CN103903255A - Ultrasound image segmentation method and system - Google Patents
Ultrasound image segmentation method and system Download PDFInfo
- Publication number
- CN103903255A CN103903255A CN201210590387.1A CN201210590387A CN103903255A CN 103903255 A CN103903255 A CN 103903255A CN 201210590387 A CN201210590387 A CN 201210590387A CN 103903255 A CN103903255 A CN 103903255A
- Authority
- CN
- China
- Prior art keywords
- particle
- module
- point
- training
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention is suitable for the technical field of image processing, and particularly relates to an ultrasound image segmentation method and system. The method includes the following steps that coarse registration is conducted on a statistical shape model and collected three-dimensional ultrasound data of a specified organ to obtain initialized coordinate transformation parameters; according to the initialized coordinate transformation parameters, an image segmentation algorithm based on particle filter is used for conducting iteration segmentation on the three-dimensional ultrasound data, wherein the statistical shape model is a combination of an average value obtained by training manual segmentation results of a plurality of high-definition three-dimensional data and a group of feature vectors of a representation change mode. Therefore, the problem that more artificial participation is needed in manual segmentation and semi-automatic segmentation is solved. Compared with an existing full-automatic segmentation method, the ultrasound image segmentation method and system solve the problems of low image resolution and segmentation accuracy under the image fuzzy condition.
Description
Technical field
The invention belongs to technical field of image processing, relate in particular to a kind of ultrasonic image division method and system.
Background technology
Image Segmentation Technology has great significance in the many-side such as image-guidance, medical diagnosis on disease.In image-guidance, only have comparatively accurate image segmentation result data to provide good three-dimensional visualization environment for navigation procedure, and follow-up path planning function etc. is also based on good segmentation result.
In medical diagnosis on disease, the shape that accurate segmentation result can allow doctor better observe internal organs, and can provide some characteristic index of internal organs to assist diagnosis.Current cutting techniques can be divided into manual, semi-automatic, full-automatic three classes.Manually cut apart and rely on doctor personal experience, need manually sketch the contours the edge of specifying internal organs to each frame of 3 d image data, waste time and energy, work load weight, and due to subjective, just may cause the problem that accuracy is too low.Semi-automatic cutting techniques relies on good initialization, therefore conventionally needs doctor that extra information is provided, and then provides segmentation result via algorithm.Although do not need a large amount of work of sketching the contours edge and so on, it relies on initialization, still needs man-machine interaction, likely normal surgical procedure is caused to interference.Full-automatic dividing technology relies on algorithm process image to obtain specifying the segmentation result of internal organs completely, but fully automatic technique often exists the problem of the aspects such as Stability and veracity, is difficult to the segmentation effect that reaches satisfied in the situation that of poor image quality.
Summary of the invention
In order to solve the problems of the technologies described above, the object of the present invention is to provide a kind of ultrasonic image division method.
The present invention is achieved in that a kind of ultrasonic image division method, said method comprising the steps of:
Statistical shape model and the three-D ultrasound data of the designated organ collecting are carried out to thick registration, obtain initialization coordinate conversion parameter;
According to initialization coordinate conversion parameter, the image segmentation algorithm of utilization based on particle filter carries out iteration to three-D ultrasound data to be cut apart,
Described statistical shape model is by the manual segmentation result of multiple high definition 3 d image datas being trained to the mean value obtaining and the combination that characterizes a stack features vector that changes mode.
Further, the foundation of described statistical shape model comprises the following steps:
Gather the 3 d image data at multiple volunteers' designated organ position by high resolution imaging method, and manually cut apart 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 point set of setting quantity, described training set need meet: the point that has same index in training set is corresponding one by one;
Training point set is carried out to principal component analysis (PCA), the mean value point set u that obtains training set with and the feature space of covariance matrix;
The proper vector that forms described feature space is pressed to the large minispread of eigenwert, select in order multiple proper vectors that character pair value is large, make its character pair value sum reach the threshold value requirement of setting, described training set mean value point set u and the proper vector v being selected
icommon formation statistical shape model.
Further, the described three-D ultrasound data by statistical shape model and the designated organ that collects carries out thick registration, and the step that obtains initialization coordinate conversion parameter comprises:
In three-D ultrasound data, manually choose the point that a range estimation approaches organ of interest geometric center most;
Translation relation in statistical shape model between the geometric center of training set mean value and this point is as thick registration.
