding
Prior model in the difference of mean value project to characteristic of correspondence vector and constitute spatially, calculate each
The weight of characteristic vector;
The rate of change between the optimal particle that before and after step D, calculating, twice circulation obtains, if less than
Set threshold value, be considered as algorithmic statement, or circulate beyond set point number, end loop, otherwise, with
Optimal particle, as the initialization changes in coordinates parameter in step A, is made with the weight obtained in step C
For each characteristic vector respective weights in step B, return to step A and re-execute.
Another object of the present invention is to provide a kind of Ultrasound Image Segmentation system, described system includes adopting
Collect the acquiring ultrasound image system of the three-dimensional ultrasound pattern of the designated organ of patient and to three collected
Dimension ultrasonoscopy carries out the image processing system processed, and described image processing system includes:
Rough registration module, for the three-D ultrasonic number by statistical shape model with the designated organ collected
According to carrying out rough registration, obtain initializing coordinate conversion parameter;
Iterative segmentation module, for according to initializing coordinate conversion parameter, utilizing based on particle filter
Image segmentation algorithm is iterated segmentation to three-D ultrasound data,
Described statistical shape model is by tying the manual segmentation of multiple fine definition 3 d image datas
Fruit is trained the combination of mean value and the stack features vector characterizing change mode obtained.
Further, described Ultrasound Image Segmentation system also includes the graphics gathering patient's designated organ
As the computed tomography (SPECT) system for setting up statistical shape model, described image processing system also includes statistics
Shape sets up module, and described statistical shape model is set up module and included:
3-D view manual segmentation editor module, is used for receiving professional person to computed tomography (SPECT) system collection
Multiple volunteers designated organ position 3 d image data segmentation editor;
Gridding processing module, for for each segmentation result, carries out resurfacing, and grid
Change;
Training point set chooses module, for for each grid after gridding, each from its summit
Choosing the summit composing training point set setting quantity, described training set need to meet: has phase in training set
With lower target point one_to_one corresponding;
Training points set analysis module, for training point set is carried out principal component analysis, obtains training set
Mean value point set u and the feature space of its covariance matrix;
Characteristic vector chooses module, big by characteristic value for the characteristic vector that will constitute described feature space
Minispread, selects the big multiple characteristic vectors of character pair value in order so that its character pair value it
With reach set threshold requirement, described training set mean value point set u and the characteristic vector being selected
viCollectively form statistical shape model.
Further, described rough registration module also includes:
Geometric center point chooses input module, for manually choosing a range estimation in three-D ultrasound data
Point closest to organ of interest geometric center;
Coordinate conversion relation computing module, in statistical shape model in the geometry of training set mean value
Translation relation between the heart and this point is as rough registration.
Further, described iterative segmentation module includes:
Particle chooses module, for centered by initializing coordinate conversion parameter, selects in parameter space
Take multiple point, as particle;
Surface point acquisition module, for each particle, by the coordinate conversion parameter that it is representative,
Mean value and characteristic vector to training point set carry out coordinate transform, the characteristic vector after conversion are taken advantage of
With each self-corresponding weight, it is added with the training points ensemble average value after conversion, obtains corresponding to each grain
The prior model of son, in for the first time circulation, weight corresponding to each characteristic vector is 0, with each
The prior model that particle is corresponding is seed point set, respectively according to the gray value information of three-dimensional ultrasound pattern,
Set up figure, Graphics Application partitioning algorithm, by minimizing given object function, figure split,
And obtain the surface point of segmentation result;
Similarity registration module, for the surface point for each particle, respectively with corresponding priori
Model registrates, and calculates the similarity of itself and corresponding prior model, by all of particle according to phase
Sort like degree and carry out resampling, selecting an optimal particle, corresponding to optimal particle
Surface point projects to the sky that characteristic of correspondence vector is constituted with the difference of mean value in corresponding prior model
On between, calculate the weight of each characteristic vector;
The optimal particle rate of change calculates judge module, the optimal grain that twice circulation obtains before and after calculating
The rate of change between son, if less than setting threshold value, is considered as algorithmic statement, or circulates beyond setting
Number of times, end loop, otherwise, choose the initialization coordinate in module using optimal particle as particle and become
Changing parameter, the weight obtained in similarity registration module is as each spy in surface point acquisition module
Levy vector respective weights, return to particle and choose module and re-execute.
