CN109191468A

CN109191468A - A kind of method, apparatus and storage medium of vessel extraction

Info

Publication number: CN109191468A
Application number: CN201810936725.XA
Authority: CN
Inventors: 魏润杰; 高琪; 李博文; 吴鹏
Original assignee: Hangzhou Sheng Shi Technology Co Ltd
Current assignee: Hangzhou Sheng Shi Technology Co Ltd
Priority date: 2018-08-16
Filing date: 2018-08-16
Publication date: 2019-01-11
Anticipated expiration: 2038-08-16
Also published as: CN109191468B

Abstract

The embodiment of the invention discloses a kind of methods of vessel extraction, this method comprises: obtaining the N layer original image of target blood, N takes the integer greater than 1；Using preset image zooming-out algorithm, the target blood in N layer original image in all original images is extracted, N layers of image to be marked are obtained；Based at least one connection label of b layers of image to be marked in N layers of image to be marked, the connected domain of a layers of image to be marked of label obtains N layers of label and completes image；Wherein, a and b takes the positive integer less than or equal to N, and the difference of a and b are 1；Image is completed based on N layers of label, three-dimensionalreconstruction is carried out to target blood, obtains the three-dimensionalreconstruction image of target blood.The embodiment of the invention also discloses a kind of vessel extraction device and storage mediums.

Description

A kind of method, apparatus and storage medium of vessel extraction

Technical field

The present invention relates to image processing techniques more particularly to the method, apparatus and storage medium of a kind of vessel extraction.

Background technique

With the continuous development of science and technology, Medical Imaging Technology achieves innovation in area of medical diagnostics and breaks through, extensive Applied to clinical diagnosis.Traditional diagnostic imaging is analyzed by the healthcare givers of profession mostly, although artificial operation With high-precision advantage, but increasingly huge image data and diagnostic requirements amount, constantly aggravate the work of image department doctor Intensity is easy to cause operator dog-tired, reduces work quality.Relatively traditional diagnostic imaging mode, computer based Auxiliary diagnosis 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. Such as: the coronary artery cutting techniques of coronary artery CT radiography (Coronary Angiography, CTA) image are conducive to cardiovascular disease Screening and diagnosis；The aorta segmentation technology of thoracic cavity nuclear magnetic resonance image (Magnetic Resonance Imaging, MRI) has Conducive to the screening and diagnosis of arotic disease.Therefore automatic, the accurate Segmentation Research of histoorgan becomes in medical image Must be particularly important, it can greatly mitigate the work of image department doctor, improve working efficiency and work quality.

Traditional image segmentation algorithm is applied to have been achieved for good effect in medical image processing.Based on threshold value Dividing method, if the histogram of image can be divided into Ganlei with one or several gray values, gray value in original image The pixel being in the same gray level is attributed to same class, this process superiority and inferiority is often depending on gray threshold.In addition, due to Threshold method is to consider single pixel, and do not consider the spatial relationship of pixel, so very sensitive to noise.The segmentation of region growing Method also has good progress on medical image, it is a series of given seed points in the picture, by the pixel of similar quality Merge and constitutes region.However the selection of final segmentation result and seed point has a much relations, while this method to noise also very Sensitivity, it is affected by noise to be likely to form discontinuity zone.Since active contour model can also obtain in the case where strong noise Continuously, smooth closure partitioning boundary, the medical image cutting techniques based on active contour model are also more and more by people Concern.The basic thought of active contour model is that an initial curve is arranged first, then makes initial curve in image force It is constantly shunk under the action of, is finally retracted to object edge.Although can be supported in active profile cutting procedure antimierophonic dry It disturbs, however the superiority and inferiority of this process depends on the position of initial curve, the presence additionally, due to noise makes suitable initial song The determination of line becomes more difficult.

For traditional dividing method although having been achieved for good segmentation effect, more or less requiring is manually dry In advance, while the effect of segmentation is dependent on given some initial values, and is easy to be influenced by noise spot.Therefore, such methods Be difficult to accomplish in medical image segmentation it is really automatic, and be merely able to allow not the healthcare givers of relevant professional knowledge go to make With, it is difficult to really put into clinical application.

Summary of the invention

In order to solve the above technical problems, an embodiment of the present invention is intended to provide a kind of method, apparatus of vessel extraction and storages Medium can weaken noise in image interference, accurately identify and extract the target blood in image.

The technical scheme of the present invention is realized as follows:

The embodiment of the invention provides a kind of methods of vessel extraction, comprising:

The N layer original image of target blood is obtained, N takes the integer greater than 1；

Using preset image zooming-out algorithm, the target blood in the N layer original image in all original images is extracted, Obtain N layers of image to be marked；

Based at least one connection label of b layers of image to be marked in the N layers of image to be marked, a layers of label to The connected domain of tag image obtains N layers of label and completes image；Wherein, a and b takes the positive integer less than or equal to N, a and b Difference be 1；

Image is completed based on N layers of label, three-dimensionalreconstruction is carried out to the target blood, obtains the three-dimensional of the target blood Reconstructed image.

In above scheme, described at least one connection mark based on b layers of image to be marked in the N layers of image to be marked Note, the connected domain of a layers of image to be marked of label obtain N layers of label and complete image, comprising: use the first labelling strategies, be based on At least one connection label of b layers of image to be marked, the connected domain of first time a layers of image to be marked of label obtain first The N tomographic image of secondary label；Using the second labelling strategies, based at least one connection label of b layers of image to be marked, for the second time The connected domain of a layers of image to be marked of label obtains the N tomographic image of second of label；N layer figure based on first time label The N tomographic image of picture and second of label, obtains N layers of label and completes image.

In above scheme, first labelling strategies are successively to mark from top to bottom, wherein a takes the positive integer greater than 1, b For a-1；Second labelling strategies are successively to mark from bottom to top, wherein a takes the positive integer less than N, b a+1.

In above scheme, before using first labelling strategies, the method also includes: using the 1st layer as initially Mark layer carries out two dimension connection label to the 1st layer of image to be marked, obtains at least one connection mark of the 1st layer of image to be marked Note；The connection label of remainder layer image to be marked in the N layers of image to be marked in addition to the 1st layer is set to 0.

In above scheme, before using second labelling strategies, the method also includes: using n-th layer as initially Mark layer, the connection label of n-th layer image in the N tomographic image that the first time is marked, n-th layer waits for when marking as second The connection of tag image marks；By the connection mark of the remainder layer image to be marked in the N layers of image to be marked in addition to n-th layer Note is set to 0.

