Quantitative analysis of coronary artery from computed tomography angiography images
Date of Issue2016
School of Physical and Mathematical Sciences
Coronary artery diseases (CAD) have been the major cause of death worldwide, accounting for about 17% of total US deaths. Automatic and accurate reconstruction of coronary arteries has been one of the aims of medical imaging, which is used to detect and quantify the potential stenosis. In clinical practice, there exist numerous ways of quantifying coronary artery lumen for stenosis detection . To evaluate the severity of a coronary stenosis, two main different approaches at present can be used: by evaluating longitudinal diameter reduction (angiography imaging) or by measuring the reduction of lumen (Computed Tomography (CT) imaging) [2, 3]. Most commonly, diameter stenosis is measured because of the popularity of projection-based invasive angiography, which is measured as the ratio of the reduction in vessel diameter to the proximal normal diameter . The estimate of area stenosis has distinct advantages and priorities, i.e., less dependency on reference site and viewing angle. When correlating with myocardial perfusion assessment or coronary blood flow, lumen area evaluation is expected to be preferable to diameter quantification [5, 6, 7, 8]. It is intuitive that diameter assessment is difficult when considering irregular arterial lumen shapes at lesion sites . Since lumen shapes of the lesion are irregular, diameter assessment may distort the true extent of stenosis in many cases . An automated, robust and efficient method to approach most Computed Tomography Angiography (CTA) images is still in demand due to the intrinsic ill-posedness and the external stenosis quantification difficulties such as manual measurement, low contrast and huge processing time, which motivates the studies in this thesis. This thesis includes several works addressing three challenging issues in the stenosis quantification topic. The proposed stenosis quantification framework used in this thesis generalizes the framework used in earlier methods. It not only preserves all theoretic merits of existing methods, but also adapts to numerical solvers in current segmentation and centerline extraction methods. In three studies, novel mathematical models are proposed and validated for the quantification purpose; efficient minimization and shortest path tools are developed; potential applications are suggested through a wide spectrum of 3D CTA images. The methods not only outperform existing methods on common stenosis quantification problems, but also first address cross sectional lumen planes cutting over vessel surface mesh. Firstly, it is a prerequisite to precisely segment coronary arteries for the quantitative evaluation of the severity of potential stenosis. An automatic coronary artery segmentation approach based on Hessian filter and connected components is proposed. A model-based vesselness filter is first used to segment coronary arteries in CTA images. Next a threshold operation based on the intensities of coronary artery is performed to create the binary representations of these structures. Besides, post-processing steps such as filling hole and surface smoothing are also performed to accurately extract the centerline. Secondly, a fast and accurate technique for coronary centerline extraction is developed. The fast marching method is first applied to calculate the timecrossing map. In the second step, a branch tracking procedure based on the Runge-Kutta method is performed. The proposed method can reduce the processing time by about 10% and achieve higher accuracy compared to the existing approaches. In addition, the application of this new approach is demonstrated on various CT images. The subvoxel precise centerlines are shown to be reliable and accurate. At last, an accurate and reproducible coronary artery cross sectional lumen area measurement algorithm at sub-voxel accuracy from human cardiac CTA images is developed. The proposed algorithm depends on centerline extraction and coronary artery surface mesh generation. Moreover, the approach is applied over morphometric data of real patients and the accuracy of the obtained cross sectional lumen area (CSLA) is validated against intravascular ultrasound (IVUS). In summary, the new approach for coronary artery stenosis quantification consists of three steps, 3D segmentation, centerline extraction and cross sectional lumen plane cutting. The application of this new algorithm is demonstrated on various CT images and the accuracy of the obtained lumen area is validated based on IVUS. This approach will facilitate quantitative evaluation of the severity of potential stenosis and help in clinical diagnosis of CAD.