The present invention uses statistical shape model to combine with image based on class particle filter pattern segmentation
Appointment internal organs in ultrasonoscopy are split by alternative manner.Using statistical model as priori,
Combine with gradation of image information, use the form of particle filter, organ is iterated segmentation.Repeatedly
For cutting procedure makes full use of the information that statistical model provides, the highest and thoughts in image definition
Interest organ remains to the effect being more satisfied with in the case of having edge blurry.Use particle filter
Form, effectively avoids local minimum problem.So, manual segmentation and semi-automatic point are avoided
Cutting the problem needing artificial participation more, compare existing full-automatic partition method, the present invention solves figure
As resolution ratio is low and the accuracy problem of segmentation under image blurring state.
Accompanying drawing explanation
Fig. 1 is the flow chart of the ultrasonic image division method that the embodiment of the present invention provides;
Fig. 2 is the method for building up flow chart of the statistical shape model that the embodiment of the present invention provides;
Fig. 3 is the implementing procedure figure of the Image Iterative dividing method that the embodiment of the present invention provides;
Fig. 4 is the structural representation of the Ultrasound Image Segmentation system that the embodiment of the present invention provides.
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing
And embodiment, the present invention is further elaborated.Should be appreciated that described herein specifically
Embodiment only in order to explain the present invention, is not intended to limit the present invention.
Fig. 1 shows the flow process of the ultrasonic image division method that the embodiment of the present invention provides, and details are as follows:
In step S101, by the three-D ultrasonic number of statistical shape model with the designated organ collected
According to carrying out rough registration, obtain initializing coordinate conversion parameter.
In the present invention, this statistical shape model is by multiple fine definition 3 d image datas
Manual segmentation result is trained the group of mean value and the stack features vector characterizing change mode obtained
Close.
As embodiments of the invention, this step includes again: manually choose one in three-D ultrasound data
Individual range estimation is closest to the point of organ of interest geometric center;And training set is average in statistical shape model
Translation relation between geometric center and this point of value is as rough registration.
This preferred version uses following scheme: in three-D ultrasound data, manually choose one estimate
Point close to organ of interest geometric center.The geometric center of training set mean value in statistical shape model
And the translation relation between this point is as rough registration.The coordinate defining coordinate system A to coordinate system B at this becomes
Change and parameter space, this preferred version use sextuple parameter space, be respectively around X-axis anglec of rotation αX,
Around Y-axis anglec of rotation αY, anglec of rotation α about the z axisZ, translate T along X-axisX, translate T along Y-axisY, along Z
Axle translation TZ.It represents: by coordinate system A by around X-axis rotation alphaX, further around postrotational Y-axis rotation alphaY,
According still further to postrotational Z axis rotation alphaZ, coordinate system the most after rotation translates T along X-axis respectivelyX,
T is translated along Y-axisY, translate T along Z axisZOverlap with coordinate system B afterwards.
During rough registration, multiple clear and that anatomical position is clear and definite index point can be chosen, use
The methods such as the such as ICP of the method for registering between point set carry out rough registration.
In step s 102, according to initializing coordinate conversion parameter, figure based on particle filter is utilized
As three-D ultrasound data is iterated splitting by partitioning algorithm.
Fig. 2 shows the method for building up flow process of the statistical shape model that the embodiment of the present invention provides, and describes in detail
As follows:
In step s 201, the designated organ of multiple volunteers is gathered by high resolution imaging method
The 3 d image data at position, and carried out manual segmentation by professional person.As embodiments of the invention,
Described high resolution imaging method is MRI or CT.
In step S202, for each segmentation result, carry out resurfacing, and gridding.
