WO2018074661A1

WO2018074661A1 - Coronary artery blood vessel subtraction device and method using vessel correspondence optimization

Info

Publication number: WO2018074661A1
Application number: PCT/KR2016/014881
Authority: WO
Inventors: 이수찬; 신승연; 노경진
Original assignee: 순천향대학교 산학협력단
Priority date: 2016-10-19
Filing date: 2016-12-19
Publication date: 2018-04-26
Also published as: KR101852689B1

Abstract

The present invention relates to a coronary artery blood vessel subtraction device and method using blood vessel angiography and, more particularly, to a coronary artery blood vessel subtraction device and method using vessel correspondence optimization (VCO), which can precisely segment, from a fluorographic X-ray image frame, the center line of a blood vessel as well as blood vessel branches that branch from or intersect at the center line of the blood vessel by applying vessel correspondence optimization for applying a Markov random field (MRF).

Description

Coronary blood vessel extraction device and method using vascular response optimization

The present invention relates to an apparatus and method for extracting coronary blood vessels using angiography, and more particularly, Vessel Correspondence Optimization to which Markov Random Field (MRF) is applied from a fluoroscopic X-ray image frame. The present invention relates to an apparatus and method for extracting coronary blood vessels using vascular optimization, which can accurately extract blood vessel branches that branch and cross from a blood vessel center line by applying a VCO.

In general, one or more of fluoroscopic X-ray (XRA) and computed tomography (CT) angiography are performed to check for lesions of the heart.

Fluoroscopy x-ray angiography is used to assess stenosis in coronary arteries and to provide guidance for percutaneous coronary interference.

In this case, blood vessels extracted by computed tomography (CTA) may be matched to blood vessels extracted by fluoroscopic X-ray angiography to detect blood vessels.

This can visualize invisible arteries due to blockage of the contrast agent in chronic total occlusion.

The lesion of the heart is determined from image frames generated by registration of image frames including cardiovascular vessels obtained by fluoroscopic X-ray and computed tomography angiography.

Most conventional coronary artery extraction devices have been developed with the aim of improving the resolution of single images and sophisticated optimization methods to determine heart lesions from images. In addition, the conventional coronary blood vessel extraction apparatus focuses on extracting blood vessels from a single image without considering continuity as described above.

As described above, there was a problem of providing inconsistent results by detecting blood vessels from a single image which does not consider continuity when detecting blood vessels of the conventional coronary arteries, and thus there is a problem in that the lesions of the coronary arteries cannot be accurately diagnosed.

Accordingly, an object of the present invention is to apply Vessel Correspondence Optimization (VCO) to which Markov Random Field (MRF) is applied from a fluoroscopic X-ray image frame. An object of the present invention is to provide an apparatus and method for extracting coronary blood vessels using vascular response optimization capable of accurately extracting branching and crossing vascular branches.

Coronary artery vessel extraction apparatus using the vascular response optimization according to the present invention for achieving the above object: Blood vessel image acquisition unit for outputting a blood vessel image frame sequence for coronary artery by taking a coronary artery; And a consistent centerline reflecting the movement of blood vessels by receiving the vessel image frame sequence, sampling the input vessel image frame sequence, and performing vessel response optimization by applying Markov random field optimization to the sampled vessel image frame sequence. And a blood vessel extracting unit for extracting and displaying coronary blood vessels matching the localized blood vessels further visualized by administration of a contrast agent.

The blood vessel extracting unit performs global stiffness by performing chamfer matching according to a comparison between a source frame including a coronary centerline and an object frame, which is the sampled blood vessel image frame, to estimate global shape and translational motion of the coronary artery of the heart. Matching part; Local non-rigid registration that detects candidate (vascular) branches along the centerline by a window of constant size with at least two keypoints, determines actual ones of the detected candidate branches, and restores the vascular centerline by contacting the determined branches with the centerline part; New blood vessel branches are detected from the object frame reflecting the contrast medium flow image by the contrast medium until the blood vessels longer than the maximum radius of the blood vessel of the source frame are not detected, and the detected blood vessel branches are connected to the center line and not connected to the center line. Post-processing unit for extracting the entire coronary artery except for the eggplant; And a display unit for visualizing and displaying on the display means a coronary artery including a centerline to which the blood vessel branch is connected.