Further, described according to initialization coordinate conversion parameter, utilize image segmentation algorithm based on particle filter to carry out to three-D ultrasound data the step that iteration cuts apart and comprise:
Steps A, centered by initialization coordinate conversion parameter, in parameter space, choose multiple points, as particle;
Step B, to each particle, by the coordinate conversion parameter of its representative, mean value and proper vector to training point set are carried out coordinate transform, proper vector after conversion is multiplied by each self-corresponding weight, be added with the training points ensemble average value after conversion, obtain the prior model corresponding to each particle, in circulation for the first time, weight corresponding to each proper vector is 0, take prior model corresponding to each particle as 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 objective function, figure is cut apart, and obtain the surface point of segmentation result,
Step C, for the surface point of each particle, carry out registration with corresponding prior model respectively, and calculate the similarity of itself and corresponding prior model, by all particles according to sequencing of similarity and carry out resampling, select a best particle, corresponding best particle surface point and the difference of mean value in corresponding prior model are projected on the space that characteristic of correspondence vector forms, calculate the weight of each proper vector;
The rate of change between the best particle that before and after step D, calculating, twice circulation obtains, if be less than setting threshold, just think algorithm convergence, or circulation exceedes set point number, end loop, otherwise, the initialization changes in coordinates parameter in using best particle as steps A, each proper vector respective weights in using the weight that obtains in step C as step B, gets back to steps A and re-executes.
Another object of the present invention is to provide a kind of Ultrasound Image Segmentation system, described system comprises the acquiring ultrasound image system of three-dimensional ultrasound pattern of the designated organ that gathers patient and the image processing system that the three-dimensional ultrasound pattern collecting is processed, and described image processing system comprises:
Thick registration module, for statistical shape model and the three-D ultrasound data of the designated organ collecting are carried out to thick registration, obtains initialization coordinate conversion parameter;
Iteration is cut apart module, and for according to initialization coordinate conversion parameter, utilize image segmentation algorithm based on particle filter to carry out iteration to three-D ultrasound data and cuts apart,
Described statistical shape model is by the manual segmentation result of multiple high definition 3 d image datas being trained to the mean value obtaining and the combination that characterizes a stack features vector that changes mode.
Further, described Ultrasound Image Segmentation system also comprises that the 3-D view that gathers patient's designated organ is for setting up the computed tomography (SPECT) system of statistical shape model, described image processing system also comprises that statistical shape model sets up module, and described statistical shape model is set up module and comprised:
3-D view is manually cut apart editor module, for receive the multiple volunteers of professional person to computed tomography (SPECT) system collection designated organ position 3 d image data cut apart editor;
Gridding processing module, for for each segmentation result, carries out resurfacing, and gridding;
Training point set is chosen module, for each grid for after gridding, respectively chooses the summit composing training point set of setting quantity from its summit, and described training set need meet: the point that has same index in training set is corresponding one by one;
Training points set analysis module, for training point set is carried out to principal component analysis (PCA), the mean value point set u that obtains training set with and the feature space of covariance matrix;
Proper vector is chosen module, for the proper vector that forms described feature space is pressed to the large minispread of eigenwert, select in order multiple proper vectors that character pair value is large, make its character pair value sum reach the threshold value requirement of setting, described training set mean value point set u and the proper vector v being selected
icommon formation statistical shape model.
Further, described thick registration module also comprises:
Geometric center point is chosen load module, approaches the point of organ of interest geometric center for manually choosing a range estimation at three-D ultrasound data most;
Coordinate transform is related to computing module, for the translation relation between geometric center and this point of statistical shape model training set mean value as thick registration.
