Research On Coronary Heart Disease Detection Based On Computed Tomography Angiography

Coronary heart disease(CHD)is one of the top death-leading diseases around the word.Early diagnose and treatment seem to be the only effective way to reduce the morbidity and mortality of CHD.With sub-millimeter-level spatial resolution and excellent image quality,modern computed tomography(CT)provides the base for examination,characterization and quantification of CHD.In recent years,with the fast development of computer vision technology,the object of computer-aided diagnosis of CHD has been transferred from traditional coronary angiography(CAG)to coronary computed tomography angiography(CCTA).The main work of this thesis is to develop the methods of coronary heart disease detection based on CCTA image data,including segmentation and quantification of coronary calcium plaque and coronary lesion cross-section detection.The major work of this thesis can be summarized as these five parts:1.Designed the methods of coronary heart disease detection,including the frameworks of coronary calcium plaque quantification and coronary lesion cross-section detection.2.Designed an automatic coronary vessel tree extraction method.Firstly,the ascending aorta is segmented slice by slice.Then,a snake model is used to fit the ascending aorta contour to find the coronary joint.Finally,a region-grow algorithm is performed to segment the coronary artery.The segment result of coronary artery provides the base for coronary calcium detection and quantification.3.Proposed an automatic method for coronary calcium plaque segmentation and quantification based on fuzzy C-means(FCM)clustering algorithm and self-adapting threshold determination.Firstly,FCM clustering algorithm is used to divide the coronary artery into different candidate volumes,which improves anti-noise ability of the algorithm;and a robust threshold determination algorithm based on the histogram is used to extract calcium plaques among those candidate volumes acquired.Finally,the coronary calcium is quantified according to the segment result.4.Proposed a coronary lesion cross-section detection method based on one-class support vector machine(OCSVM).Firstly,each coronary cross-section is described by a multi-scale image feature vector.After feature selection,the OCSVM algorithm is used to detect the coronary lesion cross-section to achieve a high sensitivityrate of lesion recognition.5.During the coronary lesion cross-section detection,a gradient flux-based coronary cross section resample method is proposed to improve the feature quality,and a mutual information based method combined with redundancy removal is adopted to select the target features,which has efficiently improved the recognition rate of OCSVM.