\chapter*{Preface}

This mini-book attempts to provide a general introduction to the
statistics side of structural brain imaging, with a heavy emphasis on
practical worked examples. It also introduces a particular toolkit,
RMINC, designed to make running these types of statistical analyses
easier. It is targeted at the general user, who may or may not have
some statistical background, but does have some data they want
analysed in a straightforward way. It is not meant to be a complete
handbook on statistics, but hopefully will provide enough of a primer
to get by, at least for a little while. 

This mini-book exists for a number of reasons. I have over the years
been asked multiple questions relating to structural brain imaging and
statistics, and have had the chance to learn answers to those
questions from countless people. This book thus exists as an attempt
to put some of those answers down on paper. Secondly, writing this
book is part and parcel of the development of RMINC; it is easier to
write code useable by others if one documents it first, and then
writes code to fit the documentation.

The book will likely be an incomplete work in progress for a long
while yet. The \LaTeX source for this book are packaged along with
RMINC itself, and contributors are most welcome!


\tableofcontents
%\listoffigures
%\listoftables


\chapter{Introduction}

The process of analysing brain imaging data is typically comprised of
a series of stages. The study is designed with various choices made
about the biology question that is to be addressed and the data
necessary to answer the questions thus posed. Then the data is then
acquired, and once that is completed, the images are processed in
various automatic, semi-automatic, or manual ways and then analysed. 

This book deals mainly with the final part, the data analysis, though
there will be several side-tracks into the other topics. A single
example will be used throughout: a mouse brain imaging study comparing
male to female mouse brains. The methods described herein should be
easily transferable to any other structural imaging study which looks
at brain shape, tissue classification, or signal intensities.

\section{Installing the tools}

All the analyses will be performed using RMINC, which is a library
designed to handle MINC\index{MINC} volumes inside the R\index{R}
statistical environment. All the tools needed are freely available,
and should run on just about any computer/operating
system. Installation and setup is described in some more detail below.

\subsection{R}

Quick background about R.

Installing R.

Where to find further reading.

\subsection{MINC}

Quick background about MINC.

In order for RMINC to work, the MINC libraries have to be compiled as
shared libraries. When configuring, make sure the following flag is added on:

\begin{verbatim}
./configure --enable-shared=yes
\end{verbatim}

Where to find more information.

\subsection{RMINC}

Quick background about RMINC.

Installing RMINC.

\section{Overview of the analysis process}


The data analysis process usually proceeds in the following way. First
the input images are assessed for correctness; any obvious processing
errors are removed from any subsequent analyses. The question of what
constitutes an outlier is often a tricky one. In order to avoid the
temptation to manipulate the data in a biased way it is best if the
person who reviews the input data is blind about the categorization of
each particular dataset.

Once all the acceptable datasets are in place a series of descriptive
statistics can be generated, usually consisting of means and standard
deviations of all images in the study as well as of all the
subgroupings. This is followed by generating statistical maps of the
main variables of interest. These are then thresholded for
significance while taking multiple comparisons into account. There is
then often a series of steps in which new statistical models are
analyzed and thresholded until the results become more
understandable. This usually involves lots of plotting of individual
datapoints.

% This book will cover those topics in turn in the following
% chapters. Chapter \ref{chp:input} describes the prepartion of the data
% for analysis, chapter \ref{chp:descriptive} shows how to run
% descriptive statistics, chapter \ref{chp:group} talks about performing
% group comparisons and ANOVAs, chapter \ref{chp:ancova} gets into the topic of
% correlations, regressions, and ANCOVAs, chapter \ref{chp:multiple}
% addresses the issue of multiple comparisons, and chapter
% \ref{chp:plotting} describes how to plot individual datapoints.

\section{Data used throughout this book}

This book will consistently work with one dataset consisting of 5 male
and 5 female C57Bl/6 mice, taken from a larger dataset published in a
2007 NeuroImage paper by Spring et al. The mice, all 12 weeks old,
were scanned using an overnight T2-weighted FSE sequence, then all
aligned into a common space using an automated image registration
algorithm (i.e. deformation based morphometry). The final metric of
interest was then the Jacobian determinant of the deformations needed
to align each mouse to the final common atlas. These details are
relatively unimportant for this book - the input might as well be
voxel density maps from VBM - but at least it gives some background
for those who care. If you want to follow along with the examples used
in this book you can download the data at
\url{http://launchpad.net/rminc}. Note that the data has been
downsampled to 120 micron voxels (from the original 32 micron voxels)
to keep the download within reasonable limits.