In above scheme, the N tomographic image of the N tomographic image based on first time label and second of label, It obtains N layers of label and completes image, comprising: the connection of the i-th obtained tomographic image is relatively marked to mark and mark for the second time for the first time The connection of the i-th obtained tomographic image marks, and retains the connected domain with identical connection label, obtains i-th layer of label and completes image； Wherein, i takes the positive integer less than or equal to N.

In above scheme, described at least one connection mark based on b layers of image to be marked in the N layers of image to be marked Note, the connected domain of a layers of image to be marked of label, comprising: when at least one connected domain in a layers of image to be marked and b layers When single connected domain overlapping in image to be marked, by the connection of connected domain single in b layer image to be marked label, as the The connection label of at least one connected domain in a layers of image to be marked；When connected domain single in a layers of image to be marked and b layers In image to be marked multiple connected domains overlapping when, by b layers of image to be marked be individually connected in a layers of image to be marked The connection of the maximum connected domain of domain overlapping area marks, and the connection as the single connected domain in a layers of image to be marked marks； When in a layers of image to be marked at least one connected domain with it is Chong Die without connected domain in b layers of image to be marked when, obtain at least one A new connection mark；At least one described new connection is identified as at least one of a layers of image to be marked The connection of connected domain marks.

In above scheme, the preset image zooming-out algorithm includes: to pre-process to original image, after obtaining processing Image；Based on the first clustering algorithm, segmentation is described treated image obtains area-of-interest (Region of Interest, ROI)；Target blood in the ROI region is enhanced, enhanced image is obtained；It is poly- based on second Class algorithm extracts the target blood in the enhanced image, obtains image to be marked.

It is described that original image is pre-processed in above scheme, the image that obtains that treated, comprising: to original image Carry out gray scale normalization processing, the image after being normalized；Fuzzy processing is carried out to the image after the normalization, is obtained Treated image.

A kind of vessel extraction device is additionally provided in the embodiment of the present invention, described device includes: processor and memory；Its In,

The processor is for executing the vessel extraction program stored in memory, to perform the steps of

In above scheme, the processor is specifically used for executing the vessel extraction program that stores in memory, with realize with Lower step: using the first labelling strategies, based at least one connection label of b layers of image to be marked, a layers of first time label The connected domain of image to be marked obtains the N tomographic image marked for the first time；Using the second labelling strategies, it is based on b layers of figure to be marked At least one connection label of picture, the connected domain of a layers of image to be marked of second of label obtain the N layer figure of second of label Picture；The N tomographic image of N tomographic image and second of label based on first time label, obtains N layers of label and completes image.

In above scheme, before using first labelling strategies, the processor is also used to execute in memory and deposit The vessel extraction program of storage carries out two to the 1st layer of image to be marked to perform the steps of using the 1st layer as initial markers layer Dimension connection label obtains at least one connection label of the 1st layer of image to be marked；The 1st layer will be removed in the N layers of image to be marked Except remainder layer image to be marked connection label be set to 0.

In above scheme, before using second labelling strategies, the processor is also used to execute in memory and deposit The vessel extraction program of storage, to perform the steps of the N layer for marking the first time using n-th layer as initial markers layer figure The connection label of n-th layer image as in, the connection label of n-th layer image to be marked when being marked as second；By described N layers to The connection label of remainder layer image to be marked in tag image in addition to n-th layer is set to 0.

In above scheme, the processor is specifically used for executing the vessel extraction program that stores in memory, with realize with Lower step: relatively marking the connection label of the i-th obtained tomographic image for the first time and marks the company of the i-th obtained tomographic image for the second time Logical label retains the connected domain with identical connection label, obtains i-th layer of label and completes image；Wherein, i, which takes, is less than or waits In the positive integer of N.

In above scheme, the processor is specifically used for executing the vessel extraction program that stores in memory, with realize with Lower step: when at least one connected domain is Chong Die with the single connected domain in b layers of image to be marked in a layers of image to be marked When, the connection of connected domain single in b layers of image to be marked is marked, is connected to as at least one in a layers of image to be marked The connection in domain marks；When connected domain single in a layers of image to be marked is Chong Die with multiple connected domains in b layers of image to be marked When, by the connection in b layers of image to be marked with the maximum connected domain of connected domain overlapping area single in a layers of image to be marked Label, the connection as the single connected domain in a layers of image to be marked mark；When at least one in a layers of image to be marked Connected domain with it is Chong Die without connected domain in b layers of image to be marked when, obtain at least one new connection mark；By described at least one A new connection mark is marked respectively as the connection of at least one connected domain in a layers of image to be marked.

In above scheme, the processor is specifically used for executing the vessel extraction program that stores in memory, with realize with Lower step: pre-processing original image, the image that obtains that treated；Based on the first clustering algorithm, divide the processing Image afterwards, obtains ROI region；Target blood in the ROI region is enhanced, enhanced image is obtained；It is based on Second of clustering algorithm extracts the target blood in the enhanced image, obtains image to be marked.

In above scheme, the processor is specifically used for executing the vessel extraction program that stores in memory, with realize with Lower step: gray scale normalization processing is carried out to original image, the image after being normalized；To the image after the normalization into Row Fuzzy processing, the image that obtains that treated.

The method, apparatus and storage medium of a kind of vessel extraction provided in an embodiment of the present invention obtain the N layer of target blood Original image, N take the integer greater than 1；Using preset image zooming-out algorithm, all original images in N layer original image are extracted In target blood, obtain N layers of image to be marked；At least one company based on b layers of image to be marked in N layers of image to be marked Logical label, the connected domain of a layers of image to be marked of label obtain N layers of label and complete image；Wherein, a and b take be less than or The difference of positive integer equal to N, a and b are 1；Image is completed based on N layers of label, three-dimensionalreconstruction is carried out to target blood, obtains mesh Mark the three-dimensionalreconstruction image of blood vessel.

It by adopting the above technical scheme, can when extracting the target blood in original image by preset image zooming-out algorithm To exclude ambient noise interference, and extraction process is not necessarily to human intervention, further, using target blood in image layer to be marked Between layer dependencies, treat tag image carry out space interlayer connection label, target blood is accurately positioned, to improve The extraction accuracy of target blood ensure that the quality of target blood three-dimensionalreconstruction image.

