WO2019242227A1 - Automatic registration method for coronary arteries - Google Patents

Automatic registration method for coronary arteries Download PDF

Info

Publication number
WO2019242227A1
WO2019242227A1 PCT/CN2018/116965 CN2018116965W WO2019242227A1 WO 2019242227 A1 WO2019242227 A1 WO 2019242227A1 CN 2018116965 W CN2018116965 W CN 2018116965W WO 2019242227 A1 WO2019242227 A1 WO 2019242227A1
Authority
WO
WIPO (PCT)
Prior art keywords
centerline
point
bifurcation
seg
image
Prior art date
Application number
PCT/CN2018/116965
Other languages
French (fr)
Chinese (zh)
Inventor
冯建江
周杰
曾邵雯
Original Assignee
清华大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 清华大学 filed Critical 清华大学
Publication of WO2019242227A1 publication Critical patent/WO2019242227A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular

Definitions

  • B s represent a point set composed of all n s bifurcation points of the center line of the coronary artery in the source image
  • B t represent a point set composed of all n t bifurcation points of the center line of the coronary artery in the target image
  • the corresponding feature descriptor is extracted from the bifurcation point of the center line, and the descriptor is represented as a vector form; where the descriptor vector of the i-th bifurcation point of the source image is recorded as The descriptor vector of the i-th bifurcation point in the target image is written as
  • a graph G (V, E, M) is established for candidate A cand , where each node in the graph node set V corresponds to each branch point pair in A cand , and each edge in the graph's edge set E corresponds to
  • the off-diagonal element M (i, j) of the attribute matrix M reflects the compatibility between the i-th bifurcation point pair and the j-th bifurcation point pair, M
  • the diagonal element M (i, i) represents the similarity of the descriptor vector of the two bifurcation points in the i-th bifurcation point pair; the expression of M is as follows:
  • FIG. 2 is a schematic diagram of a bifurcation point matching related parameter according to an embodiment of the present invention.
  • a pair of images may be the same patient interval, that is, the images obtained by two scans, or two images at different points in the cardiac cycle in the same scan.
  • Two images can be arbitrarily selected as the source image and the other as the target image.
  • FIG. 3 is the coronary centerline C s of the source image
  • FIG. 3 (b) is the Coronary centerline C t ;
  • sampling points in the figure are represented by the symbol "+”, and the successfully matched sampling points are represented by the symbol "o" and assigned a numerical label. Is not labeled at the end because Is longer than Part of the label at There is no corresponding point on it because The length is longer.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

The present invention relates to the field of medical image processing, and provides an automatic registration method for coronary arteries. The method comprises: firstly obtaining a pair of coronary artery images, performing blood vessel segmentation on each image, and extracting a coronary artery centerline; then extracting a bifurcation feature from the centerline in each image and performing matching to obtain a final matched bifurcation point pair set; using the set to register centerline segments, deleting specific bifurcations and a part of the centerline segments, and obtaining a matched centerline segment pair and a finely registered sampling point pair set; and further segmenting and registering missing segments corresponding to the deleted specific centerline segments in the other image, and obtaining a final registration result of the coronary artery centerlines. The invention can deal with the inconsistency of the topological structures of the two coronary artery centerlines, and further register the missing segments and the centerline segments of the coronary arteries, and can improve the integrity of the segmentation and registration results.

