A kind of vessel extraction method, apparatus and computer readable storage medium
Technical field
This application involves image processing techniques more particularly to a kind of vessel extraction method, apparatus and computer-readable storages
Medium.
Background technique
With the continuous development of computer technology, modern medical service image technology achieves many innovations in medical diagnosis neighborhood
It is widely used in clinical diagnosis with breakthrough.Traditional diagnostic imaging is analyzed by the healthcare givers of profession mostly, though
So artificial operation has a high-precision advantage, but the demand of increasingly huge image data and diagnosis, so that medical people
The working strength of member constantly aggravates, so that work quality be caused to reduce.So relative to traditional diagnostic imaging mode, based on
The auxiliary diagnosis of calculation machine can be greatly reduced the working strength of clinician, improve accuracy rate of diagnosis, avoid mistaken diagnosis, fail to pinpoint a disease in diagnosis phenomenon
Occur.Such as: CT angiography (CT Angiography, CTA) and nuclear magnetic resonance image (Nuclear Magnetic
Resonance Imaging, MRI) angiographic image can be provided, be conducive to the screening and diagnosis of vascular diseases;Four-dimensional blood
It flows nuclear magnetic resonance image (Four-Dimensional Flow Magnetic Resonance Imaging, 4D flow MRI)
The blood flow velocity field changed over time can be extracted in 3 dimension spaces, the visualization technique of blood flow velocity is conducive to vascular diseases
Screening and diagnosis.Therefore particularly important is become for automatic, the accurate Segmentation Research of medical image medium vessels tissue,
It can greatly mitigate the work of image department doctor, provide working efficiency and work quality.
Traditional blood-vessel image partitioning algorithm is applied to have been achieved for good effect in medical image processing.Based on threshold
The dividing method of value, if the histogram of image can be divided into Ganlei with one or several gray values, institute in original image
There is the pixel for the gray value being in the same gray level to be attributed to same class, to achieve the effect that image segmentation.But this
The superiority and inferiority of process is often depending on the gray threshold of image itself, due to having a long way to go between same group of data of medical imaging, often
There is many fluctuations (noise), and gray level method does not account for the spatial positional information of pixel, so very quick for noise
Sense.The dividing method of region growing also has good progress on medical image, it is a series of given seed points in the picture,
The pixel combination of similar quality is constituted into region.However the selection of final segmentation result and seed point has much relations, by
The instable influence of data, the selection of seed point also become more difficult, and institute is also very sensitive to noise in this way, by noise
Influence be likely to form discontinuity zone.Due to active contour model can also be obtained in the case where strong noise it is continuous, smooth
It is closed partitioning boundary, the medical image cutting techniques based on active contour model are also concerned by people more and more.It is main
The basic thought of dynamic skeleton pattern is one initial curve of setting first, then make initial curve under the action of the image field of force not
It is disconnected to shrink, finally it is retracted to object edge.Although antimierophonic interference, this mistake can be supported in active profile cutting procedure
The superiority and inferiority of journey depends on the position of initial curve, and the presence additionally, due to noise becomes the determination of suitable initial curve more
Add difficulty.
Conventional method achieves good segmentation effect in a certain range, but its common ground be segmentation effect seriously according to
Rely in Heuristics, need a large amount of manual intervention, and very sensitive for noise, divides inefficient.Therefore, this kind of
Method is difficult to accomplish in medical image segmentation really automatic, and is merely able to allow the healthcare givers of not no relevant professional knowledge to go
It uses, it is difficult to really put into clinical application.
Summary of the invention
In order to solve the above technical problems, the embodiment of the present application is intended to provide the method, apparatus and storage of a kind of vessel extraction
Medium can weaken noise jamming, improve the accuracy of vessel extraction.
The technical solution of the application is achieved in that
In a first aspect, providing a kind of vessel extraction method, which comprises
Obtaining the N layer sectioning image of target blood, wherein each layer of sectioning image includes the speed field picture in M direction,
N and M take positive integer;
Based on Image Pretreatment Algorithm, each layer of sectioning image in the N layers of sectioning image is pre-processed, is obtained
First kind N tomographic image;Wherein, described image Preprocessing Algorithm includes at least one neighborhood processing algorithm;
Based on preset image segmentation algorithm, it is partitioned into from each tomographic image in the first kind N tomographic image and includes
The target area of the target blood obtains the second class N tomographic image;
Based on the image sequence of the N layers of sectioning image, the second class N tomographic image is merged, the mesh is obtained
Mark the initial three-dimensional point cloud model of blood vessel.
Second aspect, provides a kind of vessel extraction device, and described device includes:
Acquisition unit, for obtaining the N layer sectioning image of target blood, wherein each layer of sectioning image includes M direction
Speed field picture, N and M take positive integer;
Processing unit, for be based on Image Pretreatment Algorithm, by each layer of sectioning image in the N layers of sectioning image into
Row pretreatment, obtains first kind N tomographic image;Wherein, described image Preprocessing Algorithm includes at least one neighborhood processing algorithm;
Processing unit is also used to based on preset image segmentation algorithm, from each layer of figure in the first kind N tomographic image
It is partitioned into the target area comprising the target blood as in, obtains the second class N tomographic image;
Reconfiguration unit closes the second class N tomographic image for the image sequence based on the N layers of sectioning image
And obtain the initial three-dimensional point cloud model of the target blood.
The third aspect, provides a kind of vessel extraction device, described device include processor and be configured to storage can be
The memory of the computer program run on processor,
Wherein, when the processor is configured to run the computer program, execute preceding method the step of.
Fourth aspect provides a kind of computer storage medium, is stored thereon with computer program, wherein the computer
The step of preceding method is realized when program is executed by processor.
By adopting the above technical scheme, it is carried out using N layer sectioning image of at least one neighborhood processing algorithm to target blood
Neighborhood processing can effectively inhibit the interference of noise in image, and target blood is accurately positioned, and guarantee subsequent image cutting operation
Accuracy guarantees the quality of target blood three-dimensionalreconstruction image to improve the extraction accuracy of target blood three-dimensional point cloud model.