Detailed description of the invention

Fig. 1 is the first pass schematic diagram for the method that medium vessels of the embodiment of the present invention extract；

Fig. 2 is to be connected to labeling process schematic diagram in the embodiment of the present invention between space layer；

Fig. 3 is to be connected to marking convention schematic diagram in the embodiment of the present invention between space layer；

Fig. 4 is the second procedure schematic diagram for the method that medium vessels of the embodiment of the present invention extract；

Fig. 5 is the schematic diagram before the image preprocessing in the embodiment of the present invention one；

Fig. 6 is the schematic diagram after the image preprocessing in the embodiment of the present invention one；

Fig. 7 is the binary map of the FCM cluster background separation result in the embodiment of the present invention one；

Fig. 8 is the schematic diagram of the region segmentation result behind the largest connected domain of reservation in the embodiment of the present invention one；

Fig. 9 is the schematic diagram of the blood vessel enhancing result in the embodiment of the present invention one；

Figure 10 is the schematic diagram of the vessel extraction result in the embodiment of the present invention one；

Figure 11 is the schematic diagram of the three-dimensional coronary result in the embodiment of the present invention one；

Figure 12 is the schematic diagram before the image preprocessing in the embodiment of the present invention two；

Figure 13 is the schematic diagram after the image preprocessing in the embodiment of the present invention two；

Figure 14 is the binary map of the FCM cluster background separation result in the embodiment of the present invention two；

Figure 15 is the schematic diagram of the region segmentation result behind the largest connected domain of reservation in the embodiment of the present invention two；

Figure 16 is the schematic diagram of the blood vessel enhancing result in the embodiment of the present invention two；

Figure 17 is the schematic diagram of the vessel extraction result in the embodiment of the present invention two；

Figure 18 is the schematic diagram of the three-dimensional aorta result in the embodiment of the present invention two；

Figure 19 is the composed structure schematic diagram of medium vessels of embodiment of the present invention extraction element.

Specific embodiment

The characteristics of in order to more fully hereinafter understand the embodiment of the present invention and technology contents, with reference to the accompanying drawing to this hair The realization of bright embodiment is described in detail, appended attached drawing purposes of discussion only for reference, is not used to limit the embodiment of the present invention.

Embodiment one

As shown in Figure 1, the method for vessel extraction includes:

Step 101: obtaining the N layer original image of target blood, N takes the integer greater than 1；

Step 102: using preset image zooming-out algorithm, extract the target in N layer original image in all original images Blood vessel obtains N layers of image to be marked；

Step 103: at least one connection label based on b layers of image to be marked in N layers of image to be marked, label a The connected domain of layer image to be marked obtains N layers of label and completes image；Wherein, a and b takes the positive integer less than or equal to N, a Difference with b is 1；

Step 104: completing image based on N layers of label, three-dimensionalreconstruction is carried out to target blood, obtains the three-dimensional of target blood Reconstructed image.

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 layer original image can be the N layer two-dimensional slice image of target blood, pass through slice scanner Obtained from target blood is scanned into computer.Scanning technique has: MRI, CTA, magnetic resonance angiography (Magnetic Resonance Angiography, MRA), computed tomography (Computed Tomography, CT) etc. it is one or more Combination.

In practical application, preset image zooming-out algorithm may include: to pre-process to original image, after obtaining processing Image；Based on the first clustering algorithm, image after dividing processing obtains ROI region；To the target blood in ROI region Enhanced, obtains enhanced image；Based on second of clustering algorithm, the target blood in enhanced image is extracted, is obtained To image to be marked.

Here, original image is pre-processed, the image that obtains that treated specifically includes: carrying out ash to original image Spend normalized, the image after being normalized；Fuzzy processing is carried out to the image after normalization, the figure that obtains that treated Picture.

In the embodiment of the present invention, ROI region can be gradually extracted by using the mode repeatedly clustered, it is dry to exclude noise It disturbs, removes background parts information, help to improve the extraction accuracy of target blood.

Illustratively, the image zooming-out algorithm of target blood specifically includes in original image:

A1. image preprocessing

A1-1. input picture carries out gray scale normalization processing to the image of input, eliminates imaging factors (especially with this Noise) influence to image grayscale, reduce the difference between different images in property tissue gray scale of the same race, make grayscale information at For a stable feature, effective support is provided for subsequent image procossing.

A1-2. Fuzzy processing is carried out to the image data after normalizing by step A1-1 using bilateral filtering method. The marginal information of image can preferably be retained while blurred picture by bilateral filtering, be the ROI region in next step Extraction is prepared.

A2. region segmentation

The first clustering algorithm is Fuzzy C-Means Cluster Algorithm (Fuzzy c-means algorithm, FCM), is utilized FCM algorithm extracts target area, wherein cluster number is l=2, that is, passes through FCM algorithm for background in image data and organ group Tissue region separation.Largest connected domain extraction (retaining largest connected domain) is carried out to the binary map of obtained tissue regions part, is obtained To ROI region (such as: the heart area in CTA, aorta regions in MRI).In practical application, using other clustering algorithms Target area can be extracted, above-mentioned FCM algorithm is not the restriction to the first clustering algorithm in the embodiment of the present invention.

A3. image blood vessel enhances

Blood vessel enhancing is carried out to ROI region obtained in step A2 using multiple dimensioned 3D-Frangi filtering.Frangi filtering The coronary tissue structure in image can be protruded, non-coronary institutional framework, the shape complicated and changeable suitable for coronary artery network etc. are inhibited Shape.However the scale size of blood vessel is different, the filter of single scale cannot the blood vessel structure to all scales all generate it is very big Response.When using multiple dimensioned being handled, the blood vessel under different scale always generates the filter under a certain scale maximum Response, to be enhanced.

A4 target blood is extracted

Second of clustering algorithm may be FCM algorithm, carry out vessel extraction using FCM algorithm, wherein cluster number is l =2, i.e., blood vessel is extracted from enhanced image by FCM algorithm.It, can also be real using other clustering algorithms in practical application Existing vessel extraction, above-mentioned FCM algorithm is not the restriction to second of clustering algorithm in the embodiment of the present invention.

