OPTIMISATION STRATEGIES FOR ULTRASOUND VOLUME REGISTRATION
Umer Zeeshan Ijaz, Richard W. Prager, Andrew H. Gee and Graham M. Treece
August 2009
This paper considers registration of 3D ultrasound volumes. One way to acquire 3D data is to use a mechanically-swept 3D probe. However, the usefulness of these probes is restricted by their limited field of view. While this problem can be overcome by attaching a position sensor to the probe, an external position sensor can be an inconvenience in a clinical setting and does not align the volumes correctly when there is tissue displacement or deformation. The objective of this paper is to replace the 6 degree of freedom (DOF) sensor with a combination of 3 DOF image registration and an integrated intertial sensor for measuring orientation. We examine a range of optimisation algorithms and similarity measures for registration and compare them in \emph{in vitro} and \emph{in vivo} experiments. We register based on multiple reslice images rather than a whole voxel array. In this paper, we use a large number of reslices for improved reliability at the expense of computational speed. We have found that the Levenberg-Marquardt method is very fast but is not guaranteed to give the correct solution all the time. We conclude that normalised mutual information used in the Nelder-Mead simplex algorithm is potentially suitable for the registration task with an average execution time of around five minutes, in the majority of cases, with two restarts in a C++ implementation on a 3.0 GHz Intel Core 2 Duo CPU machine.
