| Cardiovascular disease(CVD)refers to conditions involving blocked blood vessels e-volved by atherosclerosis that can lead to the heart attack and stroke,and it is the major cause of morbidity and mortality in the world.In China,one in five residents suffers from CVD,and moreover its prevalence has been continuously increasing.Thus,it is highly needed to develop the computer-aided method to predict,diagnose and treat CVD.The diagnosis of CVD based on medical imaging is one crucial research field in the computer-aided diagnosis,and it has also been attracting the attention from clinical experts.This thesis aims to investigate how to apply the medical image analysis and machine learning technology to accurately,robustly and fast extract the CVD-related information from medical images,in order to facilitate the subsequent clinical diagnosis and treatment of CVD.Common image-based CVD diagnosis technologies include:1)intravascular ultrasound(IVUS).It can help to measure the plaque burden for guid-ing the choice of the subsequent interventional therapy;2)computed tomography angiography(CTA).It can provide useful information for reconstructing three-dimensional vessel structure,and moreover help the hemodynamics simulation for determining the degree of ischemia;3)carotid ultrasound.It can record the information of vessel motion trajectory,which has been recognized as a new predictor of CVD.The main contents of this thesis include:·Developing a method for the vessel border extraction in IVUS images.Due to the influence from the high-density calcified plaque with acoustic shadowing on the ap-pearance of the vessel borders,the proposed method firstly recognizes the IVUS image with/without the high-density calcification.Then the region growing algorithm and the curve optimization are applied on the two classes of IVUS images to extract the media-adventitia border.As regards the vascular anatomy,the extraction of media-adventitia border can narrow the region of interest of the vessel lumen.Because the image contrast between the lumen region and the plaque region may change in the vessel with differ-ent levels of atherosclerotic disease.Thus,the auto-encoder artificial neural network is applied to extract the image features of the lumen and non-lumen regions.By these features,the lumen border can be extracted.The experimental results demonstrate the effectiveness of the proposed method.·Developing a method for the coronary segmentation and its three-dimensional recon-struction in CTA images.As regards the constancy of aorta morphology in CTA images,the proposed method extracts the aorta region at first,and then detects the aorta-coronary intersection for determining the location of the coronary root.Then,the cross-sectional CTA images are projected from the direction along the vessel in order to generate the image within longitudinal vessel region.By the dynamic programming,the longitudi-nal vessel borders can be extracted,which provide the information to reconstruct the three-dimensional coronary morphology.The experimental results demonstrate the ef-fectiveness of the proposed method.·Developing a method for the motion tracking of the vessel wall in the carotid ultra-sound sequence.Because the motion tracking of vessel wall can be considered as an ill-conditional problem,the prior information is required for constraining this motion.The proposed method constrains this motion by the linear elastic model,and then represents it by the state-space approach.The observation of the state-space equations is obtained by the block matching algorithm.The prediction of the motion state is acquired by the H∞ filter,which can handle the problem of the unknown noise distribution.Finally,the locations of the target tissue in all ultrasound frames are connected to generate the mo?tion trajectory.The experimental results demonstrate the effectiveness of the proposed method. |