Detailed description of the invention
Fig. 1 is the flow diagram of the embodiment of the present application medium vessels extracting method;
Fig. 2A to Fig. 2 C is the speed field picture schematic diagram in three directions in the embodiment of the present application;
Fig. 3 A to Fig. 3 B is the contrast schematic diagram of neighborhood variance method before and after the processing in the embodiment of the present application;
Fig. 4 A to Fig. 4 C is the positive and negative judging result schematic diagram of velocity field Image neighborhood in three directions in the embodiment of the present application;
Fig. 5 A to Fig. 5 C is that the positive negative judgement weighted results of velocity field Image neighborhood in three directions in the embodiment of the present application are shown
It is intended to;
Fig. 6 A to Fig. 6 C is the knot after two kinds of neighborhood processings of speed field picture in three directions in the embodiment of the present application merge
Fruit schematic diagram;
Fig. 7 A to Fig. 7 C is after the speed field picture in three directions in the embodiment of the present application fills up pit-hole and median filtering
Result schematic diagram;
Fig. 8 is the result schematic diagram in the embodiment of the present application after three direction speed field pictures merging;
Fig. 9 is the result schematic diagram in the embodiment of the present application after image segmentation;
Figure 10 is the result schematic diagram in the embodiment of the present application after morphological image corrosion;
Figure 11 is initial three-dimensional point cloud model schematic diagram in the embodiment of the present application;
Figure 12 is the result schematic diagram in the embodiment of the present application after morphological dilations;
Figure 13 is final three-dimensional point cloud model schematic diagram in the embodiment of the present application;
Figure 14 is the three-dimensional point cloud model schematic diagram in region to be reinforced in the embodiment of the present application;
Figure 15 is enhanced three-dimensional point cloud model schematic diagram in the embodiment of the present application;
Figure 16 is the three-dimensionalreconstruction model schematic of target blood in the embodiment of the present application;
Figure 17 is the composed structure schematic diagram of the embodiment of the present application medium vessels extraction element.
Specific embodiment
The characteristics of in order to more fully hereinafter understand the embodiment of the present application and technology contents, with reference to the accompanying drawing to this Shen
Please the realization of embodiment be described in detail, appended attached drawing purposes of discussion only for reference is not used to limit the embodiment of the present application.
Embodiment one
As shown in Figure 1, vessel extraction method specifically includes:
Step 101: obtaining the N layer sectioning image of target blood, wherein each layer of sectioning image includes the speed in M direction
Field picture is spent, N and M take positive integer;
Step 102: it is based on Image Pretreatment Algorithm, each layer of sectioning image in N layers of sectioning image is pre-processed,
Obtain first kind N tomographic image;Wherein, Image Pretreatment Algorithm includes at least one neighborhood processing algorithm;
Step 103: being based on preset image segmentation algorithm, be partitioned into from each tomographic image in first kind N tomographic image
Target area comprising target blood obtains the second class N tomographic image;
Step 104: the second class N tomographic image is merged, obtains target blood by the image sequence based on N layers of sectioning image
The initial three-dimensional point cloud model of pipe.
Here, the executing subject of step 101 to step 104 can be the processor of vessel extraction device.
Here, target blood can be all kinds of blood vessels, and blood vessel includes main blood vessel and capilary, and main blood vessel can be arterial blood
The blood vessels such as pipe, vein blood vessel or above-mentioned any combination.Capilary includes main vessel branch blood vessel and capillary.
In practical application, N layers of sectioning image can be the N layer two-dimensional slice image of target blood, pass through slice scanner
Obtained from digital picture where target blood is scanned into computer.Scanning technique has: MRI, four-dimensional blood flow nuclear-magnetism are total
Vibration image (Four-Dimensional Flow Magnetic Resonance Imaging, 4D flow MRI), CTA, magnetic are total
Shake angiography (Magnetic Resonance Angiography, MRA), computed tomography (Computed
Tomography, CT) etc. one or more combinations.
Here, Image Pretreatment Algorithm may include at least one neighborhood processing algorithm;Neighborhood processing algorithm has neighborhood side
Poor method, neighboring region energy method, the positive negative judgement of neighborhood etc..For example, step 102 can specifically include: it is based on the first neighborhood processing algorithm,
The speed field picture in each of N layers of sectioning image direction is subjected to neighborhood processing, obtains N number of template image;Based on second
The speed field picture in each of N layers of sectioning image direction is carried out neighborhood processing, obtains third class N by neighborhood processing algorithm
Tomographic image;By i-th of template image in N number of template image respectively with M direction of the i-th tomographic image in third class N tomographic image
Speed field picture carries out point multiplication operation, obtains first kind N tomographic image, i takes the positive integer less than or equal to N;Wherein, first is adjacent
Domain Processing Algorithm is different from the second neighborhood processing algorithm.
Specifically, the method for obtaining N number of template image include: based on the first neighborhood processing algorithm, will be in N layers of sectioning image
Each direction speed field picture carry out neighborhood processing, the image after obtaining N × M neighborhood processing, by N × M image
In the speed field picture in M direction in each tomographic image merge, obtain N number of image;It is true based on maximum variance between clusters
The optimal threshold of fixed N number of image;Each of N number of image image segmentation will be made an uproar at two parts region based on optimal threshold
Sound area pixel is set to 0, other area pixels are set to 1, obtain N number of binary image, this N number of binary image is as N number of
Template image.
Here, the first neighborhood processing algorithm can be neighborhood variance method, enable to the noise region pixel value in image
It is amplified, angiosomes pixel value, which is reduced, to be achieved the purpose that distinguish noise region and angiosomes, to improve subsequent image
The effect of binary conversion treatment.Second neighborhood processing algorithm can be the positive negative judgement of neighborhood, can effectively make the noise in image
Regional luminance reduces, that is, inhibits the noise region in image.Therefore, can be had using above two neighborhood combination treatment method
Effect inhibits noise in image, improves denoising effect.
In practical application, first kind N tomographic image is obtained, comprising: morphology filling behaviour is carried out to the image after point multiplication operation
Work and/or median filtering, obtain first kind N tomographic image.Padding can fill the pit-hole in image, and median filtering can disappear
Except partial noise remaining in image, subsequent image treatment effect is improved.
In the embodiment of the present application, each tomographic image in first kind N tomographic image includes the speed field picture in M direction, the
Each tomographic image in three classes N tomographic image also includes the speed field picture in M direction.