In practical application, step 103 is specifically included: using the first labelling strategies, at least based on b layers of image to be marked One connection label, the connected domain of first time a layers of image to be marked of label obtain the N tomographic image marked for the first time；Using Two labelling strategies, at least one connection based on b layers of image to be marked mark, a layers of image to be marked of second of label Connected domain obtains the N tomographic image of second of label；The N tomographic image of N tomographic image and second of label based on first time label, It obtains N layers of label and completes image.

Here, the first labelling strategies are successively to mark from top to bottom, wherein a takes the positive integer greater than 1, b a-1；Second Labelling strategies are successively to mark from bottom to top, wherein a takes the positive integer less than N, b a+1.

Here, before using the first labelling strategies, this method further include: using the 1st layer as initial markers layer, to the 1st Layer image to be marked carries out two dimension connection label, obtains at least one connection label of the 1st layer of image to be marked；By N layers wait mark The connection label of remainder layer image to be marked in note image in addition to the 1st layer is set to 0.I.e. first to the 1st layer of image to be marked Each connected domain carry out nonzero integer label, using the 1st layer of image to be marked as initial markers layer after the completion of label, by remaining Connection label in image to be marked is set to 0, carries out top-down label to N layers of image to be marked.

Before using the second labelling strategies, this method further include: using n-th layer as initial markers layer, by the first deutero-albumose The connection label of n-th layer image in the N tomographic image of note, the connection label of n-th layer image to be marked when being marked as second；It will The connection label of remainder layer image to be marked in N layers of image to be marked in addition to n-th layer is set to 0.It will mark for the first time Obtained in n-th layer image initial markers layer of the connection label as second connection label, to N layers of images progress to be marked Label from bottom to top.

Marking convention is specifically connected in practical application, in step 103 can be with are as follows: when in a layers of image to be marked at least When one connected domain is Chong Die with the single connected domain in b layers of image to be marked, by single connected domain in b layers of image to be marked Connection label, as in a layer image to be marked at least one connected domain connection mark；When in a layers of image to be marked When single connected domain is Chong Die with multiple connected domains in b layers of image to be marked, by b layers of image to be marked with a layers to The connection label of the maximum connected domain of single connected domain overlapping area in tag image, as the list in a layers of image to be marked The connection of a connected domain marks；When in a layers of image to be marked at least one connected domain in b layers of image to be marked without being connected to When domain is overlapped, at least one new connection mark is obtained；The connection that at least one is new is identified as a layers of figure to be marked The connection label of at least one connected domain as in.

Here, b layers of image to be marked can be understood as the image that nonzero integer label has been carried out in initial markers layer, or Person's image that other have been labeled in same secondary labeling process.

Further, the N tomographic image of N tomographic image and second of label based on first time label, obtains N layers of label and completes Image, comprising: the connection of the i-th obtained tomographic image is relatively marked to mark and mark the i-th obtained tomographic image for the second time for the first time Connection label, retain the connected domain with identical connection label, obtain i-th layer of label completion image；Wherein, i take be less than or Person is equal to the positive integer of N.

Illustratively, connection labeling process specifically includes following:

B1. the image after vessel extraction is inputted, label is using the 1st layer as initial markers layer for the first time.It is to be marked to the 1st layer Image carries out two-dimentional connection (4- connection, 8- connection) label, remembers that first tag image is Label₁ ¹, connected domain number on the image Mesh is L¹ _num, it is that each connected domain carries out nonzero integer label according to connected domain number.

B2. two dimension connection label is carried out to a layers of image to be marked, note tag image is Label_a ¹, by each connected domain mark It is denoted as 0.

B3. to Label_b ^pAnd Label_a ^pSpace interlayer connection label is carried out, wherein | a-b |=1:

I) work as Label_a ^pIt is middle to mark the target connected domain for being and Label_b ^pWhen single connected domain overlapping, then by Label_a ^pIn Target connected component labeling is Label_b ^pThe connection label of single connected domain；

Ii) work as Label_a ^pIt is middle to mark the target connected domain for being and Label_b ^pIn multiple connected domains overlapping when, compare overlapping Size is marked with the connection of Maximum overlap area, modifies Label_a ^pThe connection of middle target area marks；

Iii) work as Label_b ^pIt is middle mark be target connected domain not with Label_a ^pIn any connected domain overlapping when, then use L^p _num=L^p _num+ 1 integer value marks target connected domain, wherein p=1.Here, b layers of image to be marked are initial markers layer In carried out the image of nonzero integer label, or the image that other have been labeled in same secondary labeling process

B4. enable b=a-1, wherein a=2:N, repeat step B.2 with step B.3, marked until by all original images At.

B5. second of label remembers tag image Label using n-th layer as initial markers layer_N ²=Label_N ¹, n-th layer figure As upper connected domain number is L² _num。

B6. enable b=a+1, wherein a=N-1:1, repeat step B.2 with step B.3, marked until by all original images It completes.

B7. two kinds of label L abel of more each image_i ¹And Label_i ², retain wherein same tag region, wherein i= 1:N。

As shown in Fig. 2, carry out space interlayer connection label from top to bottom for the first time, using the 1st layer as initial markers layer, by The 2nd layer of layer label, to n-th layer, obtains the N tomographic image marked for the first time.Carry out space interlayer connection mark from bottom to top for the second time Note, using n-th layer as initial markers layer, i.e., using the last layer after first time label as initial markers next time Layer, obtains the N tomographic image of second of label by N-1 layers to the 1st layer of layer-by-layer label.Take the i-th tomographic image for marking obtain for the first time With mark the i-th obtained tomographic image for the second time, i takes the positive integer less than or equal to N, retains label same section and (marks Part for 1 and 2), obtain i-th layer of label and complete image, and so on complete the label of its N layers of image to be marked.

Further, the specific labeling process of connected domain of each image to be marked is shown in Fig. 3, as shown in figure 3, The connection for including in b layers of image to be marked in first time labeling process is identified with 1 and 2, the connection of a layers of image to be marked Mark is 0, after a layers of image to be marked and b layers of picture registration to be marked, a layers of image connectivity domain 34 to be marked and company Logical domain 35 is Chong Die with connected domain 31 single in b layers of image to be marked and single connected domain 32 respectively, then a layers of image to be marked The connection of middle connected domain 34 is identified as 2, and the connection of connected domain 35 is identified as 1；Connected domain 33 and b in a layers of image to be marked Connected domain 31 and connected domain 32 have overlapping in layer image to be marked, identify 1 with the connection of the connected domain 32 of Maximum overlap area Mark connected domain 33；Connected domain non-overlapping in connected domain 36 and b layers of image to be marked, then give birth in a layers of image to be marked The connection of Cheng Xin identifies, and new connection is identified as current connected domain sum and adds 1, it is assumed that current connected domain sum is 2, then new Connection is identified as 3, and the connection mark of connected domain 36 is 3, in this way, obtaining a tomographic image or the second deutero-albumose that identify for the first time The a tomographic image known.

