Research On 2D/3D Coronary Artries Segmentation And Registration

Coronary artery disease(CAD)is a leading cause of death worldwide.Conventional coronary angiography(CCA)is still the first screening modality on potential coronary artery disease,and it is the reference gold standard imaging technique as before due to its real-time and high resolution.However,due to the projection nature,there are overlapping and covering between vessels.In order to overcome these disadvantages,doctors can use the three-dimensional coronary computed tomography angiography(CCTA)to aid vessel tracking.Nevertheless,these two images cannot be used together.With the application and development of artificial intelligence and image processing,they provide new ways to solve these problems quantitatively,objectively and fully automatically.In computer-aided diagnosis and treatment of coronary artery disease,the segmentation can provide the basis for subsequent objectively quantitative vascular stenosis degree diagnosis,and registration can provide a modality fusion in interventional therapy.Thus segmentation and registration become the most important steps in the diagnosis and treatment.However,they are grand challenges due to image noise,artifacts,non-uniform staining in the CCA and CCTA images.Aiming at the issues aforementioned,the research focuses on 2D vascular segmentation and registration on 2D/3D modalities.In addition,because the centerline of the blood vessel is a good reference for regional positioning in the 2D segmentation,centerline tracking on 2D is implemented firstly based on connectedness matching and principal curves.And followed the accurate extraction of 2D centerline,the 2D segmentation of vessel is proposed based on multi-domains remapping and quantile regression.Furthermore,in order to take advantage of 3D vascular centerlines to further registration,a method based on heart region isolation and vessel identification is proposed.Finally,a registration method based on tree topology consistency is proposed on 2D CCA and 3D CCTA.This work mainly includes the following four parts:Firstly,centerline extraction method based on connectedness clustering and principal curves in 2D coronary arteries is proposed.Because the tubular structure is not obvious in vessel image,and the detection performance of branch points in vessels will be influenced.This paper tracks the coronary vascular centerline by means of vascular branch separation and regathering.In the beginning step,we refine Frangi's detection result to compensate the vesselness measure,to ensure connectivity and to eliminate artifacts as much as possible.Then,a vessel connectedness-based matching method to identify the each blood vessel is studied.In the end,in order to handle the gaps and holes in enhanced vessel image,a robust method based on principle curves is employed to extract the centerlines.Secondly,coronary vascular segmentation method based on multi-domains remapping and quantile regression under the reference of the centerline in 2D is proposed.In terms of compressed image in 3D,the two-dimensional image stacks the complex textures.Moreover,due to the non-uniform illumination and dissipation of contrast agent,the intensities on the boundary are not stable,and the traditional linear regression model or least squares regression cannot achieve a robust result.A robust and automatic vessel segmentation method which combined the intensity and frequency boundary information is proposed.It includes a more accuracy boundary remapping with adaptive weights and robust discrepancy correction via distance balance and quantile regression.Thirdly,centerline extraction method based on DBSCAN and Frangi's distribution of symmetry in 3D coronary arteries is proposed.Although there is no overlapping in the 3D CCTA images,the whole lung and liver area can be enhanced as same as the heart due to its enhancement approach.Therefore,the heart region is isolated firstly based on DBSCAN.Aiming at the false response of tubular-like structure on heart inner wall,the true vessels can be distinguished from artifacts via an identification function based on their different anisotropic distributions of Frangi's vesselness.To the end,a 3D directional connectedness-related matching and principal curves are employed to extract the vascular centerlines.Finally,vessel registration method based on tree topology consistency matching for 2D/3D images is proposed.Due to the changes of coronary artery shape and position caused by heart beating and respiration,the registration between real-time CCA and static CCTA is a no rigid issue.Referred by the 2D and 3D centerlines aforementioned,this paper proposes a two-level(coarse and fine)registration approach for 2D/3D coronary arteries.In the coarse level,the approach performs many-to-many nodes matching constrained by tree topological consistency instead of the existing one-to-one matching,in the fine level,the paired edges are matched by using the DTW and TPS.In this study,two modalities(2D CCA images and CCTA computed tomography images)are used as study objects of coronary artery disease(CAD).Two key technologies(Segmentation and 2D/3D registration algorithm)under the reference of vessel centerline tracking in computer-aided diagnosis are studied.And a good foundation for subsequent automated diagnosis and better service for the prevention and treatment of coronary artery disease are expected.