When practical application, step 103 is specifically included: by the speed in M direction of each tomographic image in first kind N tomographic image
Degree field picture is merged into a sub-picture, obtains N tomographic image to be split;N layer to be split is determined based on maximum variance between clusters
The optimal threshold of image;It is partitioned into from each tomographic image in N tomographic image to be split comprising target blood based on optimal threshold
The target area of pipe obtains the second class N tomographic image.Here, the target area of each tomographic image includes that target blood is cutd open in difference
Cut the vascular cross-section of position.
In the embodiment of the present application, each tomographic image in the second class N tomographic image only includes piece image.
Adaptive threshold fuzziness based on maximum between-cluster variance, specific implementation process:
The groups such as blood vessel and the liver of surrounding, pancreas, body fluid and fat, which are woven in intensity profile, to be had differences, and be can be used and is based on
The adaptive threshold fuzziness method of maximum between-cluster variance isolates blood vessel from image, and removes background parts.Algorithm steps are as follows:
1, it treats segmented image and is normalized and is calculated grey level histogram, i.e., with the picture of certain gray level
Plain number.Assuming that the total pixel of image be it is N number of, there is n altogether in the pixel that gray value is i, then the probability of occurrence of gray level i
For P (i)=n/N;
2, grey level histogram is divided into A by given threshold k, B two major classes: A:P (i) >=k and B:P (i) < k, calculates A, B class
Inter-class variance;
3, it allows k to be changed to 255 from 1, calculates separately the corresponding AB inter-class variance of k value each time, when inter-class variance maximum
Corresponding threshold value is optimal threshold K;
4, binary segmentation is carried out to image using K value, i.e. image grayscale >=K pixel is target area, remaining is background.
In practical application, step 104 is specifically included: it is based on preset Morphology Algorithm, it will be every in the second class N tomographic image
The target area of one tomographic image carries out morphological erosion processing, the second class N tomographic image after being corroded;Based on N layers of slice map
The second class N tomographic image after corrosion is merged, obtains the initial three-dimensional point cloud model of target blood by the image sequence of picture.
Here, preset Morphology Algorithm is erosion algorithm,
Here, the image sequence of N layers of sectioning image is the number of each layer of sectioning image after being sliced to image, example
Number such as from 1 to N, when merging to the second class image, the position of each tomographic image is depending on image sequence, image
Sequence consensus when sequencing and slice when merging, it is ensured that accurate three-dimensional point cloud model can be obtained after merging.
This method further include: surface reconstruction is carried out to target blood in initial three-dimensional point cloud model, is obtained smooth, closure
Three-dimension curved surface, the three-dimension curved surface are the best fit curved surfaces to three-dimensional point cloud model, while the entity that the three-dimension curved surface is included
It is exactly the objective blood vessel after reconstruct.
By adopting the above technical scheme, it is carried out using N layer sectioning image of at least one neighborhood processing algorithm to target blood
Neighborhood processing can effectively inhibit the interference of noise in image, and target blood is accurately positioned, and guarantee subsequent image cutting operation
Accuracy guarantees the quality of target blood three-dimensionalreconstruction image to improve the extraction accuracy of target blood three-dimensional point cloud model.
Embodiment two
In order to more embody the purpose of the application, on the basis of the embodiment of the present application one, further illustrated
Illustrate, which specifically includes:
Step 201: obtaining the N layer sectioning image of target blood, wherein each layer of sectioning image includes the speed in M direction
Field picture is spent, N and M take positive integer;
Step 202: the first neighborhood processing algorithm is based on, by the speed field picture in each of N layers of sectioning image direction
Neighborhood processing is carried out, N number of template image is obtained;
Specifically, the method for obtaining N number of template image include: based on the first neighborhood processing algorithm, will be in N layers of sectioning image
Each direction speed field picture carry out neighborhood processing, the image after obtaining N × M neighborhood processing, by N × M image
In the speed field picture in M direction in each tomographic image merge, obtain N number of image;It is true based on maximum variance between clusters
The optimal threshold of fixed N number of image;Each of N number of image image segmentation will be made an uproar at two parts region based on optimal threshold
Sound area pixel is set to 0, other area pixels are set to 1, obtain N number of binary image, this N number of binary image is as N number of
Template image.
Here, the first neighborhood processing algorithm can be neighborhood variance method, enable to the noise region pixel value in image
It is amplified, angiosomes pixel value, which is reduced, to be achieved the purpose that distinguish noise region and angiosomes, to improve subsequent image
The effect of binary conversion treatment.
Step 203: the second neighborhood processing algorithm is based on, by the speed field picture in each of N layers of sectioning image direction
Neighborhood processing is carried out, third class N tomographic image is obtained;
Here, the second neighborhood processing algorithm can be the positive negative judgement of neighborhood, can effectively make the noise region in image
Brightness reduces, that is, inhibits the noise region in image.Therefore, can effectively be pressed down using above two neighborhood combination treatment method
Imaged middle noise improves denoising effect.
Step 204: by the M with the i-th tomographic image in third class N tomographic image respectively of i-th of template image in N number of template image
The speed field picture in a direction carries out point multiplication operation, obtains first kind N tomographic image, i takes the positive integer less than or equal to N;
Here, point multiplication operation is that same position pixel value carries out product calculation between image.
In practical application, first kind N tomographic image is obtained, comprising: morphology filling behaviour is carried out to the image after point multiplication operation
Work and/or median filtering, obtain first kind N tomographic image.Padding can fill the pit-hole in image, and median filtering can disappear
Except partial noise remaining in image, subsequent image treatment effect is improved.
In the embodiment of the present application, each tomographic image in first kind N tomographic image includes the speed field picture in M direction, the
Each tomographic image in three classes N tomographic image also includes the speed field picture in M direction.
Step 205: being based on preset image segmentation algorithm, be partitioned into from each tomographic image in first kind N tomographic image
Target area comprising target blood obtains the second class N tomographic image;
Specifically, the speed field picture in M direction of each tomographic image in first kind N tomographic image is merged into a secondary figure
Picture obtains N tomographic image to be split;The optimal threshold of N tomographic image to be split is determined based on maximum variance between clusters;Based on most
Good threshold value is partitioned into the target area comprising target blood from each tomographic image in N tomographic image to be split, obtains second
Class N tomographic image.
