Introduction - Medical Image Registration - page 6

Between then and late 2000, desktop workstations have increased in speed by
nearly two orders of magnitude while their cost has been reduced by an order
of magnitude, so the same calculations can now be achieved in a few minutes
on desktop PCs. The first algorithms for nonrigid registration required large
amounts of interaction or were prohibitively slow, but more recent work has
resulted in highly automated algorithms that can run on standard hardware
in minutes rather than days.

1.3 Overview of the Topic

This topic is divided into three sections. Section I, Methodology, introduces
the wide variety of techniques used for medical image registration. The con-
cepts behind registration techniques are introduced for a general audience in
Chapter 2, and, for those who wish to understand the underlying algorithms
or implement registration methods themselves. Behind the techniques are
described in more detail in Chapter 3. The necessary additional considerations
behind acquiring and preparing data for image registration are discussed in
Chapter 4, and correcting errors in the scanners is addressed in Chapter 5. In
the final chapter in this section, Chapter 6, the essential problem of detecting
when the algorithms have failed is discussed, and how accurately the algo-
rithms have aligned the images is assessed.
Section II describes the relatively mature applications of image registration
in which the images can be aligned by global translation and rotation
alone—so-called rigid-body registration. The chapters in this section are
written by researchers with many years of experience in the applications of
serial MR registration (Chapter 7), functional MRI (Chapter 8), registration of
PET and MRI (Chapter 9), registration of MRI and CT (Chapter 10), registra-
tion in nuclear medicine (Chapter 11), and the use of registration in guided
therapeutic procedures (Chapter 12).
Section III focuses on the less mature but rapidly developing field of nonrigid
image registration. The topic is introduced in Chapter 13, and alternative
approaches are reviewed and examples given of one approach for a variety
of applications. The problem of combining images from multiple subjects in
cohort studies is examined in detail in Chapter 14, and the rather different
approach of using biomechanical models to achieve registration is considered
in Chapter 15. Section III tells less of a finished story than Sections I and II, due
to the rapid evolution of techniques in this area. The goal of this section is
to give insight into some of the applications that drive nonrigid registration
and the different approaches being devised.
An observation as this topic was prepared (late 2000) was that there are
literally hundreds of papers in the literature describing medical image regis-
tration methods and applications, and yet this technology is currently used
very little in clinical practice, with the exception of image-guided surgery.