Intra-subject MRI image registration

Abstract

Brain MRI imaging technology has been widely used in modern clinical field. As the key of this technology, MRI image registration method plays a critical role and dominants the success of this advanced technology.

This project was designed to improve registration accuracy and reliability by optimizing some of the input variables of registration method. The function of registration method is to seek the best transformation between two images, which makes the transformed image (one of the two images after transformation) have the best alignment and the highest similarity with the other image. Instead of viewing registration result by human eyes, cost function provides a more straightforward way to evaluate the performance. Generally it is used to measure the dissimilarities between two scans [3], and there had lots of attentions been focused on it, also a lot of researches had been performed on optimization method and interpolation method in order to further improve the accuracy and reliability of registration method. They did get some success in terms of registration accuracy; however the conclusions obtained from those researches are not always suitable for every other specific data.

In this project, with the assumption that all variables are independent to each other, different cost functions, maximum search angle, interpolation methods and angle representation schemes were re-tested on the database which contains 10 subjects of raw MRI brain image from local area. Based on the analysis of testing results, the best set of variables was chosen to be used for future registration work.