In the embodiment of the present application, each tomographic image in the second class N tomographic image only includes piece image.
Step 206: the second class N tomographic image is merged, obtains target blood by the image sequence based on N layers of sectioning image
The initial three-dimensional point cloud model of pipe;
Here, the image sequence of N layers of sectioning image is the number of each layer of sectioning image after being sliced to image, example
Number such as from 1 to N, when merging to the second class image, the position of each tomographic image is depending on image sequence, image
Sequence consensus when sequencing and slice when merging, it is ensured that accurate three-dimensional point cloud model can be obtained after merging.
Step 207: initial three-dimensional point cloud model being sliced, N number of template image is obtained;
Specifically, initial three-dimensional point cloud model is sliced, N tomographic image is obtained;It, will based on preset Morphology Algorithm
Region in each tomographic image in N tomographic image comprising target blood carries out morphological dilations processing, obtains N number of template image.
Here, preset Morphology Algorithm is expansion algorithm.
Here, the N number of template image initial three-dimensional point cloud model being sliced, the N obtained compared to step 202
A template image quality is higher, that is to say, that in carrying out second time iterative process, is obtained using first pass iterative processing
Initial three-dimensional point cloud model generate template image and execute vessel extraction operations again, can be improved denoising effect and image point
Cut precision.That is, executing the available preferably initial three-dimensional of an image processing operations if blood vessel detection environment is preferable
Point cloud model, then without executing the later step of step 206;If it is complicated that blood vessel detects environment, in obtained initial three-dimensional point cloud
In model target blood adhesion its hetero-organization or organ, then need to carry out second iteration and execute step 206 to step 210,
To remove the influence of its hetero-organization or organ.
Step 208: by the M with the i-th tomographic image in third class N tomographic image respectively of i-th of template image in N number of template image
The speed field picture in a direction carries out point multiplication operation, obtains first kind N tomographic image;
Here, N number of template image obtained in step 202 is considered as the first template, N number of Prototype drawing obtained in step 207
As being considered as the second template.
That is, step 208 is that the first template in step 204 is replaced with to the second template, and third class N tomographic image,
And calculating process is all the same.
Step 209: being based on preset image segmentation algorithm, be partitioned into from each tomographic image in first kind N tomographic image
Target area comprising target blood obtains the second class N tomographic image;
Step 210: the second class N tomographic image is merged, obtains target blood by the image sequence based on N layers of sectioning image
The final three-dimensional point cloud model of pipe.
Here, the image sequence of N layers of sectioning image is the number of each layer of sectioning image after being sliced to image, example
Number such as from 1 to N, when merging to the second class image, the position of each tomographic image is depending on image sequence, image
Sequence consensus when sequencing and slice when merging, it is ensured that accurate three-dimensional point cloud model can be obtained after merging.
This method further include: surface reconstruction is carried out to target blood in final three-dimensional point cloud model, is obtained smooth, closure
Three-dimension curved surface, the three-dimension curved surface are the best fit curved surfaces to three-dimensional point cloud model, while the entity that the three-dimension curved surface is included
It is exactly the objective blood vessel after reconstruct.
Here, the executing subject of step 201 to step 210 can be the processor of vessel extraction device.
Embodiment three
In order to more embody the purpose of the application, on the basis of the embodiment of the present application one and embodiment two, carry out into
One step for example, the vessel extraction method specifically includes:
Step 301: obtaining the N layer sectioning image of target blood, wherein each layer of sectioning image includes the speed in M direction
Field picture is spent, N and M take positive integer;
Step 302: based on user's selection information, in the speed field picture for determining each of N layers of sectioning image direction
Region to be reinforced;It wherein, include at least partly blood vessel of target blood in region to be reinforced;
It, may using above-mentioned vessel extraction method if subregional feature is unobvious in the middle part of target blood in practical application
The unconspicuous region of these features can not be accurately extracted, therefore, it is necessary to carry out enhancing processing to these regions.
User can be selected by artificial frame and determine region to be reinforced, for example, being selected by user input unit frame to be reinforced
Region, user input unit can be touch-display unit, and the region for wanting enhancing is irised out in touch-display unit, is based on user
The region to be reinforced in the region i.e. each layer of sectioning image irised out.
Step 303: be based on the first neighborhood processing algorithm, to the region to be reinforced in the speed field picture in each direction into
Row neighborhood processing obtains first kind N tomographic image;
Here, enhancing region is treated when being handled, need to only use a kind of field Processing Algorithm, can be retained more viscous
Even part.Using the first neighborhood processing algorithm the noise region pixel value in image is amplified, angiosomes pixel value
It is reduced, achievees the purpose that distinguish noise region and angiosomes, retain close with target blood adhesion in region to be reinforced
Part.If other parts adhesion in region to be reinforced can be made using the second field Processing Algorithm and field merging treatment
Target blood is taken as noise remove and falls, thus can not complete extraction go out this parts of images.
Step 304: being based on preset image segmentation algorithm, be partitioned into from each tomographic image in first kind N tomographic image
Target area comprising target blood obtains the second class N tomographic image;
Specifically, the speed field picture in M direction of each tomographic image in first kind N tomographic image is merged into a secondary figure
Picture obtains N tomographic image to be split;The optimal threshold of N tomographic image to be split is determined based on maximum variance between clusters;Based on most
Good threshold value is partitioned into the target area comprising target blood from each tomographic image in N tomographic image to be split, obtains second
Class N tomographic image.
In the embodiment of the present application, each tomographic image in the second class N tomographic image only includes piece image.
Step 305: the second class N tomographic image is merged, is obtained to be reinforced by the image sequence based on N layers of sectioning image
The three-dimensional point cloud model in region.
Here, the image sequence of N layers of sectioning image is the number of each layer of sectioning image after being sliced to image, example
Number such as from 1 to N, when merging to the second class image, the position of each tomographic image is depending on image sequence, image
Sequence consensus when sequencing and slice when merging, it is ensured that accurate three-dimensional point cloud model can be obtained after merging.
