Study On Structure Identification And Matching Of Coronary Arteries In X-ray Angiograms

Coronary Artery Disease (CAD) is a common cardiovascular disease. According to thereport of the World Health Organization, CAD accounts for12.8percent of deathsworldwide (about700million people) each year, which is the "No.1killer" threateninghuman health. Due to the high spatial and temporal resolution of X-ray coronaryangiography imaging, it has become the "gold standard" for the clinical diagnosis ofcoronary artery stenosis. Since the X-ray and contrast agent are harmful to the human body,doctors tend to collect only a limited number of angiograms for the purpose of diagnosisand treatment of lesions. The overall spatial topological structure of the coronary artery isabsent in the two dimensional angiography. Meanwhile, because of the cross and occlusionof the coronary, it is difficult to perform the recognition and matching for the vesselstructure in different angiography views. In this paper, based on the study of recognitionand matching of the coronary artery structure, the correlation between the sequences ofangiograms can be obtained, and the coronary artery structure can be reconstructed in thethree-dimensional space. All the results could provide auxiliary diagnosis and treatment forthe clinical CAD, reduce the subjectivity for disease judgment, as well as improve theefficiency. The main contributions of this paper are as follows:(1) In order to solve the problem that manual seed point setting may cause instabilityfor extracting coronary artery structure, an automatic seed points extraction method basedon the ridge points detection is proposed in this paper. Initially, the continuity of Hessianmatrix eigenvalues and the corresponding eigenvectors is proved on multi-scale vascularenhancement image. Then, on the basis of the proposed ridge existence theorem, an explicitdiscriminant is designed to extract the candidate seed points on the angiograms. At last, aseed point selection method is proposed based on the gray distribution of the detected seedpoints’ neighbors. The pseudo seed points can be removed and all the seed points arelocated inside the coronary artery.(2) Due to the uneven gray level distributions of vessels and the complexity of thetopology, large errors may be brought in the process of the structure detecting. In this paper,an adaptive tracking method of coronary artery extraction is proposed. Firstly, the initial tracking direction of each seed point is detected based on its location. Secondly, ridgepoints are tracked on the vascular enhancement image to form the initial center line points.Thirdly, according to the intensity profile perpendicular of the tracking directions, ridgepoints are adjusted to the geometrical centerline points adaptively. Finally, all the trackedcenterlines are merged and classified to remove the false tracking results by the connectingcomponents on the vascular structures, so that precise coronary structure can be obtained.(3) Due to the movement of heart and different angiography view, the correspondingrelationship between coronary arteries is uncertain. In this paper, a recognition method ofthe coronary artery based on the curvature scale space(CSS) is proposed. Initially, CSSimage is obtained by convoluting the vascular centerline with a Gaussian function on thedifferent scale levels. To match the CSS image of the open curve, the open contour andclosed contour of CSS image are classified and matched. And in order to match two imageswith larger views, we introduce a method based on the transmissibility of curve matchingwhich can improve the identification degree of each vessel. Therefore, coronary arteries inany two angiograms of the rotational sequence can be identified.(4) For traditional coronary artery methods, there are lots of error results when usingepipolar lines to find the matching relationship between vessels, because epipolar lines areoften parallel to vessels or inserted with more than one branch. In this paper, a soft coronaryartery structure matching algorithm based on the geometry constraint and projectionconstraint is proposed. Firstly, the matching energy equation between coronary arteries indifferent angiography views is constructed based on the geometry constraint and projectionconstraint. Then, the deterministic annealing method is used to optimize the matchingfunction. According to the shape context describer, the optimal correspondence can beobtained by iteratively reducing the temperature. Finally, on the basis of imaging model ofthe angiography system, the skeleton structure of the coronary artery is reconstructed andelliptical cross section of vessels in three-dimensional space is fitted to form the surface ofthe coronary artery structure.