Further, described iteration is cut apart module and is comprised:
Particle is chosen module, for centered by initialization coordinate conversion parameter, chooses multiple points, as particle in parameter space;
Surface point acquisition module, be used for each particle, by the coordinate conversion parameter of its representative, mean value and proper vector to training point set are carried out coordinate transform, proper vector after conversion is multiplied by each self-corresponding weight, be added with the training points ensemble average value after conversion, obtain the prior model corresponding to each particle, in circulation for the first time, weight corresponding to each proper vector is 0, take prior model corresponding to each particle as 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 objective function, figure is cut apart, and obtain the surface point of segmentation result,
Similarity registration module, for the surface point for each particle, carry out registration with corresponding prior model respectively, and calculate the similarity of itself and corresponding prior model, by all particles according to sequencing of similarity and carry out resampling, select a best particle, corresponding best particle surface point and the difference of mean value in corresponding prior model are projected on the space that characteristic of correspondence vector forms, calculate the weight of each proper vector;
The best particle rate of change is calculated judge module, the rate of change between the best particle obtaining for twice circulation before and after calculating, if be less than setting threshold, just think algorithm convergence, or circulation exceedes set point number, end loop, otherwise, choose the initialization changes in coordinates parameter in module using best particle as particle, the each proper vector respective weights in using the weight that obtains in similarity registration module as surface point acquisition module, gets back to particle and chooses module and re-execute.
The present invention adopts statistical shape model and the image based on class particle filter pattern to cut apart the alternative manner combining the appointment internal organs in ultrasonoscopy are cut apart.Using statistical model as priori, and the combination of gradation of image information, the form of employing particle filter, carries out iteration to organ and cuts apart.In iteration cutting procedure, make full use of the information that statistical model provides, not high and have organ of interest to have the effect that still can obtain being comparatively satisfied with in ill-defined situation in image definition.Adopt the form of particle filter, effectively avoid local minimum problem.Like this, with regard to having avoided manually cutting apart and semi-automaticly having cut apart that needs are artificial participates in more problem, compare existing full-automatic partition method, the present invention solves the accuracy problem of cutting apart under the low and image blurring state of image resolution ratio.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the ultrasonic image division method that provides of the embodiment of the present invention;
Fig. 2 is the method for building up process flow diagram of the statistical shape model that provides of the embodiment of the present invention;
Fig. 3 is the implementing procedure figure of the Image Iterative dividing method that provides of the embodiment of the present invention;
Fig. 4 is the structural representation of the Ultrasound Image Segmentation system that provides of the embodiment of the present invention.
Embodiment
In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, 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, statistical shape model and the three-D ultrasound data of the designated organ collecting are carried out to thick registration, obtain initialization coordinate conversion parameter.
In the present invention, this statistical shape model is by the manual segmentation result of multiple high definition 3 d image datas being trained to the mean value obtaining and the combination that characterizes a stack features vector that changes mode.
As embodiments of the invention, this step comprises again: the point of manually choosing a range estimation and approach most organ of interest geometric center in three-D ultrasound data; And in statistical shape model the translation relation between geometric center and this point of training set mean value as thick registration.
In this preferred version, adopt following scheme: the point of manually choosing a range estimation and approach most organ of interest geometric center in three-D ultrasound data.Translation relation in statistical shape model between the geometric center of training set mean value and this point is as thick registration.Coordinate conversion parameter space at this definition coordinate system A to coordinate system B, adopts sextuple parameter space in this preferred version, be respectively around X-axis anglec of rotation α
x, around Y-axis anglec of rotation α
y, around Z axis anglec of rotation α
z, along X-axis translation T
x, along Y-axis translation T
y, along Z axis translation T
z.Its expression: coordinate system A is pressed around X-axis rotation alpha
x, then around postrotational Y-axis rotation alpha
y, then according to postrotational Z axis rotation alpha
z, finally in postrotational coordinate system respectively along X-axis translation T
x, along Y-axis translation T
y, along Z axis translation T
zoverlap with coordinate system B afterwards.
In thick registration process, can choose multiple clear and monumented points that anatomical position is clear and definite, adopt method for registering between point set to carry out thick registration as methods such as ICP.
In step S102, according to initialization coordinate conversion parameter, the image segmentation algorithm of utilization based on particle filter carries out iteration to three-D ultrasound data to be cut apart.
Fig. 2 shows the method for building up flow process of the statistical shape model that the embodiment of the present invention provides, and details are as follows:
In step S201, gather the 3 d image data at multiple volunteers' designated organ position by high resolution imaging method, and manually cut apart 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 has adopted triangle gridding, certainly can adopt the grid of other any suitable shapes.