Description

一种冠状动脉的自动配准方法Automatic registration method of coronary arteries
相关申请的交叉引用Cross-reference to related applications
本申请要求清华大学于2018年06月21日提交的、发明名称为“一种冠状动脉的自动配准方法”的、中国专利申请号“201810643770.6”的优先权。This application claims the priority of Chinese Patent Application No. “201810643770.6” filed on June 21, 2018 and entitled “A method for automatic registration of coronary arteries” by Tsinghua University.
技术领域Technical field
本发明涉及医学图像处理领域，特别涉及一种冠状动脉的自动配准方法。The invention relates to the field of medical image processing, in particular to an automatic registration method of a coronary artery.
背景技术Background technique
冠状动脉疾病是全世界致死率最高的因素之一，而计算机断层扫描血管造影技术是当下冠状动脉成像的一种主流方式。对同一病人不同时期(例如初次就诊和复查)的冠状动脉图像进行对比，可以观察病情发展状况从而有利于调整治疗方案。对心脏运动周期的不同时相的冠状动脉图像进行对比，可以分析心动周期内的冠状动脉运动规律和管腔变化规律。不同冠状动脉图像之间存在形状、姿态等差别，这是由外部原因(例如图像获取处于心脏运动周期的不同时间点、成像时人体与仪器的相对位置和角度不同)和内部原因(例如可能发生的血管重塑、病人的心脏及呼吸运动)共同导致的。上述差别给冠状动脉的对比分析造成困难，所以事先对冠状动脉进行精确的自动配准，明确各点对应关系，具有必要性，这样可以大大提高诊断效率和准确率。Coronary artery disease is one of the most lethal factors in the world, and computed tomography angiography is a mainstream method of coronary artery imaging. Comparing the coronary arterial images of the same patient at different periods (such as initial visits and re-examinations) can observe the development of the disease and facilitate adjustment of the treatment plan. By comparing the coronary arterial images of different phases of the cardiac cycle, we can analyze the rules of coronary artery movement and luminal changes during the cardiac cycle. There are differences in shape and posture between different coronary arterial images, which are caused by external reasons (for example, the image acquisition is at different points in the cardiac motion cycle, the relative position and angle of the human body and the instrument are different during imaging), and internal reasons (such as may occur Vascular remodeling, the patient's heart and respiratory movements). The above differences cause difficulties in the comparative analysis of the coronary arteries. Therefore, it is necessary to perform accurate automatic registration of the coronary arteries in advance, and to clarify the corresponding relationship between the points, which can greatly improve the diagnosis efficiency and accuracy.
冠状动脉具有树状结构，分支多且几乎分布于全心脏表面，管腔细小。鉴于其树状结构，冠状动脉血管通常被视作很多血管片段的集合，血管片段被定义为夹在两个分叉点或一个分叉点和一个端点(起点或末端)之间的血管部分。现有的医学图像配准方法大多是针对体积较大、分布较为集中的器官，例如脑部、心脏和肺部。如果对冠状动脉应用一般的图像配准方法，由于周围其他组织的严重影响，配准效果较差。所以设计一种针对冠状动脉且精度高的配准方法十分必要。The coronary arteries have a tree-like structure, with many branches and distributed almost on the surface of the whole heart, and the lumen is small. In view of its tree-like structure, coronary arteries are generally regarded as a collection of many blood vessel segments, and a blood vessel segment is defined as a part of a blood vessel sandwiched between two bifurcation points or a bifurcation point and an end point (starting point or end point). Most existing medical image registration methods are aimed at organs with large volumes and relatively concentrated distribution, such as the brain, heart, and lungs. If the general image registration method is applied to the coronary arteries, the registration effect is poor due to the severe influence of other surrounding tissues. Therefore, it is necessary to design a highly accurate registration method for coronary arteries.
现有的一种针对冠状动脉的配准方法，该方法首先从一对计算机断层扫描血管造影图像(此处一对图像指的是同一个病人间隔一段时间，即两次扫描得到的图像，或是同一次扫描中处于心动周期不同时间点的两张图像)中分别提取出冠状动脉血管的中心线，并对两个中心线进行密集的点采样得到两组点集，以这两组点集作为配准方法的输入。然后，利用相干点漂移方法对两个点集进行匹配，利用该点集匹配计算得到的变形场被用来作为冠状动脉配准的空间变换模型。这种方法具有的不足之处包括：1)将整个树状结构的冠状 动脉中心线上的离散点作为整体点集进行匹配，没有对不同分支上的点加以区别，容易发生两个不同分支上的点被拉近的错误匹配；2)作为输入的两个树状冠状动脉中心线往往具有拓扑结构不一致的情形，比如一个冠状动脉具有的分支在另一个冠状动脉中被遗漏，该方法将冠状动脉上的所有点作为整体，所以对拓扑结构不一致的地方的点匹配和可能发生错误；3)该方法的配准只关注两个冠状动脉的共有部分，对于一个冠状动脉具有而另一个冠状动脉遗漏的血管，即使不会发生错误的点匹配，也是选择直接忽略这部分血管。但是这些血管很可能具有病变，对于诊断和医学研究尤其重要，直接忽略大大降低了实际应用价值。An existing registration method for coronary arteries. This method first scans an angiographic image from a pair of computed tomography (here, a pair of images refers to the image obtained by the same patient at intervals, that is, two scans, or Are the two images at different points in the cardiac cycle in the same scan) respectively extract the centerline of the coronary arteries, and perform dense point sampling on the two centerlines to obtain two sets of points. As an input to the registration method. Then, the coherent point drift method is used to match the two point sets, and the deformation field calculated by using the point set matching is used as a spatial transformation model for coronary artery registration. The shortcomings of this method include: 1) Matching the discrete points on the coronary centerline of the entire tree structure as a whole point set, without distinguishing the points on different branches, it is easy to occur on two different branches. The points that are closer to each other are mismatched; 2) The centerlines of the two tree-like coronary arteries as inputs often have inconsistent topological structures, such as branches of one coronary artery being missed in another coronary artery. This method will All the points on the artery as a whole, so the points in the place where the topology is inconsistent are matched and errors may occur; 3) The registration of this method focuses only on the common part of the two coronary arteries, which has one coronary artery and the other coronary artery For missing blood vessels, even if the wrong point matching does not occur, they are also chosen to ignore these blood vessels directly. However, these blood vessels are likely to have lesions, which is particularly important for diagnosis and medical research. Direct neglect greatly reduces the practical application value.
发明内容Summary of the Invention
本发明的目的是为克服已有技术的不足之处，提出一种冠状动脉的自动配准方法。本发明可以处理输入的两个冠状动脉中心线拓扑结构不一致的情形，并且可以对输入冠状动脉的遗漏血管片段和中心线片段进行进一步分割和提取，并做进一步配准，提高分割和配准结果的完整性，更具有实际的医学研究及诊断价值。The purpose of the invention is to overcome the shortcomings of the prior art, and propose an automatic registration method for coronary arteries. The invention can handle the situation where the topological structures of the two coronary arteries are inconsistent, and can further segment and extract the missing coronary artery segment and centerline segment of the input coronary arteries, and perform further registration to improve the segmentation and registration results The completeness of the research has practical medical research and diagnostic value.
一种冠状动脉的自动配准方法，其特征在于，该方法包括以下步骤：A method for automatic registration of coronary arteries, characterized in that the method includes the following steps:
1)获取冠状动脉图像，对每张冠状动脉图像进行血管分割并提取对应的中心线；具体步骤如下：1) Obtain a coronary artery image, segment each coronary artery image and extract the corresponding centerline; the specific steps are as follows:
1.1)获取冠状动脉图像；1.1) Acquire coronary artery images;
获取一对针对某个冠状动脉所做的计算机断层扫描血管造影图像，将两张图像中任意选取一张作为源图像，另一张作为目标图像；Obtain a pair of computed tomography angiography images for a certain coronary artery, and randomly select one of the two images as the source image and the other as the target image;
1.2)对步骤1.1)获取的每张图像分割冠状动脉血管，并提取每个冠状动脉血管对应的中心线；将源图像中的冠状动脉血管记作V s，中心线记作C s；目标图像中的冠状动脉血管记作V t，中心线记作C t； 1.2) For each image obtained in step 1.1), segment the coronary arteries and extract the center line corresponding to each coronary artery; record the coronary arteries in the source image as V s and the center line as C s ; the target image Coronary arterial blood vessels are denoted by V t , and the center line is denoted by C t ;
2)对步骤1)的每张冠状动脉图像中的中心线提取分叉点特征并进行匹配，得到最终匹配的分叉点点对集合；具体步骤如下：2) For each of the coronary arterial images in step 1), the branch point features are extracted and matched to obtain the final matching branch point point set; the specific steps are as follows:
2.1)提取分叉点的特征；2.1) Extract the features of the bifurcation points;
令B s代表源图像中冠状动脉中心线全部n s个分叉点构成的点集，B t代表目标图像中冠状动脉中心线全部n t个分叉点构成的点集；对每个冠状动脉中心线的分叉点提取对应的特征描述子，将该描述子表示为向量形式；其中，源图像的第i个分叉点的描述子向量记作
Figure PCTCN2018116965-appb-000001
目标图像中第i个分叉点的描述子向量记作
Figure PCTCN2018116965-appb-000002
Let B s represent a point set composed of all n s bifurcation points of the center line of the coronary artery in the source image, and B t represent a point set composed of all n t bifurcation points of the center line of the coronary artery in the target image; for each coronary artery The corresponding feature descriptor is extracted from the bifurcation point of the center line, and the descriptor is represented as a vector form; where the descriptor vector of the i-th bifurcation point of the source image is recorded as
Figure PCTCN2018116965-appb-000001
The descriptor vector of the i-th bifurcation point in the target image is written as
Figure PCTCN2018116965-appb-000002
2.2)获取候选匹配分叉点点对集合A cand； 2.2) Acquire the candidate matching fork point point set A cand ;
计算B t中的每一个分叉点b t与B s中所有分叉点的描述子向量的欧氏距离，并将计算结 果中最小欧氏距离对应的B s中的分叉点记作b t的对应分叉点b s，共得到n t个分叉点点对，将n t个分叉点点对的集合记作候选匹配分叉点点对集合A cand； B t calculated in each bifurcation point b B s T Euclidean described subvector all branching point distance, and the calculated result of the branching point corresponding to the minimum Euclidean distance is referred to as B s b t corresponding to the branch point b s, were obtained for n-bit t bifurcation and the n-bit t of the bifurcation, referred to as the set of candidate matching set of bit bifurcation a cand;
2.3)计算集合A cand的属性矩阵M； 2.3) Calculate the attribute matrix M of the set A cand ;