The blood vessel extracting unit may further include a vascular structure analyzing unit configured to analyze branch points at which branches branch and cross points intersecting branches at the centerline of the entire coronary artery, and display analysis information through the display unit. do.

The source frame further includes a vessel point, and the local non-rigid registration unit includes a vessel point sampling unit for sampling vessel points of the centerline by vessel points of the source frame; A blood vessel feature point extracting unit configured to extract a characteristic indicating whether each of the sampled sampling vessel points is a branch point, an intersection point, or an end point; A blood vessel corresponding point candidate detector searching for a corresponding vessel point corresponding to the vessel point of the source frame from an object frame; An MRF optimizer for defining a window, which is a local search region, for the vessel feature point and corresponding vessel points, and performing vessel point search for detecting optimal correspondence point candidates that are optimal correspondence points in the window; And a blood vessel centerline restoration unit for restoring the blood vessel centerline by matching the branches at the optimum corresponding point.

The MRF optimizer detects an optimal correspondence point according to Equation 2 below.

[Equation 2]

And

Denotes a source frame and a target frame, respectively, p _i and

Represents the coordinates of the i-th node of the source blood vessel structure and the xi-th corresponding candidate in the object frame, respectively. D is a function for the local feature descriptor.

Coronary artery extraction method using vascular response optimization according to the present invention for achieving the above object comprises: taking a coronary artery to obtain a blood vessel image frame step for outputting a vascular image frame sequence for the coronary artery; And receiving the blood vessel image frame sequence, sampling the input blood vessel image frame sequence, performing blood vessel response optimization on the sampled blood vessel image frame sequence, extracting blood vessels including a blood vessel centerline, and performing Markov random field optimization. And a blood vessel extraction step of extracting and displaying a consistent local blood vessel reflecting the movement, and extracting and displaying a coronary blood vessel matching the extracted blood vessel centerline and the local blood vessel.

In the blood vessel extraction step, a global shape and translational motion of the coronary artery of the heart is estimated by performing a chamfer matching according to a comparison between the source frame including the coronary centerline and the object frame, which is the sampled blood vessel image frame. A global rigid matching step; Local non-rigid registration that detects candidate (vascular) branches along the centerline by a window of constant size with at least two keypoints, determines actual ones of the detected candidate branches, and restores the vascular centerline by contacting the determined branches with the centerline step; Reflect the contrast flow image by contrast injection, detect new vessel branches until no vessel longer than the radius of the maximizing vessel is detected, connect the detected vessel branches to the centerline, and exclude the branches not connected to the centerline Post-treatment step of extracting the coronary artery; And a display step of visualizing and displaying on a display means a coronary artery including a centerline to which the blood vessel branch is connected.

The blood vessel extracting step may further include a blood vessel structure analyzing step of analyzing a branching point at which a branch branches and an intersection point at which a branch intersects at the center line of the extracted whole coronary artery, and displaying analysis information through a display unit. It is done.

The source frame further includes a vessel point, and the local non-rigid registration step includes: a vessel point sampling step of sampling a vessel point of a centerline by the vessel point of the source frame; A vein feature point extraction step of retrieving feature point branches of feature points having branching, crossing, and end point characteristics for each of the sampled sampling vessel points; A vessel corresponding point candidate detection step of searching for a corresponding vessel point for the sampling vessel point; An MRF optimization step of defining a window that is a local search region for the vessel feature point and corresponding vessel points and performing vessel point search to detect optimal correspondence point candidates that are optimal correspondence points within the window; And restoring a blood vessel centerline by matching the branches at the optimum corresponding point.

The MRF optimization step is characterized by detecting the best corresponding point by the following equation (2).