In gridding, this preferred version have employed triangle gridding, it is of course possible to uses other any conjunctions
The grid of conformal shape.
In step S203, for each grid after gridding, respectively choose from its summit and set
The summit composing training point set of determined number, described training set need to meet: has same index in training set
Some one_to_one corresponding.
In step S204, training point set is carried out principal component analysis (Principle Component
Analysis, PCA), obtain the mean value point set u of training set and the feature sky of its covariance matrix
Between.
In step S205, the characteristic vector constituting described feature space is pressed the big minispread of characteristic value,
Select multiple characteristic vectors that character pair value is big in order so that its character pair value sum reaches to set
Fixed threshold requirement, described training set mean value point set u and characteristic vector v being selectediCommon structure
Become statistical shape model.
Fig. 3 shows the implementing procedure of the Image Iterative dividing method that the embodiment of the present invention provides, and describes in detail
As follows:
In step S301, centered by initializing coordinate conversion parameter, choose in parameter space
Multiple, as particle.
In this preferred version, to each parameter, press with initialization coordinate conversion parameter as average
The normal distribution of given variance, randomly selects particle point.In circulation for the first time, with rough registration result
As initializing coordinate conversion parameter.
In step s 302, to each particle, by the coordinate conversion parameter that it is representative, to instruction
Mean value and the characteristic vector of practicing point set carry out coordinate transform, the characteristic vector after conversion are multiplied by respectively
Self-corresponding weight, is added with the training points ensemble average value after conversion, obtains corresponding to each particle
Prior model, in for the first time circulation, weight corresponding to each characteristic vector is 0, with each particle
Corresponding prior model is seed point set, respectively according to the gray value information of three-dimensional ultrasound pattern, sets up
Figure, Graphics Application partitioning algorithm, by minimizing given object function, figure is split, and take
Obtain the surface point of segmentation result.
In this preferred version, the constituted mode of figure is as follows: each pixel is as a joint of figure
Point, square being inversely proportional to of the difference of the channel capacity between adjacent two nodes and two node gray values.
In step S303, for the surface point of each particle, respectively with corresponding prior model
Registrate, and calculate the similarity of itself and corresponding prior model, by all of particle according to similarity
Sort and carry out resampling, selecting an optimal particle, by the surface corresponding to optimal particle
Point set projects to the space that characteristic of correspondence vector is constituted with the difference of mean value in corresponding prior model
On, calculate the weight of each characteristic vector.
In this preferred version, registration approach uses ICP algorithm.In this preferred version, similarity
It is defined as follows: to each some Pi in surface point, finds its closest approach Qi in prior model,
Calculate L2 distance L of point-to-point transmissioni 2=(pi-qi)2.Similarity is defined as p=∑ 1/Li 2, to there being a Pi.
In step s 304, the rate of change between the optimal particle that before and after calculating, twice circulation obtains,
If less than setting threshold value, it is considered as algorithmic statement, or circulates beyond set point number, end loop,
Otherwise, using optimal particle as the initialization changes in coordinates parameter in step S301, with step S303
In the weight that obtains as each characteristic vector respective weights in step S302, return to step S301 weight
New execution.
Fig. 4 shows the structure of the Ultrasound Image Segmentation system that the embodiment of the present invention provides, this ultrasonic figure
The acquiring ultrasound image system of the three-dimensional ultrasound pattern of the designated organ of collection patient is included as segmenting system
1 and image processing system 2 that the three-dimensional ultrasound pattern collected is processed, described image procossing
System 2 includes: rough registration module 21 and iterative segmentation module 22.
Being manually entered to coordinate, this Ultrasound Image Segmentation system also includes that human-computer interaction module is (in figure
Not shown).In order to clearly show the image result after segmentation, this Ultrasound Image Segmentation system also includes showing
Showing device (not shown), this display device can show the image result after segmentation.