首先，对候A cand建立一个图G(V,E,M)，其中图的节点集合V中的每个节点对应A cand中每个分叉点点对，图的边集合E中每条边对应每两个分叉点对之间的关系，属性矩阵M的非对角线元素M(i,j)反映第i个分叉点对和第j个分叉点对之间的兼容性，M的对角线元素M(i,i)表示第i个分叉点对中两个分叉点的描述子向量相似度；M的表达式如下： First, a graph G (V, E, M) is established for candidate A cand , where each node in the graph node set V corresponds to each branch point pair in A cand , and each edge in the graph's edge set E corresponds to The relationship between every two bifurcation point pairs, the off-diagonal element M (i, j) of the attribute matrix M reflects the compatibility between the i-th bifurcation point pair and the j-th bifurcation point pair, M The diagonal element M (i, i) represents the similarity of the descriptor vector of the two bifurcation points in the i-th bifurcation point pair; the expression of M is as follows:
Figure PCTCN2018116965-appb-000003
Figure PCTCN2018116965-appb-000003
Figure PCTCN2018116965-appb-000004
Figure PCTCN2018116965-appb-000004
Figure PCTCN2018116965-appb-000005
Figure PCTCN2018116965-appb-000005
Figure PCTCN2018116965-appb-000006
Figure PCTCN2018116965-appb-000006
其中，d i代表第i个分叉点对中两个分叉点m i1和m i2的欧氏距离，θ i是m i1和m i2方向的相对角度，
Figure PCTCN2018116965-appb-000007
是把m i2对应的向量当作极轴时m i1的极角；TH dist、TH θ和
Figure PCTCN2018116965-appb-000008
分别代表ratio(i,j)下限的阈值、|θ i-θ j|上限的阈值和
Figure PCTCN2018116965-appb-000009
上限的阈值；TH dist取值在[0.4,0.9]区间，TH θ取值在[30°,90°]区间，
Figure PCTCN2018116965-appb-000010
取值在[30°,90°]区间； Wherein, D i represents the i-th bifurcation points of the two diverging points m i1 m i2 and the Euclidean distance, a relative angle [theta] i and m i1 m i2 direction,
Figure PCTCN2018116965-appb-000007
Is the polar angle m i1 m i2 when the corresponding vector as a pole shaft; TH dist, TH θ and
Figure PCTCN2018116965-appb-000008
Represents the threshold of the lower limit of ratio (i, j), the threshold of the upper limit of | θ i -θ j |, and
Figure PCTCN2018116965-appb-000009
Upper threshold; TH dist is in the interval [0.4,0.9], and TH θ is in the interval [30 °, 90 °].
Figure PCTCN2018116965-appb-000010
The value is in the interval [30 °, 90 °];
2.4)计算最终匹配的分叉点点对集合A final；具体步骤如下： 2.4) Calculate the final matching fork point pair set A final ; the specific steps are as follows:
2.4.1)建立最终匹配的分叉点点对集合A final，并将A final初始化为空集； 2.4.1) Establish the final matching fork point pair set A final and initialize A final to the empty set;
2.4.2)对矩阵M做特征值分解，找到M的最大特征值对应的特征向量x *； 2.4.2) Eigenvalue decomposition of the matrix M to find the eigenvector x * corresponding to the largest eigenvalue of M;
2.4.3)对x *中的元素进行由大到小的排序，得到序列x′，x′从前往后的各个元素在x *中的序号组成序列L； 2.4.3) x * of the elements in descending order, to obtain a sequence x ', x' front to back in the respective elements of x * constituent sequence number L;
2.4.4)对L进行判定：若L为空，则输出当前的A final，求解结束；若L不为空，则进入步骤2.4.5)； 2.4.4) Judge L: If L is empty, then output the current A final and the solution ends; if L is not empty, go to step 2.4.5);
2.4.5)取序列L中的第一个值L(1)并判定：如果x *(L(1))＜ε，则输出当前的A final，求解结束，其中ε取值在[0.0000001,0.01]区间；否则，进入步骤2.4.6)； 2.4.5) Take the first value L (1) in the sequence L and determine: if x * (L (1)) <ε, then output the current A final and the solution ends, where ε takes the value [0.0000001, 0.01] interval; otherwise, proceed to step 2.4.6);
2.4.6)如果A cand中的第L(1)个分叉点对与A final中任意一对分叉点对包含相同的分叉点，则把L(1)从L中删除，重新返回步骤2.4.4)；否则，进入步骤2.4.7)； 2.4.6) If the L (1) th fork point pair in A cand contains the same fork point as any pair of fork points in A final , delete L (1) from L and return Step 2.4.4); otherwise, go to Step 2.4.7);
2.4.7)把A cand中的第L(1)个分叉点对加入到集合A final中，将L(1)从L中删除，重新返 回步骤2.4.4)； 2.4.7) Add the L (1) bifurcation point pair in A cand to the set A final , delete L (1) from L, and return to step 2.4.4);
3)对每张冠状动脉图像中的中心线进行更新，得到删除的中心线片段集合；利用更新后的中心线和步骤2)的最终匹配的分叉点点对集合对中心线片段进行匹配；具体步骤如下：3) Update the centerline in each coronary artery image to obtain the deleted centerline fragment set; use the updated centerline and the final matching branch point of step 2) to match the set to the centerline fragment; specifically Proceed as follows:
3.1)对每张冠状动脉图像中的中心线进行更新，得到删除的中心线片段集合；具体步骤如下：3.1) Update the centerline in each coronary artery image to obtain the deleted centerline fragment set; the specific steps are as follows:
3.1.1)获取C s或C t中的任一特有分叉点，分别计算该特有分叉点对应的两个支干中心线片段和一个主干中心线片段在分叉点处的方向向量，删除方向向量与主干中心线片段的方向向量夹角更大的一个支干中心线片段，同时删除该特有分叉点； 3.1.1) Obtain any unique bifurcation point in C s or C t , and calculate the direction vector of the two branch centerline segments and one trunk centerline segment at the bifurcation point corresponding to the unique bifurcation point, respectively. Delete a branch centerline segment with a greater angle between the direction vector and the direction vector of the trunk centerline segment, and delete the unique branch point
3.1.2)从C s和C t的剩余分叉点中继续寻找特有分叉点并判定：若存在特有分叉点，则重新返回步骤3.1.1)；若不存在特有分叉点，则结束中心线更新过程，得到更新后的源图像的冠状动脉中心线记为C′ s，更新后的目标图像的冠状动脉中心线记为C′ t，并将所有删除的中心线片段集合记作Set seg_del； 3.1.2) Continue to find unique branching points from the remaining branching points of C s and C t and determine: if there is a unique branching point, return to step 3.1.1); if there is no unique branching point, then End the centerline update process, obtain the updated centerline of the coronary arteries as C ′ s , record the updated centerline of the coronary arteries as C ′ t , and record all deleted centerline segments as Set seg_del ;
3.2)中心线片段的匹配；3.2) Matching of centerline segments;
利用步骤3.1)得到的C′ s和C′ t和步骤2)得到的最终匹配的分叉点点对集合A final，满足以下任意两个条件之一的中心线片段被认为是匹配的：a)夹在两对匹配分叉点点对之间的两个中心线片段；b)一端为冠状动脉起点或终点，另一端为匹配的分叉点点对，并且分叉点点对的两个分叉点的方向夹角小于设定的夹角阈值的两个中心线片段；将匹配后的中心线片段对的集合记作Set seg_pair； Using C ′ s and C ′ t obtained in step 3.1) and the set of final matching branch point pairs A final obtained in step 2), a centerline segment that meets any of the following two conditions is considered to be a match: a) Two centerline segments sandwiched between two pairs of matching bifurcation point pairs; b) one end is the origin or end of the coronary artery, the other end is the matching bifurcation point pair, and the two bifurcation points of the bifurcation point pair Two centerline segments whose direction included angle is smaller than a set included angle threshold; the set of matched centerline segment pairs is referred to as Set seg_pair ;
3.3)中心线片段的精细配准；3.3) Fine registration of centerline segments;
将Set seg_pair中属于源图像的中心线上片段记作Seg s，Set seg_pair中属于目标图像的中心线片段记作Seg t；对Seg s和Seg t分别进行点采样，Seg s的采样间隔为Δ s，Seg t的采样间隔为Δ t； The Set seg_pair belonging to the center line of the source image is referred to as segments Seg s, belonging to the centerline segment Set seg_pair referred to as target image Seg t; on Seg s Seg t and for each sampling point, the sampling interval is Δ Seg s s , the sampling interval of Seg t is Δ t ;
假设Seg s和Seg t分别具有N s和N t个采样点，定义一个点集之间的离散变换模型如下： Assuming Seg s and Seg t have N s and N t sampling points, respectively, the discrete transformation model defining a point set is as follows:
S：＝{S(i)＝j,i∈{0,1,…,N t},j∈{0,1,…,N s}}， S: = {S (i) = j, i∈ {0,1,…, N t }, j∈ {0,1,…, N s }},
其中，S(i)是Seg t的第i个采样点的状态，S(i)＝j表示Seg t的第i个采样点与Seg s的第j个采样点相对应； Among them, S (i) is the state of the i-th sampling point of Seg t , and S (i) = j means that the i-th sampling point of Seg t corresponds to the j-th sampling point of Seg s ;
目标函数为：The objective function is:
Figure PCTCN2018116965-appb-000011
Figure PCTCN2018116965-appb-000011
本发明的特点及有益效果在于：The features and beneficial effects of the present invention are:
本发明对不同计算机断层扫描血管造影图像中的冠状动脉进行配准，首先找出冠状动脉中心线的分叉点，利用分叉点信息把树状冠状动脉中心线分解成一个个片段，找出两个冠状动脉的中心线片段之间的匹配关系，同时也确定了拓扑结构不一致的部分。本发明对成对的中心线片段上的采样点进行点集匹配，所以不会发生不同分支上的点错误匹配的情况，匹配的过程也不会受到不一致的拓扑结构的影响。另外，对于一个冠状动脉具有而另一个冠状动脉遗漏的血管片段和中心线片段，本发明可以对其进行进一步分割和提取，并做进一步的配准，将很多具有医学价值的病变血管片段分割出来并配准，具有很重要的医学研究及诊断价值。The invention registers the coronary arteries in different computer tomography angiography images. First, the bifurcation points of the centerline of the coronary arteries are found, and the centerline of the tree-like coronary arteries is decomposed into fragments by using the information of the bifurcation points. The matching relationship between the centerline segments of the two coronary arteries also identified the parts with inconsistent topology. The present invention performs point set matching on sample points on paired centerline segments, so that point mismatching on different branches does not occur, and the matching process is not affected by inconsistent topological structures. In addition, the present invention can further segment and extract the blood vessel fragments and centerline fragments that one coronary artery has but the other coronary artery misses, and perform further registration to segment many diseased blood vessel fragments with medical value. And registration, it has very important medical research and diagnostic value.
1)对于输入的两个冠状动脉拓扑结构不一致的情形，本发明的配准效果不会受到其影响，具有很强的鲁棒性；1) The registration effect of the present invention is not affected by the inconsistency of the topological structure of the two coronary arteries, and has strong robustness;
2)本发明可以对冠状动脉中遗漏的血管和中心线片段进行再一次分割和提取，同时提升分割和配准的完整性，并且这一部分血管具有很强的医学研究及诊断价值；2) The present invention can segment and extract the missing blood vessels and centerline segments in the coronary arteries, while improving the integrity of segmentation and registration, and this part of the blood vessels has strong medical research and diagnostic value;
3)本发明可达到目前最高的配准精度。3) The present invention can achieve the highest registration accuracy at present.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1为本发明方法的整体流程图。FIG. 1 is an overall flowchart of the method of the present invention.