In step S203, for each grid after gridding, from its summit, respectively choose the summit composing training point set of setting quantity, described training set need meet: the point that has same index in training set is corresponding one by one.
In step S204, training point set is carried out to principal component analysis (PCA) (Principle ComponentAnalysis, PCA), the mean value point set u that obtains training set with and the feature space of covariance matrix.
In step S205, the proper vector that forms described feature space is pressed to the large minispread of eigenwert, select in order multiple proper vectors that character pair value is large, make its character pair value sum reach the threshold value requirement of setting, described training set mean value point set u and the proper vector v being selected
icommon formation statistical shape model.
Fig. 3 shows the implementing procedure of the Image Iterative dividing method that the embodiment of the present invention provides, and details are as follows:
In step S301, centered by initialization coordinate conversion parameter, in parameter space, choose multiple points, as particle.
In this preferred version, to each parameter, by the normal distribution of the given variance take initialization coordinate conversion parameter as average, choose at random particle point.In circulation for the first time, using thick registration result as initialization coordinate conversion parameter.
In step S302, to each particle, by the coordinate conversion parameter of its representative, mean value and proper vector to training point set are carried out coordinate transform, proper vector after conversion is multiplied by each self-corresponding weight, be added with the training points ensemble average value after conversion, obtain the prior model corresponding to each particle, in circulation for the first time, weight corresponding to each proper vector is 0, take prior model corresponding to each particle as 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 objective function, figure is cut apart, and obtain the surface point of segmentation result.
In this preferred version, the constituted mode of figure is as follows: each pixel is as a node of figure, square being inversely proportional to of the channel capacity between adjacent two nodes and the difference of two node gray-scale values.
In step S303, for the surface point of each particle, carry out registration with corresponding prior model respectively, and calculate the similarity of itself and corresponding prior model, by all particles according to sequencing of similarity and carry out resampling, select a best particle, corresponding best particle surface point and the difference of mean value in corresponding prior model are projected on the space that characteristic of correspondence vector forms, calculate the weight of each proper vector.
In this preferred version, registration approach adopts ICP algorithm.In this preferred version, similarity is defined as follows: the each some Pi that effects on surface point is concentrated, find its closest approach Qi in prior model, and calculate the L2 distance L of point-to-point transmission
i 2=(p
i-q
i)
2.Similarity is defined as p=∑ 1/L
i 2, to a have Pi.
In step S304, the rate of change between the best particle that before and after calculating, twice circulation obtains, if be less than setting threshold, just think algorithm convergence, or circulation exceedes set point number, end loop, otherwise, initialization changes in coordinates parameter in using best particle as step S301, the each proper vector respective weights in using the weight that obtains in step S303 as step S302, gets back to step S301 and re-executes.
Fig. 4 shows the structure of the Ultrasound Image Segmentation system that the embodiment of the present invention provides, this Ultrasound Image Segmentation system comprises the acquiring ultrasound image system 1 of three-dimensional ultrasound pattern of the designated organ that gathers patient and the image processing system 2 that the three-dimensional ultrasound pattern collecting is processed, and described image processing system 2 comprises: thick registration module 21 and iteration are cut apart module 22.
In order to coordinate artificial input, this Ultrasound Image Segmentation system also comprises human-computer interaction module (not shown).For the image result of clear demonstration after cutting apart, this Ultrasound Image Segmentation system also comprises display device (not shown), and this display device can show the image result after cutting apart.
Statistical shape model and the three-D ultrasound data of the designated organ collecting are carried out thick registration by this thick registration module 21, obtains initialization coordinate conversion parameter; Iteration is cut apart module 22 according to initialization coordinate conversion parameter, the image segmentation algorithm of utilization based on particle filter carries out iteration to three-D ultrasound data to be cut apart, and described statistical shape model is by the manual segmentation result of multiple high definition 3 d image datas being trained to the mean value obtaining and the combination that characterizes a stack features vector that changes mode.
As embodiments of the invention, described Ultrasound Image Segmentation system also comprises that the 3-D view that gathers patient's designated organ is for setting up the computed tomography (SPECT) system 3 of statistical shape model, described image processing system also comprises that statistical shape model sets up module 20, and described statistical shape model is set up module 20 and comprised: 3-D view is manually cut apart editor module 201, gridding processing module 202, training point set and chosen module 203, training points set analysis module 204 and proper vector and choose module 205.
