Vascular segmentation in magnetic resonance angiography

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Vascular segmentation in magnetic resonance angiography"

By

Mr. Wai-Kong Law


Abstract

Clinical assessment of vasculatures is essential for the detection and 
treatment of vascular diseases which can be potentially fatal. To 
facilitate clinical assessment of blood vessels, there is a growing need 
of developing computer assisted vessel segmentation schemes based on 
magnetic resonance angiographic (MRA) images. A vast number of approaches 
have been proposed in the past decade for the segmentation of vascular 
structures in MRA images. These approaches were devised according to 
different assumptions on the shape of blood vessels and different 
underlying prior knowledge about the desired imaging modalities. The 
development of these approaches aims at delivering more accurate and 
robust segmentation results. Nonetheless, these approaches face different 
technical challenges that prohibit them from being widely employed in the 
clinical environment. The challenges include significant contrast 
variation of vessel boundaries in MRA images, the excessive computation 
time required by some algorithms and the complicated geometry of vascular 
structures. These challenges motivate us to propose three novel edge 
detection and vascular segmentation methods.

In the first proposed method, vessel segmentation is performed grounded on 
the edge detection responses given by the weighted local variance-based 
edge detector. This detector is robust against large intensity contrast 
changes and capable of returning accurate detection responses on low 
contrast edges. Our second method is an efficient implementation of a well 
founded vessel detection approach. The proposed efficient implementation 
is a thousand times faster than the conventional implementation without 
segmentation performance deterioration. The third method is a curvilinear 
structure descriptor which is robust against the disturbance induced by 
closely located objects. Preliminary experimental results show that the 
proposed methods are very suitable for vascular segmentation in MRA 
images.


Date:			Thursday, 22 July 2010

Time:			10:30am – 12:30pm

Venue:			Room 3501
 			Lifts 25/26

Chairman:		Prof. Limin Zhang (CIVL)

Committee Members:	Prof. Albert Chung (Supervisor)
 			Prof. Pedro Sander
 			Prof. Chiew-Lan Tai
                         Prof. Oscar Au (ECE)
                         Prof. Pheng-Ann Heng (Comp. Sci. & Engg., CUHK)


**** ALL are Welcome ****