This rough registration module 21 is by the three-D ultrasonic number of statistical shape model with the designated organ collected
According to carrying out rough registration, obtain initializing coordinate conversion parameter;Iterative segmentation module 22 is according to initializing seat
Mark transformation parameter, utilizes image segmentation algorithm based on particle filter to be iterated three-D ultrasound data
Segmentation, described statistical shape model is by the manual segmentation to multiple fine definition 3 d image datas
Result is trained the combination of mean value and the stack features vector characterizing change mode obtained.
As embodiments of the invention, described Ultrasound Image Segmentation system also includes that gathering patient specifies device
The 3-D view of official is for setting up the computed tomography (SPECT) system 3 of statistical shape model, described image procossing system
System also includes that statistical shape model sets up module 20, and described statistical shape model is set up module 20 and included:
3-D view manual segmentation editor module 201, gridding processing module 202, training point set choose module
203, training points set analysis module 204 and characteristic vector choose module 205.
This 3-D view manual segmentation editor module 201 receives professional person to computed tomography (SPECT) system collection
Multiple volunteers designated organ position 3 d image data segmentation editor;Gridding processes mould
Block 202, for for each segmentation result, carries out resurfacing, and gridding;Training point set choosing
Delivery block 203, for for each grid after gridding, respectively chooses setting quantity from its summit
Summit composing training point set, described training set need to meet: has the point one of same index in training set
One is corresponding;Training points set analysis module 204, for training point set is carried out principal component analysis, is instructed
Practice mean value point set u and the feature space of its covariance matrix of collection;Characteristic vector chooses module 205
For the characteristic vector by constituting described feature space by the big minispread of characteristic value, select correspondence in order
Multiple characteristic vectors that characteristic value is big so that its character pair value sum reaches the threshold requirement set,
Described training set mean value point set u and characteristic vector v being selectediCollectively form Statistical Shape mould
Type.
As embodiments of the invention, described rough registration module 21 also includes: geometric center point is chosen defeated
Enter module 211 and coordinate conversion relation computing module 212.This geometric center point chooses input module
211 for manually choosing a range estimation closest to organ of interest geometric center in three-D ultrasound data
Point;Coordinate conversion relation computing module 212 training set mean value several in statistical shape model
What translation relation between center and this point is as rough registration.
As embodiments of the invention, this iterative segmentation module 22 includes: particle choose module 221,
Surface point acquisition module 222, similarity registration module 223 and the optimal particle rate of change calculate and sentence
Disconnected module 224.
Particle chooses module 221 for initialize centered by coordinate conversion parameter, in parameter space
Choose multiple point, as particle;Surface point acquisition module 222 is for each particle, by it
Representative coordinate conversion parameter, carries out coordinate transform to the mean value and characteristic vector training point set,
Characteristic vector after conversion is multiplied by each self-corresponding weight, with the training points ensemble average value phase after conversion
Adding, obtain the prior model corresponding to each particle, in circulation for the first time, each characteristic vector is corresponding
Weight be 0, with prior model corresponding to each particle for seed point set, surpass according to three-dimensional respectively
The gray value information of acoustic image, sets up figure, and Graphics Application partitioning algorithm, by minimizing to setting the goal
Function, splits figure, and obtains the surface point of segmentation result;Similarity registration module 223
For the surface point for each particle, registrate with corresponding prior model respectively, and calculate
Its similarity with corresponding prior model, according to sequencing of similarity and carries out resampling by all of particle,
Select an optimal particle, by the surface point corresponding to optimal particle and corresponding priori mould
In type, the difference of mean value projects to characteristic of correspondence vector composition spatially, calculates each characteristic vector
Weight;The optimal particle rate of change calculates what judge module 224 twice circulation before and after calculating obtained
The rate of change between optimal particle, if less than setting threshold value, is considered as algorithmic statement, or circulation is super
Cross set point number, end loop, otherwise, choose the initialization in module using optimal particle as particle
Changes in coordinates parameter, the weight obtained in similarity registration module is as in surface point acquisition module
Each characteristic vector respective weights, return to particle and choose module and re-execute.