In practical application, this method further include: by the three-dimensional point cloud model in region to be reinforced and initial three-dimensional point cloud model
It merges, obtains the three-dimensional point cloud model of enhanced target blood;Alternatively, by the three-dimensional point cloud model in region to be reinforced and
Final three-dimensional point cloud model merges, and obtains the three-dimensional point cloud model of enhanced target blood.
That is, by the three-dimensional point cloud model of enhanced target blood and the initial threedimensional model of target blood or finally
Three-dimensional point cloud model merges, and obtains the three-dimensional point cloud model of enhanced target blood.
This method further include: surface reconstruction is carried out to target blood in the three-dimensional point cloud model of enhanced target blood,
The three-dimension curved surface of smooth closure is obtained, which is the best fit curved surface to three-dimensional point cloud model, while the three-dimensional is bent
The entity that face is included is exactly the objective blood vessel after reconstruct.
Here, the executing subject of step 301 to step 305 can be the processor of vessel extraction device.
Example IV
In order to more embody the purpose of the application, on the basis of the embodiment of the present application one is to embodiment three, carry out into
The illustration of one step.The present embodiment is directed to the MRI image of aorta, can be realized the arterial vascular automatic identification of high-precision.
Sectioning image is torso model four-dimension blood flow nuclear magnetic resonance image (Four-Dimensional Flow Magnetic
Resonance Imaging, 4D flow MRI), N layers of sectioning image are a heart beat cycle (totally 28 of human heart region
Moment) in 64 sagittal sections faces serial NMR imaging image group, the resolution ratio of every piece image is r0×c0(this reality
Apply r in example0=512, c0=512) r, is added up to0×c0Open image.The target of the present embodiment was determined most from 28 moment
At the good moment, the arteries being accurately partitioned into the nuclear-magnetism image at the moment then will be on all 64 width images at the moment
Arteries extract, lay equal stress on and be configured to a high accuracy three-dimensional arteries.
The method of vessel extraction specifically includes following:
Step 1: two dimensional image processing
1.1 both two-dimensional image slices
By the speed field data in 4D flow MRI image, slicing treatment is done along side-looking direction, in three durection component speed
64 sectioning images are generated respectively in degree field picture, add up to 64 × 3 images.By three width velocity fields in each sectioning image
Image is denoted as U, V and W respectively, and the meaning of speed field picture is that the velocity vector of the particle of a certain position in space is broken down into
Under three directions in corresponding space coordinates, the velocity magnitude of the particle on three directions of coordinate system respectively is represented.
Discomposing effect figure is as shown in Fig. 2A to Fig. 2 C, and Fig. 2A is the direction U speed field picture in the embodiment of the present application, and Fig. 2 B is that the application is real
The direction V speed field picture in example is applied, Fig. 2 C is the direction W speed field picture in the embodiment of the present application.
1.2 neighborhood variance methods
The processing of neighborhood variance is carried out to image using neighborhood variance method (Neighborhood variance), formula is as follows:
Wherein, (x, y) is coordinate a little, and S is contiguous range, and m and n are respectively that (the present embodiment takes m=n for the dimension of neighborhood S
=o=3, i.e. S range are 3 × 3 × 3, i.e. 26 neighborhoods), fS(x, y) is the pixel of point (x, y),For the average picture in neighborhood
Element, IS(i, j) is the variance in the neighborhood centered on point (i, j).
By neighborhood variance method, the noise region pixel value in image can be made to be amplified, angiosomes pixel value quilt
Reduction achievees the purpose that distinguish noise region and angiosomes, before Fig. 3 A is neighborhood variance method processing in the embodiment of the present application
Image, Fig. 3 B are the image in the embodiment of the present application after the processing of neighborhood variance method.
The positive negative judgement of 1.3 neighborhoods
Using the positive negative judgement of neighborhood (PN, Neighborhood positive and negative) again to former data into
Row processing, formula are as follows:
PNS(i, j)=| lengthS(P)-lengthS(N)|2
Wherein, lengthSIt (P) is the positive value quantity in neighborhood S, lengthSIt (N) is the negative value quantity in neighborhood S, PNS
(i, j) is the PN value in neighborhood S.
Neighborhood S range is 8 neighborhoods in the present embodiment, so it is U in the embodiment of the present application that the codomain of PN, which is 0~81, Fig. 4 A,
The positive and negative judging result schematic diagram of direction velocity field Image neighborhood, Fig. 4 B are the direction V velocity field Image neighborhood in the embodiment of the present application
Positive and negative judging result schematic diagram, Fig. 4 C are the positive and negative judging result signal of neighborhood of the direction W speed field picture in the embodiment of the present application
Figure.Wherein PN value is bigger, it is meant that consistency of the image in the region is stronger, and PN value is smaller, it is meant that consistency is poorer, and past
Consistency toward blood vessel target area can be far longer than noise region, therefore PN value and former data are carried out dot product, can be effectively
The noise regional luminance in image is significantly reduced, namely inhibits the noise region in image, is judged by neighborhood positive and negative values,
It can achieve the effect of denoising.Fig. 5 A is that velocity field Image neighborhood positive negative judgement weighted results in the direction U are shown in the embodiment of the present application
It is intended to, Fig. 5 B is velocity field Image neighborhood positive negative judgement weighted results schematic diagram in the direction V in the embodiment of the present application, and Fig. 5 C is this Shen
Please in embodiment the direction W speed field picture the positive negative judgement weighted results schematic diagram of neighborhood.
1.4 merge two kinds of neighborhood processing methods
After being judged respectively by neighborhood variance method and neighborhood positive and negative values, first against 1.2 steps as a result, by each of which
Three directional image results in slice are summed, the image after generating 64 merging.Using maximum variance between clusters, by it
It is divided into two parts region, high-brightness region, that is, noise area pixel is set 0, and other area pixels are set to 1 to get to one two
The image of value is denoted as template I1, i.e. 64 template image I1, by the corresponding template I of each sectioning image1With 1.3 steps
Result do dot product, generate image after the processing of N × 3.It is handled by template, can achieve the purpose of further removal noise.
