Biomedical Engineering Reference
In-Depth Information
in a cost-effective manner. This is the topic of health technology assessment.
It is touched on in application chapters but is largely beyond the scope of
this topic. It is clear that image registration is invaluable in neurosciences
research and the clinical application of image-guided surgery. Many other
applications will undoubtedly be accepted as the technology matures.
2.9
Summary and Conclusion
This chapter has introduced the basic ideas that underlie image registration.
The concept of correspondence was discussed in some detail. A clear defini-
tion of correspondence is required in any new application (and many new
applications are being suggested all the time). This is particularly important
as the development of applications in nonrigid and intersubject registration
gathers pace. The chapter proceeds with a discussion of the dimensionality
of the data to be registered and the number of degrees of freedom of the trans-
formation. This topic is primarily concerned with registering 3D images
assuming that the part of the body imaged can be treated as if it were a rigid
body. Image registration has also been successfully applied to 2D images,
between 2D and 3D images, and to time series. We are also beginning to see
progress in devising useful and practical algorithms that allow images to be
aligned in cases where the structure that is imaged deforms, or when aligning
images from different individuals.
A descriptive account of the more successful image registration algorithms
is provided. The intention in this chapter is to provide insight into the con-
cepts behind the algorithms, not to provide mathematical, algorithmic, or
implementation details. Our focus is on an understanding of how these algo-
rithms work that the nontechnical individual can understand. The descrip-
tion of the algorithms for the more mathematically inclined is left to the next
chapter. We describe the well known point-based and surface-based algo-
rithms, and, in more detail, the highly successful voxel intensity or voxel sim-
ilarity approaches. We touch on recent work on nonrigid registration
algorithms, which are dealt with in more detail in Chapter 13.
Very few registration algorithms provide a direct calculation of the trans-
formation. The computer has to search iteratively for the best solution. This
is the process of optimization described conceptually in Section 2.6. Finally,
we usually need to transform one of the images into the coordinate system or
“space” of the other. Some of the issues and pitfalls in transformation are dis-
cussed in Section 2.7.
Finally, all complex computations must be validated. This is particularly
important in medical applications. This topic is introduced in Section 2.8 and
expanded throughout this topic.