For the computed tomography (SPECT) system 3 in the present invention, this preferred version uses nuclear magnetic resonance
(MRI) imaging, it would however also be possible to employ other suitable fine definition such as CT high-resolution fault imaging system
System;Ultrasonic probe in the present invention can be two-dimensional ultrasound probe, it is also possible to be three dimensional ultrasound probe (i.e.
Volume imaging probe), as long as the probe that can obtain three-D ultrasonic volume data all can use.
In sum, the dividing method that the present invention provides first to set up statistical shape model.Certainly, also
The statistical shape model being ready for can be introduced directly into Ultrasound Image Segmentation system.Thus avoid
Every equipment is required for setting up statistical shape model.The embodiment of the present invention is divided into two stages.Set up
The process of statistical shape model is: by professional person's high-resolution scan data to a large amount of volunteer's samples
Carry out manual segmentation.Segmentation result is carried out resurfacing gridding.Each grid selects
A number of summit composing training point set.Training point set must meet following condition: some concentration has phase
With lower target point one_to_one corresponding.Training point set is carried out PCA(Principle Component Analysis)
Analyze, obtain training mean value and the feature space of covariance matrix of point set.Empty from constitutive characteristic
Between vector in by character pair value size sequence choose series of features vector so that it is characteristic value sum
More than certain threshold value.The mean value training point set and the characteristic vector being selected constitute Statistical Shape mould
Type.
Image partition method is carried out by following flow process: one, initialize.System is determined by certain criterion
Rough coordinates transformation parameter between meter shape and three-dimensional ultrasound pattern, and defined parameters space.Two,
Image based on class particle filter pattern segmentation iterative splitting algorithm.Carry out as follows: 1, with just
Centered by beginningization coordinate conversion parameter, in parameter space, choose a number of point by certain standard,
As particle.In circulation for the first time, the coordinate conversion parameter obtained in step one is sat for initializing
Mark transformation parameter.2, to each particle, by the coordinate conversion parameter that it is representative, to training point set
Mean value and characteristic vector carry out coordinate transform.Characteristic vector after conversion is multiplied by respective correspondence
Weight, with conversion after training points ensemble average value be added, obtain the priori mould corresponding to each particle
Type.In for the first time circulation, weight corresponding to each characteristic vector is 0.With the elder generation that each particle is corresponding
Testing model is seed point set, respectively according to the gray value information of three-dimensional ultrasound pattern, with certain principle
Set up figure, application graph cut(image segmentation) algorithm splits, and obtains the surface of segmentation result
Point set.3, to the surface point corresponding to each particle obtained in 2, respectively with corresponding priori mould
Type registrates, and calculates the similarity of itself and corresponding prior model according to certain definition.By all of
Particle is according to sequencing of similarity and carries out resampling, selects an optimal particle.By optimal grain
Surface point corresponding to son projects to what characteristic of correspondence vector was constituted with the difference of corresponding prior model
Spatially, the weight of each characteristic vector is calculated.4, the optimal particle that before and after calculating, twice circulation obtains
Between the rate of change, if less than certain threshold value, be considered as algorithmic statement, or circulate beyond a fixed number
Amount, end loop.Otherwise, using optimal particle as the initialization changes in coordinates parameter in 1, in 3
The weight obtained, as each characteristic vector respective weights in 2, returns to 1 and re-executes.
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all
Any amendment, equivalent and the improvement etc. made within the spirit and principles in the present invention, all should comprise
Within protection scope of the present invention.