[Equation 2]

And

Denotes a source frame and a target frame, respectively, p _i and

The present invention has the effect of estimating more precise movements by overcoming an aperture problem and extracting an optimum corresponding point through vascular optimization optimization.

In addition, the present invention can detect an error generated when the three-dimensional structure is projected to the two-dimensional through the process of determining the branch point, the intersection point, etc. has the effect of more accurate motion estimation.

1 is a view showing a schematic configuration of a coronary blood vessel extraction apparatus using vascular response optimization according to the present invention.

Figure 2 is a view showing the detailed configuration of the blood vessel extraction unit of the coronary vessel extraction apparatus according to the present invention.

Figure 3 is a view showing the detailed configuration of the local non-rigid registration part of the blood vessel extraction unit according to the present invention.

4 is a diagram illustrating a sampled X-ray angiography sequence according to an embodiment of the present invention.

5 is a diagram showing an overall framework for explaining an example of blood vessel extraction according to an embodiment of the present invention.

6 is a view for explaining the effect of the hierarchical branch point search method according to an embodiment of the present invention showing the blood vessel branches of the source frame.

FIG. 7 is a view for explaining the effect of the hierarchical branch point searching method according to an embodiment of the present invention.

Hereinafter, with reference to the accompanying drawings will be described the configuration and operation of the coronary vessel extraction apparatus according to the present invention, the blood vessel extraction method in the device will be described.

Referring to FIG. 1, the coronary blood vessel extraction apparatus using vascular optimization according to the present invention includes a blood vessel image acquisition unit 100 and a blood vessel extraction unit 200.

The blood vessel image acquisition unit 100 photographs the heart using a fluoroscopic X-ray imaging apparatus (not shown), a computed tomography (CT) apparatus, and the like to output a coronary blood vessel image frame sequence. The blood vessel image acquisition unit 100 may photograph the heart after the contrast medium is added, and thus outputs the coronary blood vessel image frame sequence reflecting the contrast medium input.

The blood vessel extraction unit 200 receives a coronary vessel image frame sequence output from the blood vessel image acquisition unit 100 as a target frame sequence, and a vessel to which a Markov Random Field (MRF) according to the present invention is applied. Corresponding optimization is applied to precisely extract the vessel's vascular centerline and branches (or "vascular branches") that branch and intersect at the vessel centerline, and match the centerline and branches of the extracted vessel to extract the entire coronary vessel Display on the display means.

2 is a view showing a detailed configuration of the blood vessel extraction unit of the coronary vessel extraction apparatus according to the present invention, Figure 4 is a view showing an X-ray imaging sequence sampled according to an embodiment of the present invention. A description with reference to FIGS. 2 and 4 is as follows. Reference numeral 1 in FIG. 4 is a blood vessel centerline, 2 is a blood vessel branch, 3 is a branch point, and 4 is an intersection point.

The blood vessel extracting unit 200 includes a global rigid matching unit 300, a local non-rigid matching unit 400, a post processing unit 500, and a display unit 700, and according to an embodiment, the vascular structure analyzing unit 600. ) May be further included.

The global rigid matching unit 300 stores a source frame including a chamfer, a coronary vessel centerline 1, and a vessel point, and coronary vessel image as a target frame from the vessel image acquisition unit 100. The frame sequence is input and sampled to a certain number. 4 shows only some of the object frames selected so that the dynamics among the object frames sampled by the global rigid matching unit 300 can be clearly indicated.

The global rigid matching unit 300 compares the sampled object frame with the source frame and performs chamfer matching to estimate the global shape and translational movement of the coronary artery of the heart according to the change of heart rate, breath, or vision. do. The estimation of the translational motion may be estimated according to one or more of the chamfer of the source frame and the centerline of the coronary vessel and the direction and extent of the chamfer and the centerline of the object frame.

The local non-rigid matching unit 400 detects candidate (vascular) branches along the centerline by a window having a predetermined size having at least two keypoints, determines an actual branch among the detected candidate branches, and matches the determined branches to the centerline. To restore coronary vessels. The window is a local search region and has a size of w _K × h _K. Detailed description of the window will be described later.

