Diffusion Tensor Imaging

$\begin{figure}\centering\epsfig{file=images/title.ps,width=12cm}\end{figure}$

Nonrigid registration of diffusion tensor images

Raimundo Sierra
Electrical Engineering, Master Thesis
rsierra@osola.com

Advisor at ETH Zürich: Prof. Dr. Gabor Székely
Advisor at Harvard: Simon K. Warfield, Ph.D.

March 2001

$\includegraphics[width = 0.2\textwidth]{images/biwi.ps}$

Swiss Federal Institute of Technology (ETHZ)
Computer Vision Laboratory
Medical Image Analysis and Visualization Group
Gloriastrasse 35
ETH-Zentrum
CH - 8092 Zürich, Switzerland
http://www.vision.ee.ethz.ch/

$\includegraphics[width = 0.15\textwidth]{images/spl.ps}$

This thesis discusses diffusion tensor imaging as a Magnetic Resonance Imaging modality. Diffusion tensor imaging allows the observation of molecular diffusion in tissues in vivo and therefore the molecular organization in tissues. The main interest in this case is the observation of myelinated fibertracts in the brain of premature born babies. Myelination of fibers in the white matter of the brain is a fast process in the last few weeks preterm and the observation of this process gives an insight in the development of the human brain and allows a better and earlier detection of small injuries or abnormalities.

The goal is to match diffusion tensor images of neonates, and to build an enabling technology to ultimately generate a statistical atlas of the development of the brain in babies between 28 and 40 weeks postconceptional age. While it was not possible to build the statistical atlas in the time given, the complete process from preprocessing of the data to nonrigid alignment of diffusion tensor images has been implemented and successfully applied on some exemplary cases.

To better understand the characteristics of diffusion tensors and to be able to prove the correctness of the algorithms, a new way of displaying diffusion tensors was implemented. This method visualizes the diffusion tensor as ellipsoids in a voxel raster.

The following report outlines the medical background, the imaging acquisition process and the data processing path. The reader should be able to understand diffusion tensor imaging and the matching principles used and expand the provided software to fit specific and further needs.

To successfully build a meaningful atlas of the development of the brain in neonates, a number of three-dimensional cases needs to be processed and a statistical analysis of the results has to be performed. Therefore a correct incorporation of the different voxel dimensions has to be implemented. As the data quality of the diffusion tensor images in baby scans is very low, the incorporation and combination with other scanning modalities should be considered.

Acknowledgments

First, I would like to thank Prof. Gabor Székely at the Swiss Federal Institute of Technology in Zürich, Switzerland, for his support and motivation for a diploma thesis in the USA, at the Surgical Planning Laboratory of Brigham and Women's Hospital, Boston, Massachusetts. A very special thanks goes to my supervisor at the Surgical Planning Laboratory, Dr. Simon Warfield, who always had time to answer any questions and discuss all the problems that arose. I also want to thank Ron Kikinis, head of the Surgical Planning Laboratory, who made this stay possible.

Furthermore, I want to say thanks to everybody at the Surgical Planning Laboratory who helped me get started with the different subjects I had to learn about, especially Carl-Fredrik Westin, Steven Haker, Hatsohu Mamata, Gary Zientara and Stephan Maier.

Of course, there are lots of other people at the SPL who made this work a great experience.