Claims (6)
1. a ultrasonic image division method, it is characterised in that said method comprising the steps of:
The three-D ultrasound data of statistical shape model with the designated organ collected is carried out rough registration, at the beginning of obtaining
Beginningization coordinate conversion parameter;
According to initializing coordinate conversion parameter, utilize image segmentation algorithm based on particle filter to three-D ultrasonic
Data are iterated segmentation,
Described statistical shape model is by entering the manual segmentation result of multiple fine definition 3 d image datas
The mean value that row training obtains and the combination characterizing the stack features vector changing mode;
Described according to initializing coordinate conversion parameter, utilize image segmentation algorithm based on particle filter to three-dimensional
Ultrasound data is iterated the step of segmentation and includes:
Step A, centered by initializing coordinate conversion parameter, parameter space is chosen multiple point, as
Particle;
Step B, to each particle, by the coordinate conversion parameter that it is representative, average to training point set
Value and characteristic vector carry out coordinate transform, and the characteristic vector after conversion is multiplied by each self-corresponding weight, with
Training points ensemble average value after conversion is added, and obtains the prior model corresponding to each particle, follows for the first time
In ring, weight corresponding to each characteristic vector is 0;With prior model corresponding to each particle for seed point set,
Respectively according to the gray value information of three-dimensional ultrasound pattern, setting up figure, Graphics Application partitioning algorithm, by minimum
Change given object function, figure is split, and obtains the surface point of segmentation result;
Step C, surface point for each particle, registrate with corresponding prior model respectively, and
Calculate the similarity of itself and corresponding prior model, all of particle according to sequencing of similarity and carried out resampling,
Select an optimal particle, by the surface point corresponding to optimal particle with in corresponding prior model
The difference of mean value projects to characteristic of correspondence vector and constitutes spatially, calculates the weight of each characteristic vector;
The rate of change between the optimal particle that before and after step D, calculating, twice circulation obtains, if less than setting
Threshold value, is considered as algorithmic statement, or circulates beyond set point number, end loop, otherwise, with optimal particle
As the initialization changes in coordinates parameter in step A, the weight obtained in step C is as in step B
Each characteristic vector respective weights, returns to step A and re-executes.
Ultrasonic image division method the most according to claim 1, it is characterised in that described Statistical Shape
The foundation of model comprises the following steps:
The 3 d image data at the designated organ position of multiple volunteers is gathered by high resolution imaging method,
And carried out manual segmentation by professional person;
For each segmentation result, carry out resurfacing, and gridding;
For each grid after gridding, from its summit, respectively choose the summit composing training setting quantity
Point set, described training set need to meet: has the some one_to_one corresponding of same index in training set;
Training point set is carried out principal component analysis, obtains mean value point set u and its covariance square of training set
The feature space of battle array;
The characteristic vector constituting described feature space is pressed the big minispread of characteristic value, selects characteristic value big in order
Multiple characteristic vectors so that its character pair value sum reach set threshold requirement, described training set put down
Average point set u and characteristic vector v being selectediCollectively form statistical shape model.
Ultrasonic image division method the most according to claim 1, it is characterised in that described will add up shape
Shape model carries out rough registration with the three-D ultrasound data of the designated organ collected, and obtains initializing coordinate transform
The step of parameter includes:
The range estimation point closest to organ of interest geometric center is manually chosen in three-D ultrasound data;
In statistical shape model, the translation relation between geometric center and this point of training set mean value is as slightly joining
Accurate.
4. a Ultrasound Image Segmentation system, it is characterised in that described system includes the appointment device gathering patient
The acquiring ultrasound image system of the three-dimensional ultrasound pattern of official and the three-dimensional ultrasound pattern collected is processed
Image processing system, described image processing system includes:
Rough registration module, for entering the three-D ultrasound data of statistical shape model with the designated organ collected