图2为本发明实施例的分叉点匹配相关参数示意图。FIG. 2 is a schematic diagram of a bifurcation point matching related parameter according to an embodiment of the present invention.
图3为本发明实施例的拓扑不一致的中心线示意图。FIG. 3 is a schematic diagram of a centerline with inconsistent topologies according to an embodiment of the present invention.
图4为本发明实施例的拓扑不一致的中心线例子图。FIG. 4 is a diagram illustrating an example of a centerline with inconsistent topologies according to an embodiment of the present invention.
图5为本发明实施例的中心线片段配准结果图。FIG. 5 is a graph of centerline segment registration results according to an embodiment of the present invention.
图6为本发明实施例的进一步分割后的冠状动脉血管模型例子图。FIG. 6 is an example diagram of a coronary artery blood vessel model after further segmentation according to an embodiment of the present invention.
图7为本发明实施例的冠状动脉配准结果图。FIG. 7 is a graph of coronary artery registration results according to an embodiment of the present invention.
图8为本发明实施例的冠状动脉配准结果图。FIG. 8 is a diagram of a coronary artery registration result according to an embodiment of the present invention.
具体实施方式detailed description
本发明提出一种冠状动脉的自动配准方法，下面结合附图和具体实施例进一步详细说明如下。The present invention provides a method for automatic registration of coronary arteries, which will be described in further detail below with reference to the accompanying drawings and specific embodiments.
本发明提出一种冠状动脉的自动配准方法，整体流程如图1所示，包括以下步骤：The present invention proposes a method for automatic registration of coronary arteries. The overall process is shown in FIG. 1 and includes the following steps:
1)获取冠状动脉图像，对每张冠状动脉图像进行血管分割并提取对应的中心线；1) Acquire coronary artery images, segment each coronary artery image and extract the corresponding centerline;
获取两张冠状动脉的计算机断层扫描血管造影图像(简称冠状动脉图像)，并分别从两张图像中分割出冠状动脉血管并提取对应的冠状动脉中心线。具体步骤如下：Computed tomographic angiographic images (referred to as coronary images) of two coronary arteries were obtained, and the coronary arteries were segmented from the two images and the corresponding coronary centerlines were extracted. Specific steps are as follows:
1.1)获取冠状动脉图像；1.1) Acquire coronary artery images;
获取一对针对某个冠状动脉所做的计算机断层扫描血管造影图像。此处，一对图像可以是同一个病人间隔一段时间，即两次扫描得到的图像，可以是同一次扫描中处于心动周期不同时间点的两张图像。两张图像任意选取一张作为源图像，另一张作为目标图像即可。Acquire a pair of computed tomography angiograms of a coronary artery. Here, a pair of images may be the same patient interval, that is, the images obtained by two scans, or two images at different points in the cardiac cycle in the same scan. Two images can be arbitrarily selected as the source image and the other as the target image.
1.2)对步骤1.1)获取的每张图像分割冠状动脉血管，并提取每个冠状动脉血管对应的中心线；1.2) Segment the coronary arteries for each image obtained in step 1.1), and extract the centerline corresponding to each coronary artery;
冠状动脉的血管分割和中心线提取结果可以是人工从计算机断层扫描血管造影图像中手动标注的，也可以是利用半自动或全自动算法进行分割和提取的。将源图像中的冠状动脉血管记作V s，中心线记作C s；目标图像中的冠状动脉血管记作V t，中心线记作C t。 Coronary artery vessel segmentation and centerline extraction results can be manually labeled manually from computed tomography angiography images, or they can be segmented and extracted using semi-automatic or fully automatic algorithms. The coronary arteries in the source image are denoted as V s , the center line is denoted as C s ; the coronary arteries in the target image are denoted as V t , and the central line is denoted as C t .
2)对步骤1)的每张冠状动脉图像中的中心线提取分叉点特征并进行匹配，得到最终匹配的分叉点点对集合；具体步骤如下：2) For each of the coronary arterial images in step 1), the branch point features are extracted and matched to obtain the final matching branch point point set; the specific steps are as follows:
2.1)提取分叉点的特征；2.1) Extract the features of the bifurcation points;
步骤1)得到的每个冠状动脉血管对应的中心线(简称“冠状动脉中心线”)都是树状结构且包含一组分叉点，其中分叉点指的是由冠状动脉中心线产生分叉的点。令B s代表源图像中冠状动脉中心线全部n s个分叉点构成的点集，B t代表目标图像中冠状动脉中心线全部n t个分叉点构成的点集。 The centerline corresponding to each coronary artery obtained in step 1) (referred to as "coronary centerline") is a tree structure and contains a set of bifurcation points, where the bifurcation points refer to the points generated by the coronary centerline. Forked point. Let B s represent a point set consisting of all n s bifurcation points of the coronary centerline in the source image, and B t represent a point set consisting of all n t bifurcation points of the coronary centerline in the target image.
本发明中，分叉点被赋予位置和方向两个性质。位置用三维空间坐标表示，方向定义为沿着血管中心线并且从心脏近端指向远端的切线方向，用方向向量表示。本发明对每个冠状动脉中心线的分叉点提取对应的特征描述子(例如三维尺度不变特征变换描述子，或者其他类型的特征描述子)，将该描述子表示为向量形式。其中，源图像的第i个分叉点的描述子向量记作
Figure PCTCN2018116965-appb-000012
目标图像中第i个分叉点的描述子向量记作
Figure PCTCN2018116965-appb-000013
In the present invention, the bifurcation point is given two properties of position and direction. The position is expressed in three-dimensional spatial coordinates, and the direction is defined as a tangential direction along the centerline of the blood vessel and from the proximal end of the heart to the distal end, and is represented by a direction vector. The present invention extracts a corresponding feature descriptor (for example, a three-dimensional scale-invariant feature transformation descriptor, or other type of feature descriptor) for each bifurcation point of a coronary artery centerline, and represents the descriptor as a vector form. Among them, the descriptor vector of the i-th bifurcation point of the source image is written as
Figure PCTCN2018116965-appb-000012
The descriptor vector of the i-th bifurcation point in the target image is written as
Figure PCTCN2018116965-appb-000013
2.2)获取候选匹配分叉点点对集合A cand； 2.2) Acquire the candidate matching fork point point set A cand ;
计算B t中的每一个分叉点b t与B s中所有分叉点的描述子向量的欧氏距离，并将计算结果中最小欧氏距离对应的B s中的分叉点记作b t的对应分叉点b s，共得到n t个分叉点点对，将n t个分叉点点对的集合记作候选匹配分叉点点对集合A cand。A cand是分叉点匹配过程中初步得到的候选匹配分叉点点对集合，之后通过步骤2.3)和2.4)从中挑选出最终的匹配分叉点点对。 B t calculated in each bifurcation point b B s T Euclidean described subvector all branching point distance, and the calculated result of the branching point corresponding to the minimum Euclidean distance is referred to as B s b t corresponding to the branch point b s, were obtained for n-bit t bifurcation and the n-bit t bifurcation, referred to as the candidate set of matching bit pair set bifurcated a cand. A cand is a set of candidate matching forking point pairs initially obtained during the forking point matching process, and then the final matching forking point pairs are selected from steps 2.3) and 2.4).
2.3)计算集合A cand的属性矩阵M； 2.3) Calculate the attribute matrix M of the set A cand ;
本发明中分叉点的最终匹配结果可以用多种方法得到，这里以基于谱聚类方法的点集匹配方法为例。The final matching result of the bifurcation points in the present invention can be obtained by various methods. Here, a point set matching method based on a spectral clustering method is taken as an example.
首先，对A cand建立一个图G(V,E,M)，其中图的节点集合V中的每个节点对应A cand中每个分叉点点对，图的边集合E中每条边对应每两个分叉点对之间的关系，属性矩阵M的非对角线元素M(i,j)(i≠j)反映第i个分叉点对和第j个分叉点对之间的兼容性(取值在(0,1)区 间，数值越大代表兼容性越高)，M的对角线元素M(i,i)表示第i个分叉点对中两个分叉点的描述子向量相似度(取值在(0,1)区间，数值越大代表相似性越高)。M与三个几何参数有关，如图2所示，d i代表第i个分叉点对中分叉点m i1和m i2的欧氏距离，θ i是m i1和m i2方向的相对角度，
Figure PCTCN2018116965-appb-000014
是把m i2对应的向量当作极轴时m i1的极角。M的具体定义如下： First, build a graph G (V, E, M) for A cand , where each node in the graph's node set V corresponds to each branch point pair in A cand , and each edge in the graph's edge set E corresponds to each The relationship between two bifurcation point pairs. The off-diagonal element M (i, j) (i ≠ j) of the attribute matrix M reflects the relationship between the i-th bifurcation point pair and the j-th bifurcation point pair. Compatibility (value in the range of (0,1), the larger the value, the higher the compatibility), the diagonal element M (i, i) of M represents the value of the two branch points in the i-th branch point pair Descriptor vector similarity (value in the range of (0,1), the larger the value, the higher the similarity). M with three relevant geometric parameters, as shown, D i represents the i-th branch point of the branch point m i1 m i2 in FIG. 2 and the Euclidean distance, a relative angle [theta] i and m i1 m i2 direction ,
Figure PCTCN2018116965-appb-000014
Is the polar angle m i1 m i2 when the corresponding vector as polar. The specific definition of M is as follows:
Figure PCTCN2018116965-appb-000015
Figure PCTCN2018116965-appb-000015
Figure PCTCN2018116965-appb-000016
Figure PCTCN2018116965-appb-000016
Figure PCTCN2018116965-appb-000017
Figure PCTCN2018116965-appb-000017
Figure PCTCN2018116965-appb-000018
Figure PCTCN2018116965-appb-000018
其中，TH dist、TH θ和
Figure PCTCN2018116965-appb-000019
分别代表ratio(i,j)下限的阈值、|θ i-θ j|上限的阈值和
Figure PCTCN2018116965-appb-000020
上限的阈值。TH dist取值在[0.4,0.9]区间，TH θ取值在[30°,90°]区间，
Figure PCTCN2018116965-appb-000021
取值在[30°,90°]区间。 Among them, TH dist , TH θ and
Figure PCTCN2018116965-appb-000019
Represents the threshold of the lower limit of ratio (i, j), the threshold of the upper limit of | θ i -θ j |, and
Figure PCTCN2018116965-appb-000020
The upper threshold. The value of TH dist is in the interval [0.4,0.9], and the value of TH θ is in the interval [30 °, 90 °].
Figure PCTCN2018116965-appb-000021
The value is in the range of [30 °, 90 °].
2.4)计算最终匹配的分叉点点对集合A final； 2.4) Calculate the final matching branch point pair set A final ;
由于正确的点对之间存在紧密的关联，本发明中将分叉点匹配问题转化成图G的节点聚类问题，进而利用特征向量方法求解，最终得到所有正确匹配的分叉点点对集合，记为A final。求解算法的具体步骤如下： Due to the close relationship between the correct point pairs, in the present invention, the bifurcation point matching problem is transformed into the node clustering problem of graph G, and then the feature vector method is used to solve, and finally all correct matching bifurcation point point set sets are obtained. Called A final . The specific steps of the algorithm are as follows:
2.4.1)建立最终匹配的分叉点点对集合A final，并将A final初始化为空集； 2.4.1) Establish the final matching fork point pair set A final and initialize A final to the empty set;
2.4.2)对矩阵M做特征值分解，找到M的最大特征值对应的特征向量x *； 2.4.2) Eigenvalue decomposition of the matrix M to find the eigenvector x * corresponding to the largest eigenvalue of M;
2.4.3)对x *中的元素进行由大到小的排序，得到序列x′，x′从前往后的各个元素在x *中的序号组成序列L； 2.4.3) x * of the elements in descending order, to obtain a sequence x ', x' front to back in the respective elements of x * constituent sequence number L;
2.4.4)对L进行判定：若L为空，则输出当前的A final，结束求解算法；若L不为空，进入步骤2.4.5)； 2.4.4) Determine L: if L is empty, then output the current A final to end the algorithm; if L is not empty, go to step 2.4.5);
2.4.5)取序列L中的第一个值L(1)并判定：如果x *(L(1))<ε，则输出当前的A final，结束求解算法，其中ε取值在[0.0000001,0.01]区间；否则，进入步骤2.4.6)； 2.4.5) Take the first value L (1) in the sequence L and determine: if x * (L (1)) <ε, then output the current A final and end the solution algorithm, where ε takes a value of [0.0000001 , 0.01] interval; otherwise, proceed to step 2.4.6);
2.4.6)如果A cand中的第L(1)个分叉点对与A final中任意一对分叉点对包含相同的分叉点，则把L(1)从L中删除，重新返回步骤2.4.4)；否则，进入步骤2.4.7)； 2.4.6) If the L (1) th fork point pair in A cand contains the same fork point as any pair of fork points in A final , delete L (1) from L and return Step 2.4.4); otherwise, go to Step 2.4.7);
2.4.7)把A cand中的第L(1)个分叉点对加入到集合A final中，将L(1)从L中删除，重新返回步骤2.4.4)。 2.4.7) Add the L (1) th branch point pair in A cand to the set A final , delete L (1) from L, and return to step 2.4.4).
3)对每张冠状动脉图像中的中心线进行更新，得到删除的中心线片段集合；利用更新后的中心线和步骤2)的最终匹配的分叉点点对集合对中心线片段进行匹配；具体步骤如下：3) Update the centerline in each coronary artery image to obtain the deleted centerline fragment set; use the updated centerline and the final matching branch point of step 2) to match the set to the centerline fragment; specifically Proceed as follows:
血管片段被定义为夹在冠状动脉的两个分叉点之间的血管部分，或者夹在一个分叉点和一个冠状动脉的起点(起点指的是与主动脉相连的冠状动脉出口)或终点(终点指的是冠状动脉末梢的结束点)之间的血管部分。冠状动脉的中心线片段定义为血管片段的中心线。对于与一个分叉点相连的三个中心线片段做如下定义：如果中心线片段的另一个端点相对于该分叉点距离冠状动脉的起点更近，那么称该片段为该分叉点的“主干片段”；如果中心线片段的另一个端点相对于该分叉点距离冠状动脉的起点更远，那么称该片段为该分叉点的“支干片段”。每个分叉点具有一个主干片段和两个支干片段。具体步骤如下：A vascular segment is defined as the portion of a blood vessel sandwiched between two bifurcation points of a coronary artery, or between the bifurcation point and the starting point of a coronary artery (the starting point refers to the exit of the coronary artery connected to the aorta) or the end point (End point refers to the end point of the coronary artery tip). The centerline segment of the coronary artery is defined as the centerline of the vessel segment. The three centerline segments connected to a bifurcation point are defined as follows: If the other endpoint of the centerline segment is closer to the starting point of the coronary artery relative to the bifurcation point, then the segment is called the " Trunk segment "; if the other endpoint of the centerline segment is further from the origin of the coronary arteries relative to the bifurcation point, the segment is called the" branch segment "of the bifurcation point. Each bifurcation point has one trunk segment and two branch segments. Specific steps are as follows:
3.1)对每张冠状动脉图像中的中心线进行更新，得到删除的中心线片段集合3.1) Update the centerline in each coronary artery image to get the deleted centerline fragment set
很多情况下，待配准的两个冠状动脉的中心线存在拓扑结构不一致的情况，经过步骤2)的匹配分叉点之后，仍存在一些未匹配的分叉点，即某一张图像中的冠状动脉中心线特有的分叉点。具体处理方法如下：In many cases, the center lines of the two coronary arteries to be registered have inconsistent topological structures. After matching the bifurcation points in step 2), there are still some unmatched bifurcation points, that is, the A bifurcation point unique to the centerline of the coronary artery. The specific processing method is as follows:
3.1.1)获取C s或C t中的任一特有分叉点，分别计算该特有分叉点对应的两个支干中心线片段和一个主干中心线片段在分叉点处的方向向量(取从心脏近端指向心脏远端的方向向量)，删除方向向量与主干中心线片段的方向向量夹角更大的一个支干中心线片段，同时删除该特有分叉点。 3.1.1) Obtain any unique bifurcation point in C s or C t , and calculate the direction vector of the two branch centerline segments and one trunk centerline segment at the bifurcation point corresponding to the unique bifurcation point ( Take the direction vector from the proximal end of the heart to the distal end of the heart), delete a branch centerline segment with a larger angle between the direction vector and the direction vector of the trunk centerline segment, and delete the unique bifurcation point.
3.1.2)从C s和C t的剩余分叉点中继续寻找特有分叉点并判定：若存在特有分叉点，则重新返回步骤3.1.1)；若不存在特有分叉点，则结束中心线更新过程，得到更新后的源图像的冠状动脉中心线记为C′ s，更新后的目标图像的冠状动脉中心线记为C′ t，并将所有删除的中心线片段集合记作Set seg_del； 3.1.2) Continue to find unique branching points from the remaining branching points of C s and C t and determine: if there is a unique branching point, return to step 3.1.1); if there is no unique branching point, then End the centerline update process, obtain the updated centerline of the coronary arteries as C ′ s , record the updated centerline of the coronary arteries as C ′ t , and record all deleted centerline segments as Set seg_del ;
对于上述中心线更新的过程进行举例说明如下：An example of the above-mentioned centerline update process is as follows:
本实施例中拓扑不一致的两张冠状动脉图像中的冠状动脉中心线示意图如图3所示，图3(a)为源图像的冠状动脉中心线C s，图3(b)为目标图像的冠状动脉中心线C t； The schematic diagram of the centerline of the coronary arteries in the two coronary images with inconsistent topology in this embodiment is shown in FIG. 3, FIG. 3 (a) is the coronary centerline C s of the source image, and FIG. 3 (b) is the Coronary centerline C t ;
如图3所示，C s具有特有分叉点bf 2，而C t没有。本发明的血管片段匹配算法将移除C s的特有分叉点bf 2和与它相连的中心线片段c 5，并且将片段c 3和c 4合并成一个新的片段，从而C s被赋予和C t一致的新拓扑结构。注意此处移除c 5而非c 4，是因为与c 5相比，c 4与c 3的方向向量夹角更小。 As shown in Fig. 3, C s has a unique bifurcation point bf 2 while C t does not. The vascular segment matching algorithm of the present invention will remove the unique bifurcation point bf 2 of C s and the centerline segment c 5 connected to it, and merge the segments c 3 and c 4 into a new segment, so that C s is given A new topology consistent with C t . Note that c 5 is removed here instead of c 4 because the angle between the direction vectors of c 4 and c 3 is smaller than c 5 .
图4为本发明在实际应用中的拓扑不一致的冠状动脉中心线例子图。图4(a)为源图像的冠状动脉中心线C s，图4(b)为目标图像的冠状动脉中心线C t。在图4展示的实际案例中，虚线表示的C s中的三条中心线片段被从源图像的中心线中依次移除。 FIG. 4 is a diagram illustrating an example of a centerline of a coronary artery with inconsistent topologies in practical applications of the present invention. FIG. 4 (a) is a source image of coronary centerline C s, FIG. 4 (b) coronary centerline C t of the target image. In the actual case shown in FIG. 4, the three centerline segments in C s indicated by the dotted line are sequentially removed from the centerline of the source image.
3.2)中心线片段的匹配；3.2) Matching of centerline segments;
利用步骤3.1)得到的C′ s和C′ t和步骤2)得到的最终匹配的分叉点点对集合A final，满足以下任意两个条件之一的中心线片段被认为是匹配的：a)夹在两对匹配分叉点点对之间的两个中心线片段；b)一端为冠状动脉起点或终点，另一端为匹配的分叉点点对，并且分叉点点对的两个分叉点的方向夹角小于设定的夹角阈值(本发明夹角阈值取值范围为[10°,60°]区间)的两个中心线片段。此时得到了成对的中心线片段，将匹配后的中心线片段对的集合记作Set seg_pair； Using C ′ s and C ′ t obtained in step 3.1) and the set of final matching branch point pairs A final obtained in step 2), a centerline segment that meets any of the following two conditions is considered to be a match: a) Two centerline segments sandwiched between two pairs of matching bifurcation point pairs; b) one end is the origin or end of the coronary artery, the other end is the matching bifurcation point pair, and the two bifurcation points of the bifurcation point pair Two centerline segments whose direction included angle is smaller than a set included angle threshold (the range of the included angle threshold of the present invention is in the range of [10 °, 60 °]). At this point, a pair of centerline segments is obtained, and the set of matched centerline segment pairs is referred to as Set seg_pair ;
3.3)中心线片段的精细配准；3.3) Fine registration of centerline segments;
明确血管片段对应关系之后，需要对Set seg_pair中的每一对中心线片段分别进行精细配准。其中，将Set seg_pair中属于源图像的中心线上片段记作Seg s，Set seg_pair中属于目标图像的中心线片段记作Seg t。对Seg s和Seg t分别进行点采样，Seg s的采样间隔为Δ s，Seg t的采样间隔为Δ t，Δ s和Δ t可以相等，也可以不等(建议Δ t/Δ s取值在[1,20]区间)。假设Seg s和Seg t分别具有N s和N t个采样点，利用一系列状态定义一个点集之间的离散变换模型如下： After the corresponding relationship between the blood vessel segments is determined , each pair of centerline segments in the Set seg_pair needs to be finely registered separately. Among them, the segment on the center line belonging to the source image in Set seg_pair is denoted as Seg s , and the segment on the center line belonging to the target image in Set seg_pair is denoted as Seg t . Seg s and Seg t are sampled separately. The sampling interval of Seg s is Δ s and the sampling interval of Seg t is Δ t . Δ s and Δ t can be equal or different (recommended Δ t / Δ s value In the [1,20] interval). Assume that Seg s and Seg t have N s and N t sampling points, respectively. A series of states are used to define a discrete transformation model between a point set as follows:
S：＝{S(i)＝j,i∈{0,1,…,N t},j∈{0,1,…,N s}}， S: = {S (i) = j, i∈ {0,1,…, N t }, j∈ {0,1,…, N s }},
其中，S(i)是Seg t的第i个采样点的状态。S(i)＝j表示Seg t的第i个采样点与Seg s的第j个采样点相对应。本发明中设计了一个目标函数并将其最大化，以求解模型S的参数，即S中i和j的取值。该目标函数由图像相似度和几何相似度两部分构成，总体目标函数为： Among them, S (i) is the state of the i-th sampling point of Seg t . S (i) = j means that the i-th sampling point of Seg t corresponds to the j-th sampling point of Seg s . In the present invention, an objective function is designed and maximized to solve the parameters of the model S, that is, the values of i and j in S. The objective function consists of image similarity and geometric similarity. The overall objective function is:
Figure PCTCN2018116965-appb-000022
Figure PCTCN2018116965-appb-000022
其中，Sim1代表图像相似度，Sim2代表几何相似度，ω 1和ω 2分别为Sim1和Sim2所占的权重。ω 1和ω 2的取值都在[0,1]区间，且ω 1与ω 2的和是1。 Among them, Sim1 represents image similarity, Sim2 represents geometric similarity, and ω 1 and ω 2 are weights occupied by Sim1 and Sim2, respectively. The values of ω 1 and ω 2 are both in the interval [0,1], and the sum of ω 1 and ω 2 is 1.
图像相似度Sim1是利用特征描述子(例如采用分叉点的特征描述子，也可以采用其他特征描述子)进行衡量的，数学表达式为：The image similarity Sim1 is measured using feature descriptors (for example, feature descriptors that use bifurcation points, or other feature descriptors). The mathematical expression is:
Figure PCTCN2018116965-appb-000023
Figure PCTCN2018116965-appb-000023
其中，
Figure PCTCN2018116965-appb-000024
是Seg t的第i个采样点的描述子，
Figure PCTCN2018116965-appb-000025
是Seg s的第S(i)个采样点的描述子。为了避免中心线折叠或过度伸缩，需要对中心线加以几何约束，对应的几何相似度表达式为： among them,
Figure PCTCN2018116965-appb-000024
Is the descriptor of the i-th sampling point of Seg t ,
Figure PCTCN2018116965-appb-000025
Is the descriptor of the S (i) th sampling point of Seg s . In order to avoid the centerline from being folded or overstretched, geometric constraints need to be imposed on the centerline. The corresponding geometric similarity expression is:
Figure PCTCN2018116965-appb-000026
Figure PCTCN2018116965-appb-000026
由于状态S(i)只与状态S(i-1)有关，满足马尔科夫性，对变换S的求解问题可以看作 一种特殊的隐马尔科夫模型，并利用维特比算法求解，得到S中i和j的取值，即得到一系列{i,j}点对。将Set seg_pair中的每一对中心线片段精细配准得到的{i,j}点对进行合并，构成点对集合B。至此成对片段上的采样点对应关系得以明确。图5(a)和(b)分别展示了两对中心线片段的配准结果。图中采样点用符号“+”表示，成功匹配的采样点用符号“o”表示并且赋予数字标号，
Figure PCTCN2018116965-appb-000027
的末端没有标号是因为
Figure PCTCN2018116965-appb-000028
的长度大于
Figure PCTCN2018116965-appb-000029
Figure PCTCN2018116965-appb-000030
的部分标号在
Figure PCTCN2018116965-appb-000031
上没有对应点，是因为
Figure PCTCN2018116965-appb-000032
的长度更长。 Since the state S (i) is only related to the state S (i-1) and satisfies the Markov property, the problem of solving the transformation S can be regarded as a special hidden Markov model and solved using the Viterbi algorithm to obtain The values of i and j in S are a series of {i, j} point pairs. The {i, j} point pairs obtained by fine registration of each pair of centerline segments in Set seg_pair are combined to form a point pair set B. So far, the correspondence between the sampling points on the paired segments is clear. Figures 5 (a) and (b) show the registration results of two pairs of centerline segments, respectively. The sampling points in the figure are represented by the symbol "+", and the successfully matched sampling points are represented by the symbol "o" and assigned a numerical label.
Figure PCTCN2018116965-appb-000027
Is not labeled at the end because
Figure PCTCN2018116965-appb-000028
Is longer than
Figure PCTCN2018116965-appb-000029
Figure PCTCN2018116965-appb-000030
Part of the label at
Figure PCTCN2018116965-appb-000031
There is no corresponding point on it because
Figure PCTCN2018116965-appb-000032
The length is longer.
4)利用步骤3)删除的中心线片段集合对中心线片段进一步分割及配准，得到最终的冠状动脉中心线的配准结果；4) Use the centerline segment set deleted in step 3 to further segment and register the centerline segment to obtain the final registration result of the centerline of the coronary artery;
步骤3.1)删除的中心线片段集合Set seg_del，这些中心线片段和与之对应的血管片段被认为是初始分割和提取方法遗漏的片段；此步骤用于处理输入的两个冠状动脉中心线拓扑结构不一致的部分，即Set seg_del中的遗漏中心线片段。首先对遗漏血管片段和中心线片段进行分割和提取，然后对新提取的中心线片段再做一次配准。具体步骤如下： Step 3.1) Delete the set of centerline segments Set seg_del . These centerline segments and their corresponding blood vessel segments are considered to be missing segments in the initial segmentation and extraction method; this step is used to process the input two coronary artery centerline topologies. The inconsistent part is the missing centerline segment in Set seg_del . Segment and extract the missing vessel segment and centerline segment first, and then register the newly extracted centerline segment again. Specific steps are as follows:
4.1)对遗漏的血管片段和中心线片段进行进一步分割和提取；4.1) further segmenting and extracting the missing vessel segment and centerline segment;
假设初始的源图像的冠状动脉中心线C s具有中心线片段CS s(即冠状动脉血管V s具有血管片段VS s)，且目标图像的冠状动脉中心线C t上遗漏了对应片段CS t(即V t上遗漏了对应血管片段VS t)，此步骤尝试从C t和V t所在的图像中分割出遗漏的血管片段VS t，并提取出该遗漏中心线片段CS t。首先，利用步骤3)最后得到的点对集合B，进行数学插值计算出连续的空间变换模型(例如薄板样条模型)，记作T。利用T对VS s进行空间变换，得到另一图像中的掩模VS a。在VS s和VS a的周围分别定义感兴趣区域，记作VOI s和VOI t。可以采用的一种感兴趣区域为图像上的长方体图像块，该长方体每个面平行于坐标轴平面，包含掩模且在平行于坐标轴的方向与掩模边界留有一定间隔(取值在[0,10]区间，单位为毫米)。对两个感兴趣区域进行基于灰度的配准，得到空间变换模型，记作T inter，将T inter施加于VS s，得到
Figure PCTCN2018116965-appb-000033
将
Figure PCTCN2018116965-appb-000034
作为初始化，利用图像分割方法(例如水平集方法)从目标图像中分割出血管分支VS t。 Suppose coronary centerline C s original source image segment having a center line CS s (i.e., V s having a coronary artery blood vessel segment VS s), and the missing fragment corresponding to the CS t coronary centerline C t of the target image ( i.e. missing fragment corresponding vessel VS t) on the V t, where t this step attempts from the image t and C V, blood vessel segment divided missing VS t, centerline and extracts the missing segment CS t. First, using the point pair set B finally obtained in step 3), mathematical interpolation is performed to calculate a continuous spatial transformation model (for example, a thin plate spline model), and denoted as T. Use T to spatially transform VS s to obtain a mask VS a in another image. Regions of interest are defined around VS s and VS a , respectively, denoted VOI s and VOI t . An area of interest that can be used is a cuboid image block on the image. Each face of the cuboid is parallel to the plane of the coordinate axis, contains a mask, and is spaced a certain distance from the mask boundary in a direction parallel to the coordinate axis. [0,10] interval in millimeters). Perform gray-level registration on two regions of interest to obtain a spatial transformation model, denoted as T inter , and apply T inter to VS s to obtain
Figure PCTCN2018116965-appb-000033
will
Figure PCTCN2018116965-appb-000034
As an initialization, an image segmentation method (such as a level set method) is used to segment a blood vessel branch VS t from the target image.
对Set seg_del中所有中心线片段对应的血管重复此步骤进行分割，得到对应的新分割的血管分支，同步骤1.2)一样可以利用手工标注或者半自动及全自动算法对每个新的血管分支的中心线片段进行提取。 Repeat this step for the blood vessels corresponding to all the centerline segments in Set seg_del to obtain the corresponding newly divided blood vessel branches. As in step 1.2), you can use manual labeling or semi-automatic and full-automatic algorithms to center each new blood vessel branch. Line fragments are extracted.
图6的例子图展示了通过步骤4.1)得到的更完整的冠状动脉血管的分割结果，图6(a)、图6(b)分别是左、右冠状动脉血管，其中黑色连续表面为初始输入的冠状动脉血管，离散点表面为步骤4.1)进一步分割得到的血管分支。The example diagram in Figure 6 shows the more complete coronary artery segmentation results obtained in step 4.1). Figures 6 (a) and 6 (b) are the left and right coronary arteries, respectively, where the black continuous surface is the initial input. For the coronary artery blood vessels, the surface of the discrete points is the blood vessel branch obtained by further segmentation in step 4.1).
4.2)对于步骤4.1)新提取的中心线片段，利用步骤3)的方法对其进行中心线片段配准，得到一系列新的采样点点对。利用这些点对对点对集合B进行扩充，得到新的点对集合 B further。利用B further进行数学插值，计算出连续的空间变换模型，记作T final。利用T final对源图像进行空间变换，得到与目标图像对齐的变换图像；利用T final对源图像中的冠状动脉血管V s进行空间变换，得到与V t对齐的变换血管V a；利用T final对源图像中的冠状动脉中心线C s进行空间变换，得到与C t对齐的变换中心线C a。B further、变换图像、V a和C a即为本发明得到的最终的冠状动脉配准结果。 4.2) For the newly extracted centerline segment in step 4.1), use the method of step 3) to perform centerline segment registration to obtain a series of new sampling point pairs. Use these point-to-point-to-point-set B to expand to obtain a new point-to-point set B further . Using B further for mathematical interpolation, a continuous spatial transformation model is calculated and denoted as T final . Use T final to spatially transform the source image to obtain a transformed image aligned with the target image; use T final to spatially transform the coronary arteries V s in the source image to obtain a transformed blood vessel V a aligned to V t ; use T final Perform a spatial transformation on the coronary center line C s in the source image to obtain a transformed center line C a aligned with C t . B further, converted image, V a and C a coronary final registration result obtained by the present invention is the.
图7和图8展示了本发明的配准结果的两个例子，其中图7(a)和图8(a)为经过步骤2)、步骤3)得到的初步配准结果，图7(b)和图8(b)为经过步骤4)处理后的最终配准结果。图中目标图像中的冠状动脉中心线C t用“..”线型表示表示，C s和C s利用T进行空间变换得到的中心线用“--”线型表示，得以成功配准(配准后两个中心线距离小于0.5毫米)的中心线用实线表示。由图7和图8可以看出步骤4)使更多的中心线片段得到配准。 Figures 7 and 8 show two examples of the registration results of the present invention, where Figures 7 (a) and 8 (a) are the preliminary registration results obtained after steps 2) and 3), and ) And Figure 8 (b) are the final registration results after processing in step 4). The center line C t of the coronary arteries in the target image in the figure is represented by a “..” line type, and the center line obtained by spatial transformation of C s and C s by T is represented by a “-” line type, which is successfully registered ( (The distance between the two center lines after registration is less than 0.5 mm) is represented by a solid line. It can be seen from FIG. 7 and FIG. 8 that step 4) enables more centerline segments to be registered.

