טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentHoltzman Gazit Michal
SubjectSegmentation of Thin Structures in Volumetric Medical Images
DepartmentDepartment of Electrical Engineering
Supervisor Professor Ron Kimmel
Full Thesis textFull thesis text - English Version


Abstract

Medical image analysis is the process through which trained personnel interpret visual medical information. In this research, the goal is to extract thin objects such as blood vessels from volumetric images and develop a tool that allows medical experts to rapidly view the blood vessels as three dimensional (3D) objects. We present a new segmentation method for 3D medical images based on variational principles. The aim of the segmentation process is to extract a three dimensional implicit surface for each anatomic objects, e.g. vessels or bones. The algorithm we use is based on geometric active surfaces. These are surfaces that evolve according to geometric partial differential equations until they stop at the boundaries of the objects. Our algorithm uses a weighted sum of three forces: an alignment term which brings the evolving surface to the edges of the desired object, a minimal variance term which maintains homogeneity inside and outside the object, and a geodesic active surface term which is used for regularization. By using these terms we are able to extract very thin structures and allow medical experts to view them in 3D. In order to accelerate movement of the evolving surface we use an efficient numerical scheme, coupled with a narrow band approach and Sethian's Fast Marching algorithm. We also address the following problem: in medical images such as CTA (computerized tomography angiography) images, the bones and the vessels usually have similar gray values and usually they are also physically adjacent. When thresholding an image that includes blood vessels as bones, they are usually extracted as the same object. Here, using ideas adopted from vector quantization theory, we present a hierarchical segmentation method using variational tools, which allow us to separate bones from the blood vessels and extract them as two separate 3D objects.