This 3-D view manually cut apart editor module 201 receive the multiple volunteers of professional person to computed tomography (SPECT) system collection designated organ position 3 d image data cut apart editor; Gridding processing module 202, for for each segmentation result, is carried out resurfacing, and gridding; Training point set is chosen module 203 for each grid for after gridding, respectively chooses the summit composing training point set of setting quantity from its summit, and described training set need meet: the point that has same index in training set is corresponding one by one; Training points set analysis module 204 is for training point set is carried out to principal component analysis (PCA), the mean value point set u that obtains training set with and the feature space of covariance matrix; Proper vector is chosen module 205 for forming the proper vector of described feature space by the large minispread of eigenwert, select in order multiple proper vectors that character pair value is large, make its character pair value sum reach the threshold value requirement of setting, described training set mean value point set u and the proper vector v being selected
icommon formation statistical shape model.
As embodiments of the invention, described thick registration module 21 also comprises: geometric center point chooses load module 211 and coordinate transform is related to computing module 212.This geometric center point is chosen load module 211 and approaches most for manually choosing a range estimation at three-D ultrasound data the point of organ of interest geometric center; Coordinate transform be related to computing module 212 for the translation relation between the geometric center of statistical shape model training set mean value and this point as thick registration.
As embodiments of the invention, this iteration is cut apart module 22 and is comprised: particle is chosen module 221, surface point acquisition module 222, similarity registration module 223 and the best particle rate of change and calculated judge module 224.
Particle is chosen module 221 for centered by initialization coordinate conversion parameter, chooses multiple points, as particle in parameter space, surface point acquisition module 222 is for to each particle, by the coordinate conversion parameter of its representative, mean value and proper vector to training point set are carried out coordinate transform, proper vector after conversion is multiplied by each self-corresponding weight, be added with the training points ensemble average value after conversion, obtain the prior model corresponding to each particle, in circulation for the first time, weight corresponding to each proper vector is 0, take prior model corresponding to each particle as 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 objective function, figure is cut apart, and obtain the surface point of segmentation result, similarity registration module 223 is for the surface point for each particle, carry out registration with corresponding prior model respectively, and calculate the similarity of itself and corresponding prior model, by all particles according to sequencing of similarity and carry out resampling, select a best particle, corresponding best particle surface point and the difference of mean value in corresponding prior model are projected on the space that characteristic of correspondence vector forms, calculate the weight of each proper vector, the rate of change between the best particle that best particle rate of change calculating judge module 224 obtains for twice circulation before and after calculating, if be less than setting threshold, just think algorithm convergence, or circulation exceedes set point number, end loop, otherwise, the initialization changes in coordinates parameter in module chosen using best particle as particle, each proper vector respective weights in using the weight that obtains in similarity registration module as surface point acquisition module, gets back to particle and chooses module and re-execute.
For the computed tomography (SPECT) system 3 in the present invention, what this preferred version adopted is nuclear magnetic resonance (MRI) imaging, also can adopt other suitable high definition high resolving power computed tomography (SPECT) systems such as CT; Ultrasonic probe in the present invention can be two-dimensional ultrasound probe, can be also three dimensional ultrasound probe (being volume imaging probe), all can use as long as can obtain the probe of three-D ultrasonic volume data.
In sum, dividing method provided by the invention will first be set up statistical shape model.Certainly, also ready-made statistical shape model directly can be imported to Ultrasound Image Segmentation system.So just avoid every equipment all to need to set up statistical shape model.The embodiment of the present invention is divided into two stages.The process of setting up statistical shape model is: by professional person, the high-resolution scan-data of a large amount of volunteer's samples is manually cut apart.Segmentation result is carried out to resurfacing gridding.In each grid, select the summit composing training point set of some.Training point set must meet following condition: point is concentrated the some correspondence one by one that has same index.Training point set is carried out to PCA(Principle Component Analysis) analyze, obtain training the mean value of point set and the feature space of covariance matrix.From the vector in constitutive characteristic space, choose series of features vector by the sequence of character pair value size, make its eigenwert sum be greater than certain threshold value.The mean value of training point set and the proper vector being selected form statistical shape model.