Claims (1)

  1. 一种冠状动脉的自动配准方法，其特征在于，该方法包括以下步骤：A method for automatic registration of coronary arteries, characterized in that the method includes the following steps:
    1)获取冠状动脉图像，对每张冠状动脉图像进行血管分割并提取对应的中心线；具体步骤如下：1) Obtain a coronary artery image, segment each coronary artery image and extract the corresponding centerline; the specific steps are as follows:
    1.1)获取冠状动脉图像；1.1) Acquire coronary artery images;
    获取一对针对某个冠状动脉所做的计算机断层扫描血管造影图像，将两张图像中任意选取一张作为源图像，另一张作为目标图像；Obtain a pair of computed tomography angiography images for a certain coronary artery, and randomly select one of the two images as the source image and the other as the target image;
    1.2)对步骤1.1)获取的每张图像分割冠状动脉血管，并提取每个冠状动脉血管对应的中心线；将源图像中的冠状动脉血管记作V s，中心线记作C s；目标图像中的冠状动脉血管记作V t，中心线记作C t； 1.2) For each image obtained in step 1.1), segment the coronary arteries and extract the center line corresponding to each coronary artery; record the coronary arteries in the source image as V s and the center line as C s ; the target image Coronary arterial blood vessels are denoted by V t , and the center line is denoted by C t ;
    2)对步骤1)的每张冠状动脉图像中的中心线提取分叉点特征并进行匹配，得到最终匹配的分叉点点对集合；具体步骤如下：2) For each of the coronary arterial images in step 1), the branch point features are extracted and matched to obtain the final matching branch point point set; the specific steps are as follows:
    2.1)提取分叉点的特征；2.1) Extract the features of the bifurcation points;
    令B s代表源图像中冠状动脉中心线全部n s个分叉点构成的点集，B t代表目标图像中冠状动脉中心线全部n t个分叉点构成的点集；对每个冠状动脉中心线的分叉点提取对应的特征描述子，将该描述子表示为向量形式；其中，源图像的第i个分叉点的描述子向量记作
    Figure PCTCN2018116965-appb-100001
    目标图像中第i个分叉点的描述子向量记作
    Figure PCTCN2018116965-appb-100002
    Let B s represent a point set composed of all n s bifurcation points of the center line of the coronary artery in the source image, and B t represent a point set composed of all n t bifurcation points of the center line of the coronary artery in the target image; for each coronary artery The corresponding feature descriptor is extracted from the bifurcation point of the center line, and the descriptor is represented as a vector form; where the descriptor vector of the i-th bifurcation point of the source image is recorded as
    Figure PCTCN2018116965-appb-100001
    The descriptor vector of the i-th bifurcation point in the target image is written as
    Figure PCTCN2018116965-appb-100002
    2.2)获取候选匹配分叉点点对集合A cand； 2.2) Acquire the candidate matching fork point point set A cand ;
    计算B t中的每一个分叉点b t与B s中所有分叉点的描述子向量的欧氏距离，并将计算结果中最小欧氏距离对应的B s中的分叉点记作b t的对应分叉点b s，共得到n t个分叉点点对，将n t个分叉点点对的集合记作候选匹配分叉点点对集合A cand； B t calculated in each bifurcation point b B s T Euclidean described subvector all branching point distance, and the calculated result of the branching point corresponding to the minimum Euclidean distance is referred to as B s b t corresponding to the branch point b s, were obtained for n-bit t bifurcation and the n-bit t of the bifurcation, referred to as the set of candidate matching set of bit bifurcation a cand;
    2.3)计算集合A cand的属性矩阵M； 2.3) Calculate the attribute matrix M of the set A cand ;
    首先，对候A cand建立一个图G(V,E,M)，其中图的节点集合V中的每个节点对应A cand中每个分叉点点对，图的边集合E中每条边对应每两个分叉点对之间的关系，属性矩阵M的非对角线元素M(i,j)反映第i个分叉点对和第j个分叉点对之间的兼容性，M的对角线元素M(i,i)表示第i个分叉点对中两个分叉点的描述子向量相似度；M的表达式如下： First, a graph G (V, E, M) is established for candidate A cand , where each node in the graph node set V corresponds to each branch point pair in A cand , and each edge in the graph's edge set E corresponds to The relationship between every two bifurcation point pairs, the off-diagonal element M (i, j) of the attribute matrix M reflects the compatibility between the i-th bifurcation point pair and the j-th bifurcation point pair, M The diagonal element M (i, i) represents the similarity of the descriptor vector of the two bifurcation points in the i-th bifurcation point pair; the expression of M is as follows:
    Figure PCTCN2018116965-appb-100003
    Figure PCTCN2018116965-appb-100003
    Figure PCTCN2018116965-appb-100004
    Figure PCTCN2018116965-appb-100004
    Figure PCTCN2018116965-appb-100005
    Figure PCTCN2018116965-appb-100005
    Figure PCTCN2018116965-appb-100006
    Figure PCTCN2018116965-appb-100006
    其中，d i代表第i个分叉点对中两个分叉点m i1和m i2的欧氏距离，θ i是m i1和m i2方向的相对角度，
    Figure PCTCN2018116965-appb-100007
    是把m i2对应的向量当作极轴时m i1的极角；TH dist、TH θ和
    Figure PCTCN2018116965-appb-100008
    分别代表ratio(i,j)下限的阈值、|θ i-θ j|上限的阈值和
    Figure PCTCN2018116965-appb-100009
    上限的阈值；TH dist取值在[0.4,0.9]区间，TH θ取值在[30°,90°]区间，
    Figure PCTCN2018116965-appb-100010
    取值在[30°,90°]区间；     Wherein, D i represents the i-th bifurcation points of the two diverging points m i1 m i2 and the Euclidean distance, a relative angle [theta] i and m i1 m i2 direction,
    Figure PCTCN2018116965-appb-100007
    Is the polar angle m i1 m i2 when the corresponding vector as a pole shaft; TH dist, TH θ and
    Figure PCTCN2018116965-appb-100008
    Represents the threshold of the lower limit of ratio (i, j), the threshold of the upper limit of | θ i -θ j |, and
    Figure PCTCN2018116965-appb-100009
    Upper threshold; TH dist is in the interval [0.4,0.9], and TH θ is in the interval [30 °, 90 °].
    Figure PCTCN2018116965-appb-100010
    The value is in the interval [30 °, 90 °];
    2.4)计算最终匹配的分叉点点对集合A final；具体步骤如下： 2.4) Calculate the final matching fork point pair set A final ; the specific steps are as follows:
    2.4.1)建立最终匹配的分叉点点对集合A final，并将A final初始化为空集； 2.4.1) Establish the final matching fork point pair set A final and initialize A final to the empty set;
    2.4.2)对矩阵M做特征值分解，找到M的最大特征值对应的特征向量x *； 2.4.2) Eigenvalue decomposition of the matrix M to find the eigenvector x * corresponding to the largest eigenvalue of M;
    2.4.3)对x *中的元素进行由大到小的排序，得到序列x′，x′从前往后的各个元素在x *中的序号组成序列L； 2.4.3) x * of the elements in descending order, to obtain a sequence x ', x' front to back in the respective elements of x * constituent sequence number L;
    2.4.4)对L进行判定：若L为空，则输出当前的A final，求解结束；若L不为空，则进入步骤2.4.5)； 2.4.4) Judge L: If L is empty, then output the current A final and the solution ends; if L is not empty, go to step 2.4.5);
    2.4.5)取序列L中的第一个值L(1)并判定：如果x *(L(1))<ε，则输出当前的A final，求解结束，其中ε取值在[0.0000001,0.01]区间；否则，进入步骤2.4.6)； 2.4.5) Take the first value L (1) in the sequence L and determine: if x * (L (1)) <ε, then output the current A final and the solution ends, where ε takes the value [0.0000001, 0.01] interval; otherwise, proceed to step 2.4.6);
    2.4.6)如果A cand中的第L(1)个分叉点对与A final中任意一对分叉点对包含相同的分叉点，则把L(1)从L中删除，重新返回步骤2.4.4)；否则，进入步骤2.4.7)； 2.4.6) If the L (1) th fork point pair in A cand contains the same fork point as any pair of fork points in A final , delete L (1) from L and return Step 2.4.4); otherwise, go to Step 2.4.7);
    2.4.7)把A cand中的第L(1)个分叉点对加入到集合A final中，将L(1)从L中删除，重新返回步骤2.4.4)； 2.4.7) Add the L (1) bifurcation point pair in A cand to the set A final , delete L (1) from L, and return to step 2.4.4);
    3)对每张冠状动脉图像中的中心线进行更新，得到删除的中心线片段集合；利用更新后的中心线和步骤2)的最终匹配的分叉点点对集合对中心线片段进行匹配；具体步骤如下：3) Update the centerline in each coronary artery image to obtain the deleted centerline fragment set; use the updated centerline and the final matching branch point of step 2) to match the set to the centerline fragment; specifically Proceed as follows:
    3.1)对每张冠状动脉图像中的中心线进行更新，得到删除的中心线片段集合；具体步骤如下：3.1) Update the centerline in each coronary artery image to obtain the deleted centerline fragment set; the specific steps are as follows:
    3.1.1)获取C s或C t中的任一特有分叉点，分别计算该特有分叉点对应的两个支干中心线片段和一个主干中心线片段在分叉点处的方向向量，删除方向向量与主干中心线片段的方向向量夹角更大的一个支干中心线片段，同时删除该特有分叉点； 3.1.1) Obtain any unique bifurcation point in C s or C t , and calculate the direction vector of the two branch centerline segments and one trunk centerline segment at the bifurcation point corresponding to the unique bifurcation point, respectively. Delete a branch centerline segment with a larger angle between the direction vector and the direction vector of the trunk centerline segment, and delete the unique bifurcation point;
    3.1.2)从C s和C t的剩余分叉点中继续寻找特有分叉点并判定：若存在特有分叉点，则重新返回步骤3.1.1)；若不存在特有分叉点，则结束中心线更新过程，得到更新后的源图 像的冠状动脉中心线记为C′ s，更新后的目标图像的冠状动脉中心线记为C′ t，并将所有删除的中心线片段集合记作Set seg_del； 3.1.2) Continue to find unique branching points from the remaining branching points of C s and C t and determine: if there is a unique branching point, return to step 3.1.1); if there is no unique branching point, then End the centerline update process, obtain the updated centerline of the coronary arteries as C ′ s , record the updated centerline of the coronary arteries as C ′ t , and record all deleted centerline segments as Set seg_del ;
    3.2)中心线片段的匹配；3.2) Matching of centerline segments;
    利用步骤3.1)得到的C′ s和C′ t和步骤2)得到的最终匹配的分叉点点对集合A final，满足以下任意两个条件之一的中心线片段被认为是匹配的：a)夹在两对匹配分叉点点对之间的两个中心线片段；b)一端为冠状动脉起点或终点，另一端为匹配的分叉点点对，并且分叉点点对的两个分叉点的方向夹角小于设定的夹角阈值的两个中心线片段；将匹配后的中心线片段对的集合记作Set seg_pair； Using C ′ s and C ′ t obtained in step 3.1) and the set of final matching branch point pairs A final obtained in step 2), a centerline segment that meets any of the following two conditions is considered to be a match: a) Two centerline segments sandwiched between two pairs of matching bifurcation point pairs; b) one end is the origin or end of the coronary artery, the other end is the matching bifurcation point pair, and the two bifurcation points of the bifurcation point pair Two centerline segments whose direction included angle is smaller than a set included angle threshold; the set of matched centerline segment pairs is referred to as Set seg_pair ;
    3.3)中心线片段的精细配准；3.3) Fine registration of centerline segments;
    将Set seg_pair中属于源图像的中心线上片段记作Seg s，Set seg_pair中属于目标图像的中心线片段记作Seg t；对Seg s和Seg t分别进行点采样，Seg s的采样间隔为Δ s，Seg t的采样间隔为Δ t； The Set seg_pair belonging to the center line of the source image is referred to as segments Seg s, belonging to the centerline segment Set seg_pair referred to as target image Seg t; on Seg s Seg t and for each sampling point, the sampling interval is Δ Seg s s , the sampling interval of Seg t is Δ t ;
    假设Seg s和Seg t分别具有N s和N t个采样点，定义一个点集之间的离散变换模型如下： Assuming Seg s and Seg t have N s and N t sampling points, respectively, the discrete transformation model defining a point set is as follows:
    S：＝{S(i)＝j,i∈{0,1,…,N t},j∈{0,1,…,N s}}， S: = {S (i) = j, i∈ {0,1,…, N t }, j∈ {0,1,…, N s }},
    其中，S(i)是Seg t的第i个采样点的状态，S(i)＝j表示Seg t的第i个采样点与Seg s的第j个采样点相对应； Among them, S (i) is the state of the i-th sampling point of Seg t , and S (i) = j means that the i-th sampling point of Seg t corresponds to the j-th sampling point of Seg s ;
    目标函数为：The objective function is:
    Figure PCTCN2018116965-appb-100011
    Figure PCTCN2018116965-appb-100011
    其中，Sim1代表图像相似度，Sim2代表几何相似度，ω 1和ω 2分别为Sim1和Sim2所占的权重，ω 1和ω 2的取值都在[0,1]区间，且ω 1与ω 2的和是1； Among them, Sim1 represents image similarity, Sim2 represents geometric similarity, ω 1 and ω 2 are weights occupied by Sim1 and Sim2, respectively, and the values of ω 1 and ω 2 are in the interval [0,1], and ω 1 and The sum of ω 2 is 1.
    图像相似度Sim1是利用特征描述子进行衡量的，数学表达式为：Image similarity Sim1 is measured using feature descriptors, and the mathematical expression is:
    Figure PCTCN2018116965-appb-100012
    Figure PCTCN2018116965-appb-100012
    其中，
    Figure PCTCN2018116965-appb-100013
    是Seg t的第i个采样点的描述子，
    Figure PCTCN2018116965-appb-100014
    是Seg s的第S(i)个采样点的描述子；     among them,
    Figure PCTCN2018116965-appb-100013
    Is the descriptor of the i-th sampling point of Seg t ,
    Figure PCTCN2018116965-appb-100014
    Is the descriptor of the S (i) th sampling point of Seg s ;
    几何相似度Sim2表达式为：The geometric similarity Sim2 expression is:
    Figure PCTCN2018116965-appb-100015
    Figure PCTCN2018116965-appb-100015
    求解S中i和j的取值，即得到一系列{i,j}点对；将Set seg_pair中的每一对中心线片段精细配准得到的{i,j}点对进行合并，构成点对集合B； Solve the values of i and j in S to obtain a series of {i, j} point pairs; combine the {i, j} point pairs obtained by fine registration of each pair of centerline segments in Set seg_pair to form points For set B;
    4)利用步骤3)删除的中心线片段集合对中心线片段进一步分割及配准，得到最终的冠状动脉中心线的配准结果；具体步骤如下：4) Use the centerline segment set deleted in step 3 to further segment and register the centerline segment to obtain the final registration result of the coronary centerline;
    4.1)假设源图像的冠状动脉中心线C s具有中心线片段CS s，即冠状动脉血管V s具有血管片段VS s，且目标图像的冠状动脉中心线C t上遗漏了对应片段CS t，即V t上遗漏了对应血管片段VS t，利用步骤3)得到的点对集合B，进行数学插值计算出连续的空间变换模型，记作T；利用T对VS s进行空间变换，得到另一图像中的掩模VS a，在VS s和VS a的周围分别定义感兴趣区域，记作VOI s和VOI t；对两个感兴趣区域进行基于灰度的配准，得到空间变换模型，记作T inter，将T inter施加于VS s，得到
    Figure PCTCN2018116965-appb-100016
    将
    Figure PCTCN2018116965-appb-100017
    作为初始化，利用图像分割方法从目标图像中分割出血管分支VS t；     Coronary centerline C s 4.1) assuming that the source image segment having a center line CS s, i.e. V s having a coronary artery blood vessel segment VS s, and the missing fragment corresponding to the CS t coronary centerline C t of the target image, i.e., V t corresponding to the omission of the blood vessel segment VS t, using step 3) the obtained set of points B, continuous spatial transformation model to calculate the mathematical interpolation, denoted by T; T of VS s using spatial transformation to give another image In the mask VS a , the regions of interest around VS s and VS a are defined respectively as VOI s and VOI t ; gray-scale registration is performed on the two regions of interest to obtain a spatial transformation model, which is denoted as T inter , apply T inter to VS s and get
    Figure PCTCN2018116965-appb-100016
    will
    Figure PCTCN2018116965-appb-100017
    As an initialization, an image segmentation method is used to segment a blood vessel branch VS t from the target image;
    对Set seg_del中所有中心线片段对应的血管重复此步骤，得到对应的新分割的血管分支，然后重复步骤1.2)，得到所有新分割的血管分支的中心线片段； Repeat this step for all blood vessels corresponding to the centerline segment in Set seg_del to obtain the corresponding newly segmented blood vessel branches, and then repeat step 1.2) to obtain the centerline segments of all newly segmented blood vessel branches;
    4.2)对于步骤4.1)新提取的中心线片段，重复步骤3)对其进行中心线片段配准，得到一系列新的采样点点对并添加进集合B中，得到新的点对集合B further；利用B further进行数学插值，计算出连续的空间变换模型，记作T final；利用T final对源图像进行空间变换，得到与目标图像对齐的变换图像；利用T final对源图像中的冠状动脉血管V s进行空间变换，得到与V t对齐的变换血管V a；利用T final对源图像中的冠状动脉中心线C s进行空间变换，得到与C t对齐的变换中心线C a；B further、变换图像、V a和C a即为最终得到的冠状动脉中心线的配准结果。 4.2) For step 4.1) newly extracted centerline segments, repeat step 3) to perform centerline segment registration to obtain a series of new sampling point pairs and add them to set B to obtain a new set of point pairs B further ; Use B further for mathematical interpolation to calculate a continuous spatial transformation model, denoted as T final ; use T final to spatially transform the source image to obtain a transformed image aligned with the target image; and use T final to coronary artery blood vessels in the source image V s performs spatial transformation to obtain a transformed blood vessel V a aligned with V t ; T final is used to spatially transform the coronary center line C s in the source image to obtain a transformed center line C a aligned with C t ; B further , converted image, the registration result of coronary centerline C a and V a is the finally obtained.
PCT/CN2018/116965 2018-06-21 2018-11-22 Automatic registration method for coronary arteries WO2019242227A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810643770.6A CN108898626B (en) 2018-06-21 2018-06-21 A kind of autoegistration method coronarius
CN201810643770.6 2018-06-21

