Biomedical Engineering Reference
In-Depth Information
Electronic atlases have two main purposes:
Automatic segmentation of a given subject by matching the segmentation
and labeling of anatomical structures of a template. Labels of the template
can be deformed into another subject, under the assumption that there is a
total relation between the points of the atlas (source) and the points of the
studied subject (target). This objective has been pursued for a long time
in medicine and was traditionally treated by paper atlases with generally
rather simple transformations. The most known example is the atlas of
Talairach with its famous AC-PC referential and its related proportional
squaring [131].
Understanding of brain functions. Many techniques have been developed
to record brain activity (SPECT, PET, MEG/EEG, fMRI). However, the links
between anatomy and functional organization are often not well known:
the superimposition of multiindividual neurofunctional recordings on the
same anatomy is useful to better understand the human brain functional or-
ganization. In this case, inherent anatomical variability between individu-
als may disturb this interpretation. Therefore, spatial normalization, which
is the goal of non-rigid registration methods, makes it possible to study the
functional variability. A better knowledge of this anatomy-function rela-
tionship is of great interest for the researcher in cognitive neuroscience,
as well as for the surgeon and the neurologist who intend to delineate
relevant functional areas before surgery.
This chapter is divided into two sections: an overview with classification
of non-rigid registration techniques will be presented first. The Romeo algo-
rithm (robust multigrid elastic registration based on optical flow) will then be
described.
8.2
Overview of Non-Rigid Registration
Methods
8.2.1
Introduction
Non-rigid registration is a very active field of research and numerous methods
have been proposed. This section does not intend to propose an exhaustive