Undertaken by following flow process for image partition method: one, initialization.Determine the rough coordinates transformation parameter between statistical shape model and three-dimensional ultrasound pattern by certain criterion, and defined parameters space.Two, the image based on class particle filter pattern is cut apart iteration partitioning algorithm.Carry out as follows: 1, centered by initialization coordinate conversion parameter, in parameter space, choose the point of some by certain standard, as particle.In circulation for the first time, take the coordinate conversion parameter that obtains in step 1 as initialization coordinate conversion parameter.2,, to each particle, by the coordinate conversion parameter of its representative, mean value and the proper vector of training point set are carried out to coordinate transform.Proper vector after conversion is multiplied by each self-corresponding weight, is added with the training points ensemble average value after conversion, obtain the prior model corresponding to each particle.In circulation for the first time, weight corresponding to each proper vector is 0.Take prior model corresponding to each particle as seed point set, respectively according to the gray value information of three-dimensional ultrasound pattern, set up figure with certain principle, application graph cut(image is cut apart) algorithm cuts apart, and obtains the surface point of segmentation result.3, to the surface point corresponding to each particle obtaining in 2, carry out registration with corresponding prior model respectively, and calculate the similarity of itself and corresponding prior model according to certain definition.All particles according to sequencing of similarity and carry out resampling, are selected to a best particle.Corresponding best particle surface point and the difference of corresponding prior model are projected on the space that characteristic of correspondence vector forms, calculate the weight of each proper vector.4, the rate of change before and after calculating between the best particle that obtains of twice circulation, if be less than certain threshold value, just think algorithm convergence, or circulation exceedes some, end loop.Otherwise, using best particle as the initialization changes in coordinates parameter in 1, using the weight that obtains in 3 as the each proper vector respective weights in 2, get back to 1 and re-execute.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any modifications of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.
Claims (8)
1. a ultrasonic image division method, is characterized in that, said method comprising the steps of:
Statistical shape model and the three-D ultrasound data of the designated organ collecting are carried out to thick registration, obtain initialization coordinate conversion parameter;
According to initialization coordinate conversion parameter, the image segmentation algorithm of utilization based on particle filter carries out iteration to three-D ultrasound data to be cut apart,
Described statistical shape model is by the manual segmentation result of multiple high definition 3 d image datas being trained to the mean value obtaining and the combination that characterizes a stack features vector that changes mode.
2. ultrasonic image division method according to claim 1, is characterized in that, the foundation of described statistical shape model comprises the following steps:
Gather the 3 d image data at multiple volunteers' designated organ position by high resolution imaging method, and manually cut apart 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 point set of setting quantity, described training set need meet: the point that has same index in training set is corresponding one by one;
Training point set is carried out to principal component analysis (PCA), the mean value point set u that obtains training set with and the feature space of covariance matrix;
The proper vector that forms described feature space by the large minispread of eigenwert, is selected to multiple proper vectors that eigenwert is large in order, make its character pair value sum reach the threshold value requirement of setting, described training set mean value point set u and the proper vector v being selected
icommon formation statistical shape model.
3. ultrasonic image division method according to claim 1, is characterized in that, the described three-D ultrasound data by statistical shape model and the designated organ that collects carries out thick registration, and the step that obtains initialization coordinate conversion parameter comprises:
In three-D ultrasound data, manually choose the point that a range estimation approaches organ of interest geometric center most;
Translation relation in statistical shape model between the geometric center of training set mean value and this point is as thick registration.
4. ultrasonic image division method according to claim 1, is characterized in that, described according to initialization coordinate conversion parameter, utilizes image segmentation algorithm based on particle filter to carry out to three-D ultrasound data the step that iteration cuts apart and comprises:
Steps A, centered by initialization coordinate conversion parameter, in parameter space, choose multiple points, as particle;
Step B, to each particle, by the coordinate conversion parameter of its representative, mean value and proper vector to training point set are carried out coordinate transform, proper vector after conversion is multiplied by each self-corresponding weight, be added with the training points ensemble average value after conversion, obtain the prior model corresponding to each particle, in circulation for the first time, weight corresponding to each proper vector is 0; Take prior model corresponding to each particle as 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 objective function, cuts apart figure, and obtains the surface point of segmentation result;
Step C, for the surface point of each particle, carry out registration with corresponding prior model respectively, and calculate the similarity of itself and corresponding prior model, by all particles according to sequencing of similarity and carry out resampling, select a best particle, corresponding best particle surface point and the difference of mean value in corresponding prior model are projected on the space that characteristic of correspondence vector forms, calculate the weight of each proper vector;
The rate of change between the best particle that before and after step D, calculating, twice circulation obtains, if be less than setting threshold, just think algorithm convergence, or circulation exceedes set point number, end loop, otherwise, the initialization changes in coordinates parameter in using best particle as steps A, each proper vector respective weights in using the weight that obtains in step C as step B, gets back to steps A and re-executes.