Publications (1)

Publication Number Publication Date
WO2019242227A1 true WO2019242227A1 (en) 2019-12-26

Family

ID=64345225

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/116965 WO2019242227A1 (en) 2018-06-21 2018-11-22 Automatic registration method for coronary arteries

Country Status (2)

Country Link
CN (1) CN108898626B (en)
WO (1) WO2019242227A1 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11210786B2 (en) * 2020-01-07 2021-12-28 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11302002B2 (en) 2020-01-07 2022-04-12 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11317883B2 (en) 2019-01-25 2022-05-03 Cleerly, Inc. Systems and methods of characterizing high risk plaques
US11861833B2 (en) 2020-01-07 2024-01-02 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11922627B2 (en) 2022-03-10 2024-03-05 Cleerly, Inc. Systems, devices, and methods for non-invasive image-based plaque analysis and risk determination

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109745120B (en) * 2018-12-24 2020-07-31 罗雄彪 Image registration conversion parameter optimization method and system
CN109978829B (en) * 2019-02-26 2021-09-28 深圳市华汉伟业科技有限公司 Detection method and system for object to be detected
CN109993730B (en) 2019-03-20 2021-03-30 北京理工大学 3D/2D blood vessel registration method and device
CN110197495B (en) * 2019-05-30 2021-03-09 数坤（北京）网络科技有限公司 Adjusting method and device for blood vessel extraction
CN111242958B (en) * 2020-01-15 2022-04-08 浙江工业大学 Carotid artery cascade learning segmentation method based on structural feature optimization
CN111932497B (en) * 2020-06-30 2021-02-09 数坤（北京）网络科技有限公司 Coronary artery identification method and device

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050249391A1 (en) * 2004-05-10 2005-11-10 Mediguide Ltd. Method for segmentation of IVUS image sequences
CN101763642A (en) * 2009-12-31 2010-06-30 华中科技大学 Matching method for three-dimensional coronary angiography reconstruction
CN104574413A (en) * 2015-01-22 2015-04-29 深圳大学 Blood vessel bifurcation extracting method and system of lung CT picture
CN105741251A (en) * 2016-03-17 2016-07-06 中南大学 Blood vessel segmentation method for liver CTA sequence image
CN105761254A (en) * 2016-02-04 2016-07-13 浙江工商大学 Image feature based eyeground image registering method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050249391A1 (en) * 2004-05-10 2005-11-10 Mediguide Ltd. Method for segmentation of IVUS image sequences
CN101763642A (en) * 2009-12-31 2010-06-30 华中科技大学 Matching method for three-dimensional coronary angiography reconstruction
CN104574413A (en) * 2015-01-22 2015-04-29 深圳大学 Blood vessel bifurcation extracting method and system of lung CT picture
CN105761254A (en) * 2016-02-04 2016-07-13 浙江工商大学 Image feature based eyeground image registering method
CN105741251A (en) * 2016-03-17 2016-07-06 中南大学 Blood vessel segmentation method for liver CTA sequence image

Cited By (32)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11317883B2 (en) 2019-01-25 2022-05-03 Cleerly, Inc. Systems and methods of characterizing high risk plaques
US11759161B2 (en) 2019-01-25 2023-09-19 Cleerly, Inc. Systems and methods of characterizing high risk plaques
US11751831B2 (en) 2019-01-25 2023-09-12 Cleerly, Inc. Systems and methods for characterizing high risk plaques
US11642092B1 (en) 2019-01-25 2023-05-09 Cleerly, Inc. Systems and methods for characterizing high risk plaques
US11350899B2 (en) 2019-01-25 2022-06-07 Cleerly, Inc. Systems and methods for characterizing high risk plaques
US11737718B2 (en) 2020-01-07 2023-08-29 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11751826B2 (en) 2020-01-07 2023-09-12 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11302002B2 (en) 2020-01-07 2022-04-12 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11302001B2 (en) 2020-01-07 2022-04-12 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11367190B2 (en) 2020-01-07 2022-06-21 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11501436B2 (en) 2020-01-07 2022-11-15 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11288799B2 (en) 2020-01-07 2022-03-29 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11660058B2 (en) 2020-01-07 2023-05-30 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11672497B2 (en) 2020-01-07 2023-06-13 Cleerly. Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11690586B2 (en) 2020-01-07 2023-07-04 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11730437B2 (en) 2020-01-07 2023-08-22 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11210786B2 (en) * 2020-01-07 2021-12-28 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11751830B2 (en) 2020-01-07 2023-09-12 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11315247B2 (en) 2020-01-07 2022-04-26 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11751829B2 (en) 2020-01-07 2023-09-12 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11276170B2 (en) 2020-01-07 2022-03-15 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11232564B2 (en) 2020-01-07 2022-01-25 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11766230B2 (en) 2020-01-07 2023-09-26 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11766229B2 (en) 2020-01-07 2023-09-26 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11779292B2 (en) 2020-01-07 2023-10-10 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11832982B2 (en) 2020-01-07 2023-12-05 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11861833B2 (en) 2020-01-07 2024-01-02 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11896415B2 (en) 2020-01-07 2024-02-13 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11969280B2 (en) 2020-01-07 2024-04-30 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11967078B2 (en) 2020-01-07 2024-04-23 Cleerly, Inc. Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking
US11948301B2 (en) 2022-03-10 2024-04-02 Cleerly, Inc. Systems, devices, and methods for non-invasive image-based plaque analysis and risk determination
US11922627B2 (en) 2022-03-10 2024-03-05 Cleerly, Inc. Systems, devices, and methods for non-invasive image-based plaque analysis and risk determination

Also Published As

Publication number Publication date
CN108898626B (en) 2019-09-27
CN108898626A (en) 2018-11-27

Similar Documents

Publication Publication Date Title
WO2019242227A1 (en) Automatic registration method for coronary arteries
Groher et al. Deformable 2D-3D registration of vascular structures in a one view scenario
Robben et al. Simultaneous segmentation and anatomical labeling of the cerebral vasculature
JP5072449B2 (en) Medical image processing apparatus and medical image processing method
KR102050649B1 (en) Method for extracting vascular structure in 2d x-ray angiogram, computer readable medium and apparatus for performing the method
CN106097298A (en) The coronary artery automatic segmentation divided based on spherical space and anatomic landmarks method
Hoffmann et al. Electrophysiology catheter detection and reconstruction from two views in fluoroscopic images
CN102722882A (en) Elastic registration method of CAG image sequence
Wu et al. Fast catheter segmentation from echocardiographic sequences based on segmentation from corresponding X-ray fluoroscopy for cardiac catheterization interventions
CN103020958B (en) A kind of blood vessel automatic matching method based on curvature scale space
Cardenes et al. 3D reconstruction of coronary arteries from rotational X-ray angiography
Fang et al. Greedy soft matching for vascular tracking of coronary angiographic image sequences
CN107392891B (en) Blood vessel tree extraction method, device, equipment and storage medium
Gu et al. Vision–kinematics interaction for robotic-assisted bronchoscopy navigation
Nazir et al. Living donor-recipient pair matching for liver transplant via ternary tree representation with cascade incremental learning
Guo et al. Coarse-to-fine airway segmentation using multi information fusion network and CNN-based region growing
Wang et al. Naviairway: a bronchiole-sensitive deep learning-based airway segmentation pipeline for planning of navigation bronchoscopy
CN111260704A (en) Vascular structure 3D/2D rigid registration method and device based on heuristic tree search
Qi et al. Examinee-examiner network: Weakly supervised accurate coronary lumen segmentation using centerline constraint
Perchet et al. Advanced navigation tools for virtual bronchoscopy
Shen et al. Automatic cerebral artery system labeling using registration and key points tracking
Du et al. Accurate non-rigid registration based on heuristic tree for registering point sets with large deformation
Zeng et al. Towards accurate and complete registration of coronary arteries in CTA images
Brieva et al. Coronary extraction and stenosis quantification in X-ray angiographic imaging
US10957057B2 (en) Post-mapping automatic identification of pulmonary veins

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18923683

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18923683

Country of ref document: EP

Kind code of ref document: A1