Title: Vascular Reconstruction for Minimally-Invasive Computer-Assisted Intervention Planning

Date: Monday, 19 March 2001

Time: 4:00pm - 5:00pm

Venue: Lecture Theater F (Leung Yat Sing Lecture Theater), Academic Concourse (near lift nos. 25/26), HKUST

Abstract:
This seminar aims at showing how we exploited computer vision techniques for three-dimensional (3D) reconstruction of vessels and aneurysms in the brain. The main purpose of this research was to improve the current endovascular coil embolisation procedures. With an accurate 3D vascular model, (a) the size and shape of the aneurysms can be measured before the treatments, and (b) the vessels in the vicinity of aneurysms can be clearly seen. The knowledge of these two factors is extremely important in the success of the treatments and in reducing the number of invasive, X-ray based digital subtraction angiography (DSA) acquisitions during the treatments.

Magnetic resonance angiography (MRA) is commonly used for the diagnosis of arterial diseases and 3D reconstruction of the brain vasculature because most of the vasculature can be clearly seen in the images and (unlike DSA) it is non-invasive. However, the low or complex flow inside the aneurysms makes their visualisation and reconstruction difficult. There is a need to develop a new reconstruction method suitable for both normal vessels and aneurysms.

As such, we created a new statistical reconstruction method making use of the knowledge of the physics of MRA image formation and other available information (speed and velocity field) in the MRA images. This is a new way of segmenting MRA data. Firstly, on the basis of the knowledge of the physics of MRA image formation, we investigated a tailored statistical description of the signals in MRA data, namely MGU statistical model. It is shown that the Maxwell-Gaussian mixture distribution (a) models the background signal more accurately than the conventional Maxwell distribution, (b) exhibits a better fit to clinical data and (c) gives fewer false positive voxels (misclassified vessel voxels) in segmentation.

Secondly, additional velocity field information was exploited to improve the segmentation quality of brain vessels and aneurysms. We have demonstrated that, rather than relying on speed information alone, as done by others, including velocity field information as a priori knowledge in a Markov random field (MRF) model can improve the quality of segmentation.

The new segmentation algorithm was tested on the phantom and clinical datasets. The experimental results show that the proposed method can provide a better quality of segmentation.

Biography:
(Albert) Chi-Shing Chung received his B.Engg. degree (First-Class Honours) in Computer Engineering from the University of Hong Kong (HKU) in 1995. He was an exchange student at the University of British Columbia (UBC), Canada from 1992-93 with the Asian Cathay Pacific Award. He joined Hong Kong University of Science and Technology (HKUST) in 1996, and received his M. Phil. degree in Computer Science in 1998. Mr. Chung is a Croucher scholar. He joined the Medical Vision Laboratory at the University of Oxford as a doctoral research student in 1998. His research interests include medical image analysis (vascular segmentation and registration), statistical analysis of MR signals, Markov Random Fields and Level Set Methods.