Biomedical Engineering Reference
In-Depth Information
8.4
Conclusion
This chapter has presented an overview of the classification of non-rigid reg-
istration methods with particular focus on non-rigid registration of brains of
different subjects. Methods can be broadly classified into two groups: geomet-
ric methods that rely on the extraction and matching of sparse features (points,
curves, surfaces), and photometric (or intensity-based) methods that rely on
image luminance directly.
Geometric methods reduce the dimensionality of the problem and are con-
sistent in the vicinity of features used for registration. However, the registration
might be incorrect far from used features. Photometric methods use all the avail-
able information present in the volume but lead to a complex problem involving
a very large number of variables.
We have presented here the Romeo algorithm (Robust Multigrid Elastic reg-
istration based on optical flow) that refers to photometric methods. Romeo
uses the optical flow as a similarity measure and relies on an efficient multires-
olution and multigrid optimization scheme. Robust estimators are introduced
to limit the effect of erroneous data and to preserve discontinuities of the de-
formation field when needed. Prior to the non-rigid registration step, two pre-
processing steps are performed: a rigid registration by maximization of mutual
information and an intensity correction so that the luminance of the volumes
to be registered are comparable. An extension to multimodal data has been
presented. The multiresolution and multigrid framework is flexible enough to
be adapted to multimodal similarity measures such as mutual information for
instance.
It has been shown that photometric methods fail in matching cortical struc-
tures such as cortical sulci, for instance [74]. This can be explained by the high
variability of cortical structures among subjects. Anatomists have pointed out
[103] that cortical sulci of different subjects are very different in shapes. This
has motivated mixed approaches where a photometric registration method in-
corporates sparse anatomical structures so as to drive the registration process
[22, 29, 36, 68, 73, 79, 141].
In this context, it must be noted that validation is difficult and should be
investigated further. Validation of non-rigid registration methods on anatomi-
cal structures have been conducted [74, 119]. However, the impact of non-rigid