Row rough registration, obtains initializing coordinate conversion parameter;
Iterative segmentation module, for according to initializing coordinate conversion parameter, utilizing image based on particle filter
Partitioning algorithm is iterated segmentation to three-D ultrasound data,
Described statistical shape model is by entering the manual segmentation result of multiple fine definition 3 d image datas
The mean value that row training obtains and the combination characterizing the stack features vector changing mode;
Described iterative segmentation module includes:
Particle chooses module, for, centered by initializing coordinate conversion parameter, choosing many in parameter space
Individual, as particle;
Surface point acquisition module, for each particle, by the coordinate conversion parameter that it is representative, right
Mean value and the characteristic vector of training point set carry out coordinate transform, the characteristic vector after conversion are multiplied by each
Corresponding weight, is added with the training points ensemble average value after conversion, obtains the priori mould corresponding to each particle
Type, in for the first time circulation, weight corresponding to each characteristic vector is 0;With the priori that each particle is corresponding
Model is seed point set, respectively according to the gray value information of three-dimensional ultrasound pattern, sets up figure, and Graphics Application divides
Cut algorithm, by minimizing given object function, figure is split, and obtain the surface point of segmentation result
Collection;
Similarity registration module, for the surface point for each particle, respectively with corresponding prior model
Registrate, and calculate the similarity of itself and corresponding prior model, by all of particle according to sequencing of similarity
And carry out resampling, select an optimal particle, by the surface point corresponding to optimal particle with right
In the prior model answered, the difference of mean value projects to characteristic of correspondence vector composition spatially, calculates each spy
Levy the weight of vector;
The optimal particle rate of change calculates judge module, the optimal particle that twice circulation obtains before and after calculating it
Between the rate of change, if less than set threshold value, be considered as algorithmic statement, or circulate beyond set point number, knot
Shu Xunhuan, otherwise, chooses the initialization changes in coordinates parameter in module, with phase using optimal particle as particle
Like the weight obtained in degree registration module as each characteristic vector respective weights in surface point acquisition module,
Return to particle choose module and re-execute.
Ultrasound Image Segmentation system the most according to claim 4, it is characterised in that described ultrasonoscopy
Segmenting system also includes that the 3-D view gathering patient's designated organ becomes for the tomography setting up statistical shape model
As system, described image processing system also includes that statistical shape model sets up module, described statistical shape model
Set up module to include:
3-D view manual segmentation editor module, gathers many for receiving professional person to computed tomography (SPECT) system
The segmentation editor of the 3 d image data at the designated organ position of individual volunteer;
Gridding processing module, for for each segmentation result, carries out resurfacing, and gridding;
Training point set chooses module, for for each grid after gridding, respectively chooses from its summit
Setting the summit composing training point set of quantity, described training set need to meet: has same index in training set
Point one_to_one corresponding;
Training points set analysis module, for training point set is carried out principal component analysis, obtains the average of training set
Value point set u and the feature space of its covariance matrix;
Characteristic vector chooses module, for the characteristic vector by constituting described feature space by the big float of characteristic value
Row, select multiple characteristic vectors that character pair value is big so that its character pair value sum reaches to set in order
Fixed threshold requirement, described training set mean value point set u and characteristic vector v being selectediCollectively form system
Meter shape.
Ultrasound Image Segmentation system the most according to claim 4, it is characterised in that described rough registration mould
Block also includes:
Geometric center point chooses input module, connects most for manually choosing a range estimation in three-D ultrasound data
The point of nearly organ of interest geometric center;
Coordinate conversion relation computing module, in statistical shape model the geometric center of training set mean value with
Translation relation between this point is as rough registration.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210590387.1A CN103903255B (en) | 2012-12-31 | 2012-12-31 | A kind of ultrasonic image division method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210590387.1A CN103903255B (en) | 2012-12-31 | 2012-12-31 | A kind of ultrasonic image division method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103903255A CN103903255A (en) | 2014-07-02 |