Fig. 6 A is the result schematic diagram in the embodiment of the present application after the two kinds of neighborhood processings merging of the direction U speed field picture, and Fig. 6 B is this Shen
Please result schematic diagram after speed field picture two kinds of neighborhood processings in the direction V merge in embodiment, Fig. 6 C is W in the embodiment of the present application
Result schematic diagram after the two kinds of neighborhood processings merging of direction speed field picture.
1.5 fill up pit-hole and median filtering
The result of 1.4 steps is subjected to morphology padding and median filter process, padding is i.e. in the picture
A connected region is found, this connected region, which meets it, can divide into original image two regions, then judge the two
Whether region contains the edge of entire blood-vessel image, then the pixel value for not including the region at blood-vessel image edge wherein is set to
Global maximum fills this background area, filling criterion is as follows:
Wherein, if p point is located on boundary line, the pixel value of p point is the original pixel value f (p) of p point, by boundary line
The pixel value in the region not comprising blood-vessel image is set to global maximum t in additionmax, here, padding is on bianry image
It is operated, pixel maximum is 1.
By the above criterion, original image is carried out in certain connected region to fill up pit-hole operation, by the step for can be with
So that the features such as pit-hole in two dimensional image are padded.In the present embodiment, it is filled up using 8 connection.It, can be with by padding
Effectively the pit-hole in image is filled.
After carrying out pit-hole filling to image, image is filtered using filter operator, it here can be to knot
Fruit carries out median filter process, it is intended to eliminate image retention and obtain feature of noise.Median filtering is a kind of nonlinear smoothing technology, it
Set the gray value of each pixel to the intermediate value of all pixels point gray value in the point neighborhood window, median filtering is public
Formula is as follows:
Y (i, j)=MedS(f(i-m,j-n),…f(i,j),…f(i+m,j+n))
Wherein, f (i, j) is the pixel value of point (i, j), and S is that (wherein m and n is respectively the dimension of neighborhood S, originally in neighborhood region
M=n=3 in embodiment, i.e. 8 neighborhoods), Y (i, j) is value of the point (i, j) in neighborhood S, is the intermediate value in neighborhood.
By the morphology padding to image, the pit-hole in image can be made to be filled, it can by median filtering
To eliminate the remaining partial noise of image.Fig. 7 A is that the direction U speed field picture fills up pit-hole and intermediate value filter in the embodiment of the present application
Result schematic diagram after wave, Fig. 7 B are that the direction V speed field picture fills up the knot after pit-hole and median filtering in the embodiment of the present application
Fruit schematic diagram, Fig. 7 C are that the direction W speed field picture fills up the result schematic diagram after pit-hole and median filtering in the embodiment of the present application.
1.6 maximum variance between clusters Threshold segmentations
The result of step 1.5 is merged first, three directional images of each sectioning image are added, N number of figure is generated
Picture, as shown in Figure 8.Then Threshold segmentation is carried out to it using maximum variance between clusters, it is divided into two parts region automatically,
Middle that high-brightness region pixel is set 1, low brightness area pixel sets 0.High-brightness region be it is considered that blood vessel region.
Threshold segmentation is wherein carried out to it using maximum variance between clusters (being also Da-Jin algorithm), maximum variance between clusters are by Japanese scholars
Big saliva proposes, is a kind of method that adaptive threshold value determines.It is the gamma characteristic by image, divides the image into background and mesh
Mark 2 parts.Inter-class variance between background and target is bigger, illustrates that the difference for constituting 2 parts of image is bigger.Side between maximum kind
Formula of variance in poor method is set as:
G=w0w1(μ0-μ1)2
Wherein, w0The ratio of all pixels, w are accounted for for the number of pixels that gray values all in image are less than threshold value1For in image
The number of pixels that all gray values are greater than threshold value accounts for the ratio of all pixels, μ0For the flat of the pixels less than threshold value all in image
Equal gray value, μ1For the average gray value of the pixels greater than threshold value all in image.
By traversing all pixels point, it can be deduced that the optimal threshold of image-region is divided, as so that formula of variance reaches
To the gray value of the pixel of maximum value.By Threshold segmentation, blood vessel target area can be completely extracted, Fig. 9 is the application
Result schematic diagram in embodiment after image segmentation.
1.7 morphological erosion
Morphological erosion is carried out to the result of step 1.6, by its boundary contraction, etching operation can be such that due to accidentally
Unnecessary adhesion caused by difference is cut off.Corrosion is defined as follows:
Wherein, A is 64 width images after segmentation, and B is that (in the present embodiment, structure B is defined as [0 10 to structural body;1
1 1;0 1 0]).
The process of corrosion in the pixel data of image A that is, carry out convolution operation for structure B, if with wherein a certain
When structure B and the intersection of image A centered on point are fully belonged in the region of A, then retain the point, all conditions that meet
The set of point is result of the image after excessive erosion.Figure 10 is the result schematic diagram in the embodiment of the present application after Image erosion.
1.8 form preliminary three-dimensional point cloud
64 images of 1.7 step results are reconsolidated according to the image sequence of sectioning image in step 1.1 as 3 dimensions
Form, and retain therein 3 and tie up maximum UNICOM domain, the blood vessel mesh of 3 dimensions can be tentatively extracted by retaining the 3 largest connected domains of dimension
Region is marked, the initial three-dimensional point cloud model of blood vessel is formed.
By step 1.1~1.8, the initial three-dimensional point cloud model of blood vessel can be tentatively generated.Figure 11 is the application implementation
Initial three-dimensional point cloud model schematic diagram in example, wherein light areas is the aorta to be extracted among image, is glued on aorta
Dark parts even are pulmonary artery, need to extract aorta the part steps in execution above-mentioned steps again, purpose
It is to separate aorta with pulmonary artery, if aorta detection environment is preferable, executing an image processing operations can be with
Preferably aorta three-dimensional point cloud model is obtained, then is not necessarily to next vessel extraction operations.
Step 2: angiosomes extract
2.1 slicing treatments again
To the initial three-dimensional point cloud model tentatively extracted in step 1.8, in such a way that step 1.1 is to image slice again
It is sliced, 64 two dimensional images are obtained.
The production of 2.2 target templates
For 64 two dimensional images obtained in 2.1 steps, due in step 1.7, in order to enable two-dimensional effect picture
Wrong connection in three dimensions is avoided as far as possible and has carried out morphological erosion operation, therefore carries out morphological dilation herein, shape
State expansion formula is as follows:
Wherein A is the two dimensional image that step 2.1 obtains, and B is that (in the present embodiment, structure B is defined as [0 1 to structural body
0;1 1 1;0 1 0]).
The process of expansion is that structure B is carried out convolution operation on structure A, if the structural body centered on certain point
B is in the moving process of former data A, and there are overlapping regions with former data A, then retains the point, all points for meeting the condition
Set is result of the image after morphological dilations.Will be after morphological dilations as a result, place as blood vessel target area
Template is managed, remembers that the template is I2, that is, generate 64 template image I2.As the processing template of target area.Figure 12 is that the application is real
Apply the result schematic diagram in example after morphological dilations.
2.3 interative computations generate maximal end point cloud
The 64 template image I obtained with step 2.22Instead of 64 template image I in step 1.41, in step 1.1
64 sectioning images execute the processing again of step 1.3~1.6 steps again, then carry out 1.8 step process to last result,
Obtain arterial vascular final three-dimensional point cloud model.Figure 13 is final three-dimensional point cloud model schematic diagram in the embodiment of the present application.
The artificial enhancing of 2.4 innominate arteries, arteria carotis communis and subclavian artery
For step 2.3 or step 1.8 obtain as a result, be added an artificial option, enhance to innominate artery, neck always moves
The extraction of arteries and veins and subclavian artery is added one manually if these three arterial branches feature in former data is unobvious
The step of enhancing, schematic diagram is as shown in figure 14, select region to be reinforced by artificial frame, and three raise upward in region to be reinforced
The portion of similar triangle is respectively innominate artery, arteria carotis communis and subclavian artery, treats enhancing region and carries out execution step again
Rapid 1.1~1.2,1.5~1.8 processing operations, obtain the three-dimensional point cloud model in region to be reinforced.
The 2.5 initial three-dimensional point clouds for obtaining the three-dimensional point cloud model in the region to be reinforced that step 2.4 obtains and step 1.8
Model merges, and obtains the three-dimensional point cloud model of enhanced target blood;Alternatively, the area to be reinforced that step 2.4 is obtained
The final three-dimensional point cloud model that the three-dimensional point cloud model and step 2.3 in domain obtain merges, and obtains enhanced target blood
Three-dimensional point cloud model.Figure 15 is enhanced three-dimensional point cloud model schematic diagram in the embodiment of the present application.
Step 3: blood vessel three-dimensionalreconstruction
To the three-dimensional point cloud model that step 1.8, step 2.3 or step 2.4 obtain, using Poisson surface reformation algorithm, to 3
Dimension point cloud model is handled, and the three-dimension curved surface of smooth closure is obtained, which is the best fit curved surface to cloud,
The entity that the three-dimension curved surface is included simultaneously is exactly the objective blood vessel after reconstruct.Figure 16 is target blood in the embodiment of the present application
The three-dimensionalreconstruction model schematic of pipe.
The processing of neighborhood variance method is carried out to the speed field picture in 4D flow MRI image first in the present embodiment.Again will
Image carries out the judgement of neighborhood positive and negative values, and the result after positive and negative values are judged is weighted to image.Then image progress morphology is filled out
Operation is filled to fill up pit-hole, and uses median filtering, so that image is full and more smooth.Then maximum variance between clusters are used
Automatic Threshold segmentation is carried out to image.Then retain 3 dimension connected regions after image being carried out morphological erosion operation, as figure
The initial three-dimensional point cloud model of picture.Initial three-dimensional point cloud model carries out to identical as original sectioning image direction and dimension is consistent cuts
Piece operation, does morphology expansion process later, using treated image as the region limitation template of former sectioning image.According to mould
Image is carried out an iteration operation again by plate, is ultimately produced the final three-dimensional point cloud model of target blood, is utilized Poisson surface
Restructing algorithm obtains three-dimension curved surface, and the entity which includes is the three-dimensionalreconstruction body of target blood.
The vessel extraction method provided in the embodiment of the present application has the advantages that
1, using the extraction of two kinds of neighborhood informations, the structural information of image is efficiently utilized, improves noise suppression effect,
Comparing simple threshold method or growth method has apparent advantage.
2, automatic threshold segmentation is carried out using maximum variance between clusters, the grayscale information of image is utilized well, realized
Complete automated image segmentation, has preferably segmentation effect.
3,3 dimension connected regions retain, and the error for handling introducing on 2d can be effectively avoided, realize 3 dimensions
The processing of middle data.
4, high degree reduces manual intervention, so that the extraction of model is more stable, it is more efficient.
Embodiment five
Based on the same inventive concept, the embodiment of the present application also provides a kind of vessel extraction devices.As shown in figure 17, the dress
It sets and includes:
Acquisition unit 171, for obtaining the N layer sectioning image of target blood, wherein each layer of sectioning image includes M
The speed field picture in direction, N and M take positive integer;
Processing unit 172, for being based on Image Pretreatment Algorithm, by each layer of slice map in the N layers of sectioning image
As being pre-processed, first kind N tomographic image is obtained;Wherein, described image Preprocessing Algorithm includes that at least one neighborhood processing is calculated
Method;
Processing unit 172, is also used to based on preset image segmentation algorithm, from each in the first kind N tomographic image
It is partitioned into the target area comprising the target blood in tomographic image, obtains the second class N tomographic image;
Reconfiguration unit 173 carries out the second class N tomographic image for the image sequence based on the N layers of sectioning image
Merge, obtains the initial three-dimensional point cloud model of the target blood.
In some embodiments, processing unit 172 are specifically used for being based on the first neighborhood processing algorithm, described N layers are sliced
The speed field picture in each of image direction carries out neighborhood processing, obtains N number of template image;It is calculated based on the second neighborhood processing
The speed field picture in each of N layers of sectioning image direction is carried out neighborhood processing, obtains third class N tomographic image by method;
By i-th of template image in N number of template image respectively with M direction of the i-th tomographic image in the third class N tomographic image
Speed field picture carries out point multiplication operation, obtains first kind N tomographic image, i takes the positive integer less than or equal to N;Wherein, described
One neighborhood processing algorithm is different from the second neighborhood processing algorithm.
In some embodiments, reconfiguration unit 173 are specifically used for being based on preset Morphology Algorithm, by the second class N
The target area of each tomographic image in tomographic image carries out morphological erosion processing, the second class N tomographic image after being corroded;Base
In the image sequence of the N layers of sectioning image, the second class N tomographic image after the corrosion is merged, the target is obtained
The initial three-dimensional point cloud model of blood vessel.
In some embodiments, after obtaining the initial three-dimensional point cloud model of the target blood, processing unit 172,
It is also used to for the initial three-dimensional point cloud model being sliced, obtains N number of template image;It will be i-th in N number of template image
Template image carries out point multiplication operation with the speed field picture in M direction of the i-th tomographic image in the third class N tomographic image respectively,
Obtain first kind N tomographic image;Based on preset image segmentation algorithm, from each tomographic image in the first kind N tomographic image
It is partitioned into the target area comprising the target blood, obtains the second class N tomographic image;
Correspondingly, reconfiguration unit 173, is also used to the image sequence based on the N layers of sectioning image, by the second class N
Tomographic image merges, and obtains the final three-dimensional point cloud model of the target blood.
In some embodiments, processing unit 172 are obtained specifically for the initial three-dimensional point cloud model to be sliced
To N tomographic image;It will include the target blood in each tomographic image in the N tomographic image based on preset Morphology Algorithm
Region carry out morphological dilations processing, obtain N number of template image.
In some embodiments, processing unit 172, specifically for carrying out morphology filling behaviour to the image after point multiplication operation
Work and/or median filtering, obtain first kind N tomographic image.
In some embodiments, processing unit 172 are also used to select information based on user, determine the N layers of sectioning image
Each of region to be reinforced in the speed field picture in direction;It wherein, include the target blood in the region to be reinforced
At least partly blood vessel of pipe;Based on the first neighborhood processing algorithm, to the region to be reinforced in the speed field picture in each direction
Neighborhood processing is carried out, first kind N tomographic image is obtained;Based on preset image segmentation algorithm, from the first kind N tomographic image
It is partitioned into the target area comprising the target blood in each tomographic image, obtains the second class N tomographic image;
Correspondingly, reconfiguration unit 173, is also used to the image sequence based on the N layers of sectioning image, by the second class N
Tomographic image merges, and obtains the three-dimensional point cloud model in region to be reinforced.
In some embodiments, reconfiguration unit 173 are also used to the three-dimensional point cloud model in the region to be reinforced and described
Initial three-dimensional point cloud model merges, and obtains the three-dimensional point cloud model of the enhanced target blood;Alternatively, will it is described to
The three-dimensional point cloud model and the final three-dimensional point cloud model for enhancing region merge, and obtain the enhanced target blood
Three-dimensional point cloud model.
In some embodiments, processing unit 172, specifically for by each tomographic image in the first kind N tomographic image
The speed field picture in M direction be merged into a sub-picture, obtain N tomographic image to be split;It is true based on maximum variance between clusters
The optimal threshold of the fixed N tomographic image to be split;Schemed based on optimal threshold from each layer in the N tomographic image to be split
It is partitioned into the target area comprising the target blood as in, obtains the second class N tomographic image.
By adopting the above technical scheme, it is carried out using N layer sectioning image of at least one neighborhood processing algorithm to target blood
Neighborhood processing can effectively inhibit the interference of noise in image, and target blood is accurately positioned, and guarantee subsequent image cutting operation
Accuracy guarantees the quality of target blood three-dimensionalreconstruction image to improve the extraction accuracy of target blood three-dimensional point cloud model.
Based on the hardware realization of each unit in above-mentioned vessel extraction device, the embodiment of the present application also provides another blood vessels
Extraction element, the device include: processor and the memory for being configured to the computer program that storage can be run on a processor;
Wherein, when processor is configured to operation computer program, the method and step in previous embodiment is executed.
Certainly, when practical application, the various components in the vessel extraction device are coupled by bus system.It can manage
Solution, bus system is for realizing the connection communication between these components.Bus system further includes electricity in addition to including data/address bus
Source bus, control bus and status signal bus in addition.
In practical applications, above-mentioned processor can be application-specific IC (ASIC, Application
Specific Integrated Circuit), digital signal processing device (DSPD, Digital Signal Processing
Device), programmable logic device (PLD, Programmable Logic Device), field programmable gate array
At least one of (Field-Programmable Gate Array, FPGA), controller, microcontroller, microprocessor.It can
To understand ground, for different equipment, the electronic device for realizing above-mentioned processor function can also be other, the application reality
Example is applied to be not especially limited.
Above-mentioned memory can be volatile memory (volatile memory), such as random access memory (RAM,
Random-Access Memory);Or nonvolatile memory (non-volatile memory), such as read-only memory
(ROM, Read-Only Memory), flash memory (flash memory), hard disk (HDD, Hard Disk Drive) or solid
State hard disk (SSD, Solid-State Drive);Or the combination of the memory of mentioned kind, and to processor provide instruction and
Data.
In the exemplary embodiment, the embodiment of the present application also provides a kind of computer readable storage medium, for example including
The memory of computer program, above-mentioned computer program can be executed by the processor in vessel extraction device, to complete aforementioned side
Method step.
It should be understood by those skilled in the art that, embodiments herein can provide as method, system or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the application
Formula.Moreover, the application, which can be used, can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present application
Schematic diagram and/or block diagram describe.It should be understood that can be realized by computer program instructions in flow diagram and/or block diagram
Each flow and/or block and process and/or box in flow diagram and/or block diagram combination.It can provide this
A little computer program instructions are to general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices
Processor to generate a machine so that the finger executed by the processor of computer or other programmable data processing devices
It enables generating and refer to for realizing in flow diagram one process or multiple processes and/or block diagrams one box or multiple boxes
The device of fixed function.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
The manufacture of device is enabled, which realizes in one side of flow diagram one process or multiple processes and/or block diagrams
The function of being specified in frame or multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one process of flow diagram or multiple processes and/or box
The step of function of being specified in figure one box or multiple boxes.
The above, the only preferred embodiment of the application, are not intended to limit the protection scope of the application.