Coronary artery stenosis detection and visualization / Tang Sze Ling

Tang, Sze Ling (2015) Coronary artery stenosis detection and visualization / Tang Sze Ling. PhD thesis, University of Malaya.

PDF (Full Text) Restricted to Repository staff only until 01 January 2018. Download (1867Kb)

Abstract

The manual operation of Coronary Artery Disease (CAD) diagnosis from volumetric Computed Tomography Angiography (CTA) is a very time consuming process and lead to inter- and intra-observer variability of the readers. This thesis investigates the computeraided CAD diagnosis in order to reduce reading time and manual operators’ variability. Particularly, this thesis aims to (1) minimize the interaction level of the computer-aided procedure, (2) visualize the arteries without obstruction and (3) detect the stenosis with consistent observation using the computer-aided procedure. First, a novel idea, the Orthogonal Planar Search (OPS) mechanism is proposed for coronary artery centerline extraction. Particularly, the OPS is conducted locally on the identified orthogonal plane (axial, sagittal or coronal). A metrics to select the best representation of vessel cross section is incorporated with tracking direction estimation throughout the centerline extraction process. The centerline extraction requires predefined root seed (ostia) for each coronary artery tree. The proposed method benchmarked with the state-of-the-art methods and achieved comparable results. The experimental results show that the proposed method performs the best tradeoff between accuracy and efficiency. Second, the stretched Curved Planar Reformation (CPR) is adapted to overcome the issue on obstruction. The sequence of points is resampled evenly to preserve the length of centerline. With this, the points are projected directly within a single image without other calculation required. In order to avoid the issue of stenosis underestimation or overestimation from a single image, two CPR projection angles are employed for the coronary artery evaluation. Third, a stenosis detection algorithm from two images is proposed in this thesis. In contrast to the conventional methods which perform detection from a single image, the stenosis detection algorithm using two images from various view angles to avoid false positive (stenosis overestimated) and false negative (stenosis underestimated). The performance evaluation results show that the stenosis detection algorithm performs better average sensitivity than several state-of-the-art algorithms.

Actions (For repository staff only : Login required)