5. a Ultrasound Image Segmentation system, it is characterized in that, described system comprises the acquiring ultrasound image system of three-dimensional ultrasound pattern of the designated organ that gathers patient and the image processing system that the three-dimensional ultrasound pattern collecting is processed, and described image processing system comprises:
Thick registration module, for statistical shape model and the three-D ultrasound data of the designated organ collecting are carried out to thick registration, obtains initialization coordinate conversion parameter;
Iteration is cut apart module, and for according to initialization coordinate conversion parameter, utilize image segmentation algorithm based on particle filter to carry out iteration to three-D ultrasound data and cuts apart,
Described statistical shape model is by the manual segmentation result of multiple high definition 3 d image datas being trained to the mean value obtaining and the combination that characterizes a stack features vector that changes mode.
6. Ultrasound Image Segmentation system according to claim 5, it is characterized in that, described Ultrasound Image Segmentation system also comprises that the 3-D view that gathers patient's designated organ is for setting up the computed tomography (SPECT) system of statistical shape model, described image processing system also comprises that statistical shape model sets up module, and described statistical shape model is set up module and comprised:
3-D view is manually cut apart editor module, for receive the multiple volunteers of professional person to computed tomography (SPECT) system collection designated organ position 3 d image data cut apart editor;
Gridding processing module, for for each segmentation result, carries out resurfacing, and gridding;
Training point set is chosen module, for each grid for after gridding, respectively chooses the summit composing training point set of setting quantity from its summit, and described training set need meet: the point that has same index in training set is corresponding one by one;
Training points set analysis module, for training point set is carried out to principal component analysis (PCA), the mean value point set u that obtains training set with and the feature space of covariance matrix;
Proper vector is chosen module, for the proper vector that forms described feature space is pressed to the large minispread of eigenwert, select in order multiple proper vectors that character pair value is large, make its character pair value sum reach the threshold value requirement of setting, described training set mean value point set u and the proper vector v being selected
icommon formation statistical shape model.
7. Ultrasound Image Segmentation system according to claim 5, is characterized in that, described thick registration module also comprises:
Geometric center point is chosen load module, approaches the point of organ of interest geometric center for manually choosing a range estimation at three-D ultrasound data most;
Coordinate transform is related to computing module, for the translation relation between geometric center and this point of statistical shape model training set mean value as thick registration.
8. Ultrasound Image Segmentation system according to claim 5, is characterized in that, described iteration is cut apart module and comprised:
Particle is chosen module, for centered by initialization coordinate conversion parameter, chooses multiple points, as particle in parameter space;
Surface point acquisition module, be used for each particle, by the coordinate conversion parameter of its representative, mean value and proper vector to training point set are carried out coordinate transform, proper vector after conversion is multiplied by each self-corresponding weight, is added with the training points ensemble average value after conversion, obtain the prior model corresponding to each particle, in circulation for the first time, weight corresponding to each proper vector is 0; Take prior model corresponding to each particle as 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 objective function, cuts apart figure, and obtains the surface point of segmentation result;
Similarity registration module, for the surface point for each particle, carry out registration with corresponding prior model respectively, and calculate the similarity of itself and corresponding prior model, by all particles according to sequencing of similarity and carry out resampling, select a best particle, corresponding best particle surface point and the difference of mean value in corresponding prior model are projected on the space that characteristic of correspondence vector forms, calculate the weight of each proper vector;
The best particle rate of change is calculated judge module, the rate of change between the best particle obtaining for twice circulation before and after calculating, if be less than setting threshold, just think algorithm convergence, or circulation exceedes set point number, end loop, otherwise, choose the initialization changes in coordinates parameter in module using best particle as particle, the each proper vector respective weights in using the weight that obtains in similarity registration module as surface point acquisition module, gets back to particle and chooses module and re-execute.