In practical application, three-dimensional reconstruction method specifically includes following in step 104:

The N layer label that C1, input extract target blood region completes image, by what is obtained in each label completion image Pixel coordinate in marked region maps back initial three-dimensional space and obtains the three-dimensional point cloud of target blood.

C2, with Poisson surface reformation algorithm, obtain the three-dimension curved surface of smooth closure, which is to the midpoint C1. cloud Best fit curved surface, while the three-dimension curved surface entity that is included is exactly the objective blood vessel after reconstruct.

The technical solution proposed in the embodiment of the present invention obtains the N layer original image of target blood, and N takes whole greater than 1 Number；Using preset image zooming-out algorithm, extract the target blood in N layer original image in all original images, obtain N layers to Tag image；Based at least one connection label of b layers of image to be marked in N layers of image to be marked, a layers of label to be marked The connected domain of image obtains N layers of label and completes image；Wherein, a and b takes the positive integer less than or equal to N, the difference of a and b Value is 1；Image is completed based on N layers of label, three-dimensionalreconstruction is carried out to target blood, obtains the three-dimensionalreconstruction image of target blood.

Embodiment two

In order to more embody the purpose of the present invention, on the basis of the embodiment of the present invention one, further illustrated Illustrate, the present embodiment is directed to the image data of CTA, can be realized high-precision coronary artery and automatically identifies.

Original image is the CT angiographic image group of human body chest area, and the resolution ratio of every piece image is r₀×c₀ (r in the present embodiment₀=512, c₀=512) 180 original images, are amounted to.The target of the present embodiment is from 180 images, certainly It moves, be accurately partitioned into coronary artery, it is final extract that obtained blood vessel pixel, which is then mapped back initial three-dimensional space, High accuracy three-dimensional coronary artery out.

As shown in figure 4, the method for vessel extraction specifically include it is following:

Step 401: image preprocessing

Here, N layer original image is 180 images that resolution ratio is 512 × 512, and image preprocessing process is will to scheme Useful information leaves as in, the process that extra information is rejected.Wherein, the identification and segmentation of noise on image have very big Interference.The specific settlement steps to deal of problems is as follows in the present invention:

The normalization of step 1.1 image grayscale；It is (outstanding to eliminate imaging factors with this using the processing mode of image normalization It is noise) influence to image grayscale, reduces the difference between different images in property tissue gray scale of the same race, makes grayscale information The feature stable as one, specific as follows:

Wherein, x, y are respectively the gray value of image before normalizing, after normalization, x_max、x_minRespectively original image gray scale Maximum value and minimum value.

Step 1.2 further removes noise using bilateral filtering method.Bilateral filtering is a kind of nonlinear filtering algorithm, it There is preferable effect in Medical Image Processing, retains histoorgan while it can make image smooth-out as far as possible Marginal information.For arbitrary image, distance between pixels similitude and grey similarity in two-sided filter based on image f plus Power combination output are as follows:

Weight coefficient w (i_f,j_f,k_f,n_f) depend on domain core d and codomain core r product:

Wherein, i_f,j_f,k_f,n_fIndicate neighborhood territory pixel position coordinates, σ in the present embodiment_d=10, σ_r=100.

Fig. 5 is the schematic diagram before the image preprocessing in the embodiment of the present invention one；Fig. 6 is the figure in the embodiment of the present invention one As pretreated schematic diagram.It can be seen that after image preprocessing from the comparison of Fig. 5 and Fig. 6, preferably remain image Marginal information.

Step 402: region segmentation

Region segmentation process be by image specific organization or organ differentiate, detection and isolated process.In this hair It is bright middle region segmentation to be carried out using FCM algorithm.FCM algorithm has a wide range of applications in image segmentation, it is a kind of soft poly- Class algorithm indicates sample point to the subjection degree of each class with subordinated-degree matrix.The main thought of FCM algorithm is to pass through cluster The continuous renewal of center and subordinated-degree matrix realizes that the foundation of data clusters is most to achieve the purpose that minimize target function value Big degree of membership principle, the Clustering Effect when target function value minimum are best.Specifically, the objective function of FCM clustering algorithm are as follows:

The constraint condition for needing to meet are as follows:

Wherein, U=(u_il) it is n × k matrix, it is called subordinated-degree matrix, u_ilIt is that i-th of sample point is under the jurisdiction of l class Be subordinate to angle value.d_il=| | x_i-v_l| | it is Euclidean distance of i-th of sample point to first of cluster centre.M is fuzzy factor and m ∈[1,∞)；The general default value of m is 2.

When solving, u is solved by construction Lagrangian_ilAnd v_l, solution formula are as follows:

Above formula is reused, cluster centre v is updated_lWith the degree of membership u of each pixel_il, until meeting | | u_il ^t-u_il ^t+1|| Less than one threshold value of the degree of membership error that < ε, i.e. twice adjacent calculation are obtained, algorithmic statement stop iteration, complete image-region Segmentation.

It is l=2, the initial subordinated-degree matrix of random initializtion that number is clustered in the present embodiment.Reach background by FCM algorithm Isolated purpose, we are removed non-targeted organ-tissue (such as vertebra) by the way of retaining largest connected domain later, are retained Target area.

FCM clusters the result binary map of background separation as shown in fig. 7, region segmentation result figure after retaining largest connected domain As shown in Figure 8.

Step 403: blood vessel enhancing

Vessel enhancement procedure is enhanced the blood vessel structure in image so as to blood vessel structure and other organ-tissue knots Structure is preferably distinguished, and the quality of blood vessel reinforcing effect directly influences the result of final vessel extraction.The present invention uses 3D- Frangi filtering enhances blood vessel.

