AUTOMATIC DETECTION OF SMALL VESSELS IN MEDICAL IMAGES

Abstract

Diseases that affect human vasculature are among the most concerned public health problems worldwide. An accurate delineation of vessel structures in medical images will provide guidance for disease diagnosis and surgical planning. Due to the complex structures of the vessels, automatic vessel segmentation has been a challenging task for many years. This thesis addresses the challenges in vessel segmentation by incorporating the local directional information of vessels into the segmentation framework. A robust edge tangent flow method is proposed to accurately represent the vessel's direction in centerline tracking. A novel vessel enhancement technique is used with edge constraint for retinal vessel segmentation. Gradient vector flow is used with region-based graph cuts to segment liver vessels. The results show better performance in extracting small vessels, which proves the effectiveness of applying the local directional information. The author also explores the convolutional neural network as an emerging framework for vessel segmentation applications.