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 true CN103903255A (en) | 2014-07-02 |
| CN103903255B 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 (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107766874A (en) * | 2017-09-07 | 2018-03-06 | 沈燕红 | A kind of measuring method and measuring system of ultrasound volume biological parameter |
| CN109472791A (en) * | 2018-10-31 | 2019-03-15 | 深圳大学 | Ultrasonic image division method and computer equipment |
| 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 |
| CN109934861A (en) * | 2019-01-22 | 2019-06-25 | 广东工业大学 | A kind of incidence multi-modality medical image autoegistration 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 |
|---|
| 曹会志等: "基于配准与ASM方法的医学图像分割", 《中国医学影像技术》 * |
| 李春丽等: "基于统计信息及个体信息的统计形状模型的肝脏分割", 《南方医科大学学报》 * |
| 李选富: "散乱点云自动配准技术研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (9)
| 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 |
| CN107766874A (en) * | 2017-09-07 | 2018-03-06 | 沈燕红 | A kind of measuring method and measuring system of ultrasound volume biological parameter |
| 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 |
| CN109472791A (en) * | 2018-10-31 | 2019-03-15 | 深圳大学 | Ultrasonic image division method and computer equipment |
| CN109472791B (en) * | 2018-10-31 | 2020-10-27 | 深圳大学 | Ultrasonic image segmentation method and computer equipment |
| CN109934861A (en) * | 2019-01-22 | 2019-06-25 | 广东工业大学 | A kind of incidence multi-modality medical image autoegistration method |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103903255B (en) | 2016-08-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20190355117A1 (en) | Techniques for Segmentation of Lymph Nodes, Lung Lesions and Other Solid or Part-Solid Objects | |
| CN108038862B (en) | Interactive medical image intelligent segmentation modeling method | |
| CN102525662B (en) | Three-dimensional visual tissue organ operation navigation system | |
| US8358819B2 (en) | System and methods for image segmentation in N-dimensional space | |
| US6362821B1 (en) | Surface model generation for visualizing three-dimensional objects using multiple elastic surface nets | |
| JP2008511366A (en) | Feature-weighted medical object contour detection using distance coordinates | |
| CN104424647A (en) | Method and apparatus for registering medical images | |
| CN103903255A (en) | Ultrasound image segmentation method and system | |
| CN107067398A (en) | Complementing method and device for lacking blood vessel in 3 D medical model | |
| CN104978730A (en) | Division method and device of left ventricular myocardium | |
| WO2021018101A1 (en) | Data processing method, apparatus, device, and storage medium | |
| CN103229210A (en) | Image registration apparatus | |
| CN104408398A (en) | Liver boundary identification method and system | |
| CN107708550A (en) | For the surface modeling for the segmentation acoustic echo structure for detecting and measuring anatomic abnormalities | |
| CN104240287A (en) | Method and system for generating coronary panorama through CT images | |
| CN103295234A (en) | Medical image segmentation system and medical image segmentation method based on deformation surface models | |
| CN106133789A (en) | For splitting image processing apparatus and the method for area-of-interest | |
| CN105389821A (en) | Medical image segmentation method based on combination of cloud module and image segmentation | |
| CN103345774B (en) | A kind of modeling method of three-dimensional multi-scale vector quantization | |
| CN108597589B (en) | Model generation method, target detection method and medical imaging system | |
| Marker et al. | Contour-Based Surface Reconstruction using Implicit Curve Fitting, and Distance Field Filtering and Interpolation. | |
| Cameron et al. | Patient-Specific Anatomic Models-Geometric Surface Generation from Three-Dimensional Medical Images Using a Specified Polygonal Budget | |
| Jing et al. | A novel 3D reconstruction algorithm of motion-blurred CT image | |
| CN111369662A (en) | Three-dimensional model reconstruction method and system for blood vessels in CT (computed tomography) image | |
| DE102012203122B4 (en) | Method and system for determining a boundary surface mesh |
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 |