Frangi filter is a kind of filter based on Hessian matrix, it is mainly used to extract linear structure Object, mainly by between each characteristic value of Hessian matrix and its relationship of corresponding linear goal completes similitude The construction of function, and then extract linear structure.According to the definition of Frangi cast filter function, to each shop of 3 d image data Hessian matrix carries out Eigenvalues analysis, can get characteristic value information | λ₁|≤|λ₂|≤|λ₃|.Given constant α, β, c and ruler Spend corrected parameter γ, and limit scale range [σ_min,σ_max], calculate each picture point shape indexObtain final multiple dimensioned cast likelihood function, expression are as follows:

It is as shown in Figure 9 that blood vessel enhances result.

Step 404: target blood is extracted

Objective extraction is split to ROI region in image, the process of extraction.The present invention carries out mesh using FCM algorithm Mark extracts, and for the enhanced blood-vessel image obtained in step 3, carries out vessel extraction, the present embodiment using FCM algorithm again Middle cluster number is l=2, the initial subordinated-degree matrix of random initializtion.

The results are shown in Figure 10 for vessel extraction, and image shown in Figure 10 is image to be marked.

Step 405: space interlayer connection label

By the above process, several potential blood vessels are extracted from original image, it can by space interlayer connection label Target blood is extracted in the blood vessel from latent.The present invention is based on the connections of space interlayer to mark, and proposes a kind of improved method, i.e., By the layer dependencies using volumetric pixel, the characteristics of combining target blood vessel mark position, successively label is carried out repeatedly, it is final quasi- It determines position destination organization, effectively extracts target blood.Target blood is coronary artery in the present embodiment.Detailed process step It is as follows:

Step 5.1 inputs the image after vessel extraction, and label is using the 1st layer as initial markers layer for the first time.To the 1st layer to Tag image carries out two-dimentional connection (4- connection) and marks, and remembers that first tag image is Label₁ ¹, connected domain number on the image For L¹ _num, it is that each connected domain carries out nonzero integer label according to connected domain number.

Step 5.2 carries out two dimension connection label to a layers of image to be marked, and note tag image is Label_a ¹, by each connection Field mark is 0.

Step 5.3 is to Label_b ^pAnd Label_a ^pCarry out space interlayer connection label:

Iii) work as Label_b ^pIt is middle mark be target connected domain not with Label_a ^pIn any connected domain overlapping when, then use L^p _num=L^p _num+ 1 integer value marks target connected domain, wherein p=1.Here, when occurring new connected domain in image, even Logical domain number L^p _numFrom adding 1, updated L^p _numAs new connection mark.

Step 5.4 enables b=a-1, wherein a=2:180, repeats step 5.2 and step 5.3, until by all original images Label is completed.

For second of step 5.5 label using the 180th layer as initial markers layer, note tag image is Label₁₈₀ ²= Label₁₈₀ ¹, the connected domain number on the image is L² _num。

Step 5.6 enables b=a+1, wherein a=179:1, repeats step 5.2 and step 5.3, until by all original images Label is completed.

Two kinds of label L abel of the more each image of step 5.7_i ¹And Label_i ², retain wherein same tag region, Middle i=1:180.

Step 406: three-dimensionalreconstruction

The N layer label that step 6.1 input extracts target blood region completes image, and each label is completed to obtain in image The pixel coordinate in marked region obtained maps back initial three-dimensional space and obtains the three-dimensional point cloud of target blood.

Step 6.2 Poisson surface reformation algorithm, obtains the three-dimension curved surface of smooth closure, which is in 6.1 The best fit curved surface of three-dimensional point cloud, while the three-dimension curved surface entity that is included is exactly the objective blood vessel after reconstruct.Three The three-dimensional coronary result that dimension reconstruct obtains is as shown in figure 11.

Technical solution provided in an embodiment of the present invention carries out gray scale normalization processing to original image first, reuses Bilateral filtering method carries out Fuzzy Processing to image, then by the FCM algorithm separating background of fixed cluster classification number, then protects Largest connected domain is stayed, to remove unwanted tissue；Then blood vessel enhancing is carried out using multiple dimensioned 3D-Frangi filtering；Later Vessel extraction is carried out again by the FCM algorithm of fixed cluster classification number；Label, which is connected to, followed by space interlayer realizes doctor The Target Segmentation for learning image, obtains the three-dimensional point cloud of target blood；Poisson surface reformation algorithm is finally utilized, is obtained three-dimensional bent Face, the entity which is included are exactly the three-dimensionalreconstruction body of target blood.

Embodiment three

In order to more embody the purpose of the present invention, on the basis of the embodiment of the present invention one, further illustrated Illustrate, the present embodiment is directed to the image data of MRI, can be realized high-precision coronary artery and automatically identifies.

Original image is 30 sagittal sections faces in a heart beat cycle (totally 25 moment) in human heart region Serial NMR imaging image group, the resolution ratio of every piece image are r₀×c₀(r in the present embodiment₀=256, c₀=256), always Count 25 × 30=750 images.The target of the present embodiment is an any given moment (t from 25 moment_iMoment), essence True Ground Split goes out the arteries in the nuclear-magnetism image at the moment, then by the arteries on all 30 width images at the moment It extracts, lays equal stress on and be configured to a high accuracy three-dimensional arteries.

Step 401: image preprocessing

Here, N layer original image be resolution ratio be 256 × 256 30 original images, image preprocessing process be by Useful information leaves in image, the process that extra information is rejected.Wherein, the identification and segmentation of noise on image have very big Interference.The specific settlement steps to deal of problems is as follows in the present invention:

Figure 12 is the schematic diagram before the image preprocessing in the embodiment of the present invention two；Figure 13 is in the embodiment of the present invention two Schematic diagram after image preprocessing.It can be seen that after image preprocessing from the comparison of Figure 12 and Figure 13, preferably remain The marginal information of image.

Step 402: region segmentation

The constraint condition for needing to meet are as follows:

The result binary map that FCM clusters background separation is as shown in figure 14, the region segmentation result after retaining largest connected domain Figure is as shown in figure 15.

Step 403: blood vessel enhancing

It is as shown in figure 16 that blood vessel enhances result.

Step 404: target blood is extracted

Vessel extraction result is as shown in figure 17, and image shown in Figure 17 is image to be marked.

Step 405: space interlayer connection label

Step 5.4 enables b=a-1, wherein a=2:30, repeats step 5.2 and step 5.3, until by all original image marks Note is completed.

For second of step 5.5 label using the 30th layer as initial markers layer, note tag image is Label₃₀ ²=Label₃₀ ¹, Connected domain number on the image is L² _num。

Step 5.6 enables b=a+1, wherein a=29:1, repeats step 5.2 and step 5.3, until by all original image marks Note is completed.