CN103903255B true CN103903255B (en) | 2016-08-31 |
Family
ID=50994562
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210590387.1A Active CN103903255B (en) | 2012-12-31 | 2012-12-31 | A kind of ultrasonic image division method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103903255B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109716389A (en) * | 2016-09-05 | 2019-05-03 | 光线搜索实验室公司 | The image processing system and method for interaction outline for 3 D medical data |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107766874B (en) * | 2017-09-07 | 2021-06-04 | 深圳度影医疗科技有限公司 | Measuring method and measuring system for ultrasonic volume biological parameters |
CN109472791B (en) * | 2018-10-31 | 2020-10-27 | 深圳大学 | Ultrasonic image segmentation method and computer equipment |
CN109934861B (en) * | 2019-01-22 | 2022-10-18 | 广东工业大学 | Head and neck multi-modal medical image automatic registration method |
CN113284243A (en) * | 2021-03-02 | 2021-08-20 | 上海交通大学 | Method and system for establishing statistical shape model of three-dimensional surface model, storage medium and terminal |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2279695A1 (en) * | 2009-07-31 | 2011-02-02 | Medison Co., Ltd. | Sensor coordinate calibration in an ultrasound system |
CN102110298A (en) * | 2009-12-25 | 2011-06-29 | 新奥特(北京)视频技术有限公司 | Method and device for projecting three-dimensional model in virtual studio system |
-
2012
- 2012-12-31 CN CN201210590387.1A patent/CN103903255B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2279695A1 (en) * | 2009-07-31 | 2011-02-02 | Medison Co., Ltd. | Sensor coordinate calibration in an ultrasound system |
CN102110298A (en) * | 2009-12-25 | 2011-06-29 | 新奥特(北京)视频技术有限公司 | Method and device for projecting three-dimensional model in virtual studio system |
Non-Patent Citations (3)
Title |
---|
基于统计信息及个体信息的统计形状模型的肝脏分割;李春丽等;《南方医科大学学报》;20120101;第32卷(第1期);第23-27页 * |
基于配准与ASM方法的医学图像分割;曹会志等;《中国医学影像技术》;20060731;第22卷(第7期);第1108-1111页 * |
散乱点云自动配准技术研究;李选富;《中国优秀硕士学位论文全文数据库》;20120215;论文中第5、7-8页 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109716389A (en) * | 2016-09-05 | 2019-05-03 | 光线搜索实验室公司 | The image processing system and method for interaction outline for 3 D medical data |
CN109716389B (en) * | 2016-09-05 | 2020-04-14 | 光线搜索实验室公司 | Image processing system and method for interactive contouring of three-dimensional medical data |
Also Published As
Publication number | Publication date |
---|---|
CN103903255A (en) | 2014-07-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109272510B (en) | Method for segmenting tubular structure in three-dimensional medical image | |
JP6877868B2 (en) | Image processing equipment, image processing method and image processing program | |
US9697639B2 (en) | Three-dimensional model data generation device, method and program | |
CN105559813B (en) | Medical diagnostic imaging apparatus and medical image-processing apparatus | |
US8837789B2 (en) | Systems, methods, apparatuses, and computer program products for computer aided lung nodule detection in chest tomosynthesis images | |
EP2916738B1 (en) | Lung, lobe, and fissure imaging systems and methods | |
Bade et al. | Comparison of fundamental mesh smoothing algorithms for medical surface models. | |
CN105096310B (en) | Divide the method and system of liver in magnetic resonance image using multi-channel feature | |
CN106651895B (en) | Method and device for segmenting three-dimensional image | |
CN109493347A (en) | The method and system that the object of sparse distribution is split in the picture | |
EP1365356A2 (en) | Semi-automatic segmentation algorithm for pet oncology images | |
CN103903255B (en) | A kind of ultrasonic image division method and system | |
CN109598697B (en) | Determination of a two-dimensional mammography data set | |
CN103514597A (en) | Image processing device | |
CN103562960B (en) | For generating the assigned unit between the image-region of image and element class | |
JPH0731739B2 (en) | Method and apparatus for extracting faces in a two-dimensional tomographic slice | |
CN105074775A (en) | Registration of medical images | |
US9773325B2 (en) | Medical imaging data processing apparatus and method | |
JP2008511366A (en) | Feature-weighted medical object contour detection using distance coordinates | |
EP2401719B1 (en) | Methods for segmenting images and detecting specific structures | |
CN107708550A (en) | For the surface modeling for the segmentation acoustic echo structure for detecting and measuring anatomic abnormalities | |
CN110313941B (en) | Data processing method, device, equipment and storage medium | |
CN113272869A (en) | Three-dimensional shape reconstruction from topograms in medical imaging | |
CN106408576A (en) | Automatic region of interest segmentation method and system based on three-dimensional ultrasonic image | |
CN109191475A (en) | Terminal plate of vertebral body dividing method, device and computer readable storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |