A study of distortion in fMRI using elastic matching method

• Main Author: 楊惠婷
• Other Authors: 陳志宏
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/68936656275337471286

碩士 === 國立臺灣大學 === 電機工程學研究所 === 89 === The fMRI (functional magnetic resonance image) is a powerful approach to studying functional activities in the brain. This method enables us to observe how functional stimulants change the blood flow in the brain, and finds the neuronal activated areas. In the fMRI experiments, we usually use the EPI (echo-planar imaging) pulse sequence. The advantage is that we can scan the whole brain in a short time. However, this approach suffers serious geometric distortions due to the inhomogeneous magnetic fields, which makes it difficult to determine the correct anatomic location for each observed functional activity. The objective of this thesis is to reduce the geometric distortion, which results from inhomogeneous magnetic fields. The present methods of correcting EPI distortion are shim, field map and image post-process. The first two methods cannot completely remove the geometric distortion. As a result, we use elastic matching methods to correct the distortion. In 1989, Bookstein proposed the TPS method to overcome the problem of image transformation. His theory enables us to match EPI image and anatomy by TPS method. Because the TPS method is limited by the transformation model and the contour which we use, we propose a new approach, which is called Spatially-Variant Model-based Spline (SVMS) method, to solve the problem. Besides Thin-plate Spline (TPS), this method adds three more energy; one is contour energy, the other is control point displacement energy, and another is continuity energy. By minimizing the energy function, we can get the way of image transformation. In this process, SVMS provides a family of deformable models. Through these models, we adopt three steps to do image elastic matching, including global match, spatial variance model, and contour local match. In addition, we use gradient vector flow (GVF) field and bending energy to decide optimal deformable model. Comparing with the TPS method, this method provides us a family of deformable models to choose. As a consequence, the result of SVMS is more precise than that of TPS.