Two kinds of label L abel of the more each image of step 5.7_i ¹And Label_i ², retain wherein same tag region, Middle i=1:30.

Step 406: three-dimensionalreconstruction

The N layer label that step 6.1 input extracts target blood region completes image, and each N layers of label is completed image Pixel coordinate in the marked region of middle acquisition maps back initial three-dimensional space and obtains the three-dimensional point cloud of target blood.

Step 6.2 Poisson surface reformation algorithm, obtains the three-dimension curved surface of smooth closure, which is in 6.1 The best fit curved surface of three-dimensional point cloud, while the three-dimension curved surface entity that is included is exactly the objective blood vessel after reconstruct.Three The three-dimensional coronary result that dimension reconstruct obtains is as shown in figure 18.

The advantages of vessel extraction method proposed by the present invention, is:

1. entire vessel extraction process is fully automated, manual intervention is not needed, the interference of human factor is eliminated；

2. the mode of fixed cluster number repeatedly clustered can gradually extract interested, exclusion noise jamming, removal back Scape partial information；

3.3D-Frangi filtering can effectively enhance target vascular structure, inhibit non-targeted blood vessel structure, be subsequent place Reason provides a good basis；

4. interlayer connection label in space utilizes the layer dependencies of volumetric pixel, the characteristics of combining target blood vessel mark position, Repeatedly successively mark, destination organization is finally accurately positioned, extracts target blood.

Example IV

Based on the same inventive concept, the embodiment of the invention also provides a kind of vessel extraction devices.Figure 19 is that the present invention is real The composed structure schematic diagram of a medium vessels extraction element is applied, as shown in figure 19, which includes: processor 191 With memory 192, wherein

Processor 191 is for executing the vessel extraction program stored in memory 192, to perform the steps of

Using preset image zooming-out algorithm, the target blood in N layer original image in all original images is extracted, is obtained N layers of image to be marked；

Based at least one connection label of b layers of image to be marked in N layers of image to be marked, a layers of label to be marked The connected domain of image obtains N layers of label and completes image；Wherein, a and b takes the positive integer less than or equal to N, the difference of a and b Value is 1；

Image is completed based on N layers of label, three-dimensionalreconstruction is carried out to target blood, obtains the Three-dimensional Gravity composition of target blood Picture.

In some embodiments, processor 191 is specifically used for executing the vessel extraction program stored in memory 192, with It performs the steps of using the first labelling strategies, based at least one connection label of b layers of image to be marked, the first deutero-albumose Remember the connected domain of a layers of image to be marked, obtains the N tomographic image marked for the first time；Using the second labelling strategies, it is based on b layers At least one connection label of image to be marked, the connected domain of a layers of image to be marked of second of label obtain second of label N tomographic image；The N tomographic image of N tomographic image and second of label based on first time label, obtains N layers of label and completes image.

In some embodiments, the first labelling strategies are successively to mark from top to bottom, wherein a takes the positive integer greater than 1, b For a-1；Second labelling strategies are successively to mark from bottom to top, wherein a takes the positive integer less than N, b a+1.

In some embodiments, before using the first labelling strategies, processor 191 is also used to execute in memory 192 The vessel extraction program of storage carries out the 1st layer of image to be marked with performing the steps of using the 1st layer as initial markers layer Two dimension connection label obtains at least one connection label of the 1st layer of image to be marked；Will in N layers of image to be marked except the 1st layer it The connection label of outer remainder layer image to be marked is set to 0.

In some embodiments, before using the second labelling strategies, processor 191 is also used to execute in memory 192 The vessel extraction program of storage, to perform the steps of using n-th layer as initial markers layer, by the N tomographic image of first time label The connection of middle n-th layer image marks, the connection label of n-th layer image to be marked when marking as second；By N layers of figure to be marked The connection label of remainder layer image to be marked as in addition to n-th layer is set to 0.

In some embodiments, processor 191 is specifically used for executing the vessel extraction program stored in memory 192, with It performs the steps of and relatively marks the connection of the i-th obtained tomographic image to mark for the first time and i-th layer obtained is marked to scheme for the second time The connection of picture marks, and retains the connected domain with identical connection label, obtains i-th layer of label and completes image；Wherein, i, which takes, is less than Or the positive integer equal to N.

In some embodiments, processor 191 is specifically used for executing the vessel extraction program stored in memory 192, with It performs the steps of when at least one connected domain in a layers of image to be marked and the single connected domain in b layers of image to be marked When overlapping, the connection of connected domain single in b layers of image to be marked is marked, as at least one in a layers of image to be marked The connection of connected domain marks；When connected domain single in a layers of image to be marked and multiple connected domains in b layers of image to be marked When overlapping, by b layers of image to be marked with the maximum connected domain of connected domain overlapping area single in a layers of image to be marked Connection label, the connection as the single connected domain in a layers of image to be marked mark；When in a layers of image to be marked at least One connected domain with it is Chong Die without connected domain in b layers of image to be marked when, obtain at least one new connection mark；To at least one A new connection mark is marked respectively as the connection of at least one connected domain in a layers of image to be marked.

In some embodiments, processor 191 is specifically used for executing the vessel extraction program stored in memory 192, with It performs the steps of and original image is pre-processed, the image that obtains that treated；Based on the first clustering algorithm, segmentation portion Image after reason obtains region of interest ROI region；Target blood in ROI region is enhanced, enhanced figure is obtained Picture；Based on second of clustering algorithm, the target blood in enhanced image is extracted, obtains image to be marked.

In some embodiments, processor 191 is specifically used for executing the vessel extraction program stored in memory 192, with It performs the steps of and gray scale normalization processing is carried out to original image, the image after being normalized；To the image after normalization Fuzzy processing is carried out, the image that obtains that treated.

In practical applications, above-mentioned memory can be volatile memory (volatile memory), such as deposit at random Access to memory (RAM, Random-Access Memory)；Or nonvolatile memory (non-volatile memory), example 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 provides instruction and data.

Above-mentioned processor can be application-specific IC (ASIC, Application Specific Integrated Circuit), digital signal processing device (DSPD, Digital Signal Processing Device), programmable logic dress Set (PLD, Programmable Logic Device), field programmable gate array (Field-Programmable Gate Array, FPGA), controller, at least one of microcontroller, microprocessor.It is to be appreciated that being used for different equipment In realize the electronic device of above-mentioned processor function can also be it is other, the embodiment of the present invention is not especially limited.

Embodiment five

Based on the same inventive concept, the embodiment of the invention 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 of vessel extraction device, to complete aforementioned one Or the method and step in more embodiments.

It should be understood by those skilled in the art that, the embodiment of the present invention 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 present invention Formula.Moreover, the present invention, 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 present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product 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.

More than, only presently preferred embodiments of the present invention is not intended to limit the scope of the present invention.

Claims

1. a kind of method of vessel extraction, which is characterized in that the described method includes:

Using preset image zooming-out algorithm, the target blood in the N layer original image in all original images is extracted, is obtained N layers of image to be marked；

Based at least one connection label of b layers of image to be marked in the N layers of image to be marked, a layers of label to be marked The connected domain of image obtains N layers of label and completes image；Wherein, a and b takes the positive integer less than or equal to N, the difference of a and b Value is 1；

Image is completed based on N layers of label, three-dimensionalreconstruction is carried out to the target blood, obtains the three-dimensionalreconstruction of the target blood Image.

2. the method according to claim 1, wherein it is described based on b layers in the N layers of image to be marked wait mark Remember at least one connection label of image, the connected domain of a layers of image to be marked of label obtains N layers of label and completes image, packet It includes:

Using the first labelling strategies, based at least one connection label of b layers of image to be marked, a layers of first time label to The connected domain of tag image obtains the N tomographic image marked for the first time；

Using the second labelling strategies, based at least one connection label of b layers of image to be marked, a layers of second label to The connected domain of tag image obtains the N tomographic image of second of label；

The N tomographic image of N tomographic image and second of label based on first time label, obtains N layers of label and completes image.

3. according to the method described in claim 2, it is characterized in that, first labelling strategies be from top to bottom successively label, Wherein, a takes the positive integer greater than 1, b a-1；

Second labelling strategies are successively to mark from bottom to top, wherein a takes the positive integer less than N, b a+1.

4. according to the method described in claim 3, it is characterized in that, using before first labelling strategies, the method Further include: using the 1st layer as initial markers layer, two dimension connection label is carried out to the 1st layer of image to be marked, obtains the 1st layer wait mark Remember at least one connection label of image；

The connection label of remainder layer image to be marked in the N layers of image to be marked in addition to the 1st layer is set to 0.

5. according to the method described in claim 3, it is characterized in that, using before second labelling strategies, the method Further include:

Using n-th layer as initial markers layer, the connection label of n-th layer image in N tomographic image that the first time is marked, as The connection label of n-th layer image to be marked when second of label；

The connection label of remainder layer image to be marked in the N layers of image to be marked in addition to n-th layer is set to 0.

6. according to the method described in claim 2, it is characterized in that, the N tomographic image and institute based on first time label The N tomographic image for stating second of label obtains N layers of label and completes image, comprising:

The connection of the i-th obtained tomographic image is relatively marked to mark and mark being connected to for the i-th obtained tomographic image for the second time for the first time Label retains the connected domain with identical connection label, obtains i-th layer of label and completes image；Wherein, i takes less than or equal to N Positive integer.

7. the method according to claim 1, wherein it is described based on b layers in the N layers of image to be marked wait mark Remember at least one connection label of image, the connected domain of a layers of image to be marked of label, comprising:

It, will when at least one connected domain is Chong Die with the single connected domain in b layers of image to be marked in a layers of image to be marked The connection label of single connected domain, the company as at least one connected domain in a layers of image to be marked in b layers of image to be marked Logical label；

When connected domain single in a layers of image to be marked is Chong Die with multiple connected domains in b layers of image to be marked, by b It marks, makees with the connection of the maximum connected domain of connected domain overlapping area single in a layers of image to be marked in layer image to be marked It is marked for the connection of the single connected domain in a layers of image to be marked；

When in a layers of image to be marked at least one connected domain with it is Chong Die without connected domain in b layers of image to be marked when, obtain extremely Few one new connection mark；In using at least one described new connection mark as a layers of image to be marked at least The connection label of one connected domain.

8. the method according to claim 1, wherein the preset image zooming-out algorithm includes:

Original image is pre-processed, the image that obtains that treated；

Based on the first clustering algorithm, segmentation is described treated image obtains region of interest ROI region；

Target blood in the ROI region is enhanced, enhanced image is obtained；

Based on second of clustering algorithm, the target blood in the enhanced image is extracted, image to be marked is obtained.

9. according to the method described in claim 8, it is characterized in that, described pre-process original image, after obtaining processing Image, comprising:

Gray scale normalization processing is carried out to original image, the image after being normalized；

Fuzzy processing is carried out to the image after the normalization, the image that obtains that treated.

10. a kind of vessel extraction device, which is characterized in that described device includes: processor and memory；Wherein,

11. device according to claim 10, which is characterized in that the processor is specifically used for executing to be stored in memory Vessel extraction program, to perform the steps of

12. device according to claim 11, which is characterized in that first labelling strategies are successively to mark from top to bottom Note, wherein a takes the positive integer greater than 1, b a-1；

13. device according to claim 12, which is characterized in that before using first labelling strategies, the place Reason device is also used to execute the vessel extraction program stored in memory, to perform the steps of

Using the 1st layer as initial markers layer, two dimension connection label is carried out to the 1st layer of image to be marked, obtains the 1st layer of figure to be marked At least one connection label of picture；

14. device according to claim 12, which is characterized in that before using second labelling strategies, the place Reason device is also used to execute the vessel extraction program stored in memory, to perform the steps of

15. device according to claim 11, which is characterized in that the processor is specifically used for executing to be stored in memory Vessel extraction program, to perform the steps of

16. device according to claim 10, which is characterized in that the processor is specifically used for executing to be stored in memory Vessel extraction program, to perform the steps of

17. device according to claim 10, which is characterized in that the processor is specifically used for executing to be stored in memory Vessel extraction program, to perform the steps of

Original image is pre-processed, the image that obtains that treated；

Target blood in the ROI region is enhanced, enhanced image is obtained；

18. device according to claim 17, which is characterized in that the processor is specifically used for executing to be stored in memory Vessel extraction program, to perform the steps of

19. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program quilt The step of claim 1 to 9 described in any item methods are realized when processor executes.