The post processor 500 detects new blood vessel branches until the blood vessel longer than the radius of the maximized vessel is not detected from the object frame reflecting the contrast medium flow image by the contrast agent input, connects the detected blood vessel branches to the centerline, and connects the centerline. Extract the entire coronary artery except for the branches that are not connected to it. In other words, the post-processing unit extracts and visualizes local blood vessel branches newly appearing as a contrast agent is added.

The post-processing unit 500 connects the obtained vessel points after the VCO in the regional non-rigid matching unit 400 to construct the vessel centerline 1 structure by a high speed marching method. The fast marching scheme is described in Yatziv, L., Bartesaghi, A., Sapiro, G .: O (N) Implementation of the Fast Marching Algorithm. As detailed in the Journal of Computational Physics 212 (2), 393-399 (2005), its detailed description is omitted.

The post-processing unit 500 combines Vesselness and high speed marching to expand the vessel centerline 1 to include vessels that are additionally visible due to the administration of contrast medium. Vesselness is calculated by Frangi, A.F., Niessen, W.J., Vincken, K.L., Viergever, M.A .: Multiscale Vessel Enhancement Filtering. In: Wells, W. M., Colchester, A., Delp, S. (eds.) MICCAI 1998. LNCS, vol. 1496, pp. 130-137. Springer, Heidelberg (1998) is described in detail, so the detailed description is omitted.

Binary segment masks are obtained by thresholding vesselness and exclude areas that are not connected to the vessel centerline 1.

The distance transformation DT of the target shape is calculated with the binary segment mask having a non-vascular region as the seed. Based on the binary segment mask and the DT, the following is repeated until no branches longer than the maximal vessel radius are found.

i) DT value as velocity matching and as seed

Take a quick match

ii)

Have a recent arrival time in

Found the shortest path from the pixel to be added to. The method of finding the shortest path in the pixel is Van Uitert, R., Bitter, I .: Subvoxel precise skeletons of volumetric data based on fast marching methods. Medical physics 34 (2) and 627-638 (2007) are described in detail, and thus description thereof is omitted.

The display unit 700 visualizes the coronary artery including the centerline to which the blood vessel branch is connected, and displays the coronary artery on the display means.

The vascular structure analyzer 600 analyzes the branch point where the branch branches and the intersection point where the branches intersect in the extracted centerline of the entire coronary artery, and displays the analysis information on the display unit 700.

3 is a view showing a detailed configuration of the local non-rigid matching portion of the blood vessel extraction unit according to the present invention, Figure 5 is a view showing a whole framework for explaining an example of the blood vessel extraction according to an embodiment of the present invention, Figure 6 is a view for explaining the effect of the hierarchical branch point search method according to an embodiment of the present invention, showing the blood vessel branches of the source frame, Figure 7 is a hierarchical branch point search according to an embodiment of the present invention A diagram for describing the effect of the method is a diagram for explaining a method for retrieving blood vessel branch points in an object frame. Hereinafter, a description will be given with reference to FIGS. 3 and 5 to 7.

The regional non-stiff matching unit 400 of the present invention includes a blood vessel point sampling unit 410, a blood vessel feature point extractor 420, a blood vessel correspondence point candidate detector 430, an MRF optimizer 440, and a blood vessel centerline restoration unit 450. ).

The vessel point sampling unit 410 is a blood vessel extracted when the vessel centerline (1-2: green) is extracted from the target frame based on the vessel centerline (1-1: gray) of the source frame in the global rigid matching unit 300. The center line 1-2 and the vessel center line 1-1 of the source frame overlap each other, and the vessel points of the vessel center line of the target frame are sampled by the vessel points of the source frame.

The vessel feature point extractor 420 is a branch point for branching of a vessel branch from a vessel centerline, an intersection point at which a vessel centerline intersects a vessel branch, and an end of a vessel centerline and a vessel branch for each of the sampled vessel points. Extract vessel features indicating whether a point.

The vessel corresponding point candidate detector 430 detects a corresponding vessel point corresponding to the vessel point of the source frame from the target frame.

The MRF optimizer 440 defines the window, which is a local search region, for the vessel feature point and the corresponding vessel points, and performs a vessel point search that detects optimal correspondence point candidates that are optimal correspondence points in the window. .

In more detail about MRF, the MRF pair graph is constructed from the vessel centerline of the source frame. The nodes correspond to sample points from the centerline, and the edges represent the points of connection between the vessel points.

MRF energy is defined as in Equation 1 below.

Where x is a vector containing a set of random variables xi at each node with index i. Each xi may be labeled with a different value of Np + 1, and the optimal x is determined by minimizing Equation 1 above. The Np + 1 labeling includes one dummy label assigned to a node when no candidate has a local shape and appearance consistent with the Np corresponding candidates. If an optimal dummy label is found, the node is excluded from the set of VCO points created.

Unary cost function

Depends on the similarity between the local shapes of the i th node. When the corresponding points having similar shapes are found, as shown in Equation 2

Is defined as decreasing.

And

Denotes a source frame and a target frame, respectively, p _i and

Respectively represent the coordinates of the i-th node of the source frame blood vessel structure and the xi-th corresponding candidate in the object frame.

D is a function for the local feature descriptor.

To ensure robustness against outliers

Seems to have been cut by

Outliers can occur when there is no corresponding point due to severe area (region) deformation.

Fairwise Cost

Reinforces similar displacement vectors between neighboring points. This may be defined as in Equation 3.

Where p _i and p _j are the coordinates of the i th and j th sourceframe nodes,

And

Is the coordinate in the object frame of the corresponding candidates of x _i and x _j .

Is the displacement vector of the i th source node. Another cut is the threshold

Included on the basis of

The parameter lambda (

) Controls the amount of normalization in Equation 1 above.

The vessel centerline restoration unit 450 restores the vessel centerline by matching branches at the optimum correspondence point.

As described above, the present invention applies a hierarchical correspondence search method including a chamfer matching global search, a branch search by vascular keypoint correspondence, and a point search. The present invention defines vascular junctions and end points that include intersections and branching points in 3D-2D projections and have intersections and branching points that have a unique shape as vascular keypoints.

Vascular branches provide a line connecting two tubercle keypoints.

The present invention is to distinguish from the general vessel point p _i

Denotes the α th keypoint. The m-th branch is represented by b _m .

Chamfer Matching Global Search : The present invention performs chamfer matching to estimate large global translational motion from heart rate, breathing, or changes in vision.

The sample shape is a collection of blood vessel points of the source frame.

The target shape is configured by sequentially applying blood vessel enhancement, bordering and skeletalization to the object frame.

The present invention minimizes the sum of the distances between each template (sample) point and the target shape by a force force on the distance transformation DT of the target shape and finds a global displacement vector.

5A shows the results of an example of this process.

Branch search by vascular keypoint correspondence : The present invention provides two keypoints of branch b _m in a local search region of size w _K × h _K.

And

Correspondence points for

And

Search for.

Correspondence is determined by local similarity and measured using equation (2).

Suppression that is not maximum can be applied to avoid adjacent matches,

Possible correspondence up to two keypoints

And

Is obtained for.

The set of candidate branches consists of all the displacement vectors for b _m ,

Is generated for b _m by applying.

It should be appreciated that the present invention may be up to Nk * 2 candidate branches and may depend on the number of keypoint correspondences.

The present invention includes a branch with nothing to replace if all keypoint matches caused by Nk * 2 branch candidates are reliable.

Generation of Corresponding Candidates by Vascular Point Search: Vascular Points in Branches (b _m )

), The set of corresponding points based on the candidate branches

to be.

The present invention defines a size w _k × h _k size local search region at these points and determines N _i optimal correspondence points based on Equation (2).

6 and 7 show an example where the corresponding candidate is improved to a smaller local search area based on the previous branch search.

The maximum number of search result candidates is

to be. Due to non-maximal suppression, the actual number of candidates Nc may be less than Np based on the composited image. Therefore, the present invention fixes the label number to Np for the vessel points. However, by assigning an infinite unary cost, it invalidates a label larger than Nc without an actual corresponding candidate.

Referring to FIG. 5 in more detail, FIG. 5 shows the entire framework. The vessel centerline of the object frame is extracted based on the centerline of the source frame that we assume to be given.

5A illustrates global search by chamfer matching as described above, and indicates that the centerline (green) and the tracking vessel centerline (red) of the source frame overlap.

(B) shows blood vessel branch search by keypoint correspondence (white). (C) shows the corresponding candidate (green) by vessel point search, (D) shows the optimal point correspondence (white) from MRF optimization, and (E) shows the extraction of new visual vessel branches.

6 and 7 will be described in more detail, Figure 6 shows an example of the vascular branch in the source frame, Figure 7 (a) shows the vascular branch is established in the target frame, Figure 7 (b) Is the window at the branch vessel point and (c) is the response candidates obtained. In addition, the upper drawings of (a), (b), and (c) show a case where hierarchical search is not performed, and the lower drawings show a case where hierarchical search is performed. Branch alignment helps to reduce the size of the window.

On the other hand, the present invention is not limited to the above-described typical preferred embodiment, but can be carried out in various ways without departing from the gist of the present invention, various modifications, alterations, substitutions or additions in the art Anyone who has this can easily understand it. If the implementation by such improvement, change, replacement or addition falls within the scope of the appended claims, the technical idea should also be regarded as belonging to the present invention.

[Description of the code]

100: blood vessel image acquisition unit 200: blood vessel extraction unit

300: global rigid matching 400: local non-rigid matching

410: blood vessel point sampling unit 420: blood vessel feature point extraction unit

430: blood vessel correspondence point candidate detector 440: MRF optimizer

450: blood vessel centerline restoration unit 500: post-processing unit

600: blood vessel structure analysis unit 700: display unit

Claims

A blood vessel image acquisition unit for photographing the coronary artery and outputting a blood vessel image frame sequence for the coronary artery; And

The vascular image frame sequence is input, the vascular image frame sequence is sampled, and the vascular response optimization by applying Markov random field optimization to the sampled vascular image frame sequence includes a consistent center line reflecting the movement of blood vessels. And a blood vessel extracting unit for extracting and displaying a coronary vessel in which local blood vessels are visually matched by administration of a contrast agent.
The method of claim 1,

The blood vessel extracting unit,

A global stiffness matching unit which estimates global shape and translational motion of the coronary artery of the heart by performing chamfer matching according to a comparison between a source frame including a coronary centerline and an object frame which is the sampled blood vessel image frame;

Local non-rigid registration that detects candidate (vascular) branches along the centerline by a window of constant size with at least two keypoints, determines actual ones of the detected candidate branches, and restores the vascular centerline by contacting the determined branches with the centerline part;

New blood vessel branches are detected from the object frame reflecting the contrast medium flow image by the contrast medium until the blood vessels longer than the maximum radius of the blood vessel of the source frame are not detected, and the detected blood vessel branches are connected to the center line and not connected to the center line. Post-processing unit for extracting the entire coronary artery except for the eggplant; And

And a display unit for visualizing and displaying on the display means a coronary artery including a centerline to which the blood vessel branch is connected.
The method of claim 2,

The blood vessel extracting unit,

Analyzing the vascular response optimization, characterized in that further comprising a vascular structure analysis unit for analyzing the branching point and the branch intersects branch intersected in the extracted center line of the entire coronary artery, and displays the analysis information through the display unit Coronary blood vessel extraction device used.
The method of claim 2,

The source frame further includes a blood vessel point,

The local non-rigid registration part,

A blood vessel point sampling unit configured to sample blood vessel points of a center line by the vessel points of the source frame;

A blood vessel feature point extracting unit configured to extract a characteristic indicating whether each of the sampled sampling vessel points is a branch point, an intersection point, or an end point;

A blood vessel corresponding point candidate detector searching for a corresponding vessel point corresponding to the vessel point of the source frame from an object frame;

An MRF optimizer for defining a window, which is a local search region, for the vessel feature point and corresponding vessel points, and performing vessel point search for detecting optimal correspondence point candidates that are optimal correspondence points in the window; And

Coronary blood vessel extraction apparatus using a vessel corresponding optimization, characterized in that it comprises a blood vessel center line restoration unit for restoring the vessel center line by matching the branches at the optimum corresponding point.
The method of claim 4, wherein

The MRF optimizer,

An apparatus for extracting coronary blood vessels using vascular response optimization, characterized by detecting an optimum correspondence point according to Equation 2 below.

[Equation 2]

And
Denotes a source frame and a target frame, respectively, p i and
Represents the coordinates of the i-th node of the source blood vessel structure and the xi-th corresponding candidate in the object frame, respectively. D is a function for the local feature descriptor.
Taking a coronary artery and acquiring a blood vessel image frame sequence for outputting a vascular image frame sequence for the coronary artery; And

Receiving the blood vessel image frame sequence, sampling the input blood vessel image frame sequence, performing blood vessel response optimization on the sampled blood vessel image frame sequence, extracting blood vessels including a blood vessel centerline, and performing Markov random field optimization. And a blood vessel extraction step of extracting and displaying a consistent local vessel reflecting movement, and extracting and displaying a coronary vessel in which the extracted vessel centerline and the local vessel are matched.
The method of claim 6,

The blood vessel extraction step,

A global stiffness matching step of estimating global shape and translational movement of the coronary artery of the heart by performing chamfer matching according to a comparison between a source frame including a coronary centerline and an object frame that is the sampled vascular image frame;

Local non-rigid registration that detects candidate (vascular) branches along the centerline by a window of constant size with at least two keypoints, determines actual ones of the detected candidate branches, and restores the vascular centerline by contacting the determined branches with the centerline step;

Reflect the contrast flow image by contrast injection, detect new vessel branches until no vessel longer than the radius of the maximizing vessel is detected, connect the detected vessel branches to the centerline, and exclude the branches not connected to the centerline Post-treatment step of extracting the coronary artery; And

And a display step of visualizing and displaying the coronary artery including the centerline to which the blood vessel branch is connected to the display means.
The method of claim 7, wherein

The blood vessel extraction step,

Analyzing the blood vessel response optimization, characterized in that further comprising the step of analyzing the branching point branching and the branching point intersecting the branch in the center line of the entire coronary artery extracted, and displaying the analysis information on the display unit Coronary blood vessel extraction method.
The method of claim 7, wherein

The source frame further includes a blood vessel point,

The regional non-rigid registration step,

A vessel point sampling step of sampling a vessel point of a centerline by the vessel point of the source frame;

A vein feature point extraction step of retrieving feature point branches of feature points having branching, crossing, and end point characteristics for each of the sampled sampling vessel points;

A vessel corresponding point candidate detection step of searching for a corresponding vessel point for the sampling vessel point;

An MRF optimization step of defining a window that is a local search region for the vessel feature point and corresponding vessel points and performing vessel point search to detect optimal correspondence point candidates that are optimal correspondence points within the window; And

And a vessel centerline restoration step of restoring the vessel center line by matching the branches at the optimum correspondence point.
The method of claim 9,

The MRF optimization step,

A method for extracting coronary blood vessels using vascular response optimization, characterized by detecting an optimum correspondence point according to Equation 2 below.

[Equation 2]

And
Denotes a source frame and a target frame, respectively, p i and
Represents the coordinates of the i-th node of the source blood vessel structure and the xi-th corresponding candidate in the object frame, respectively. D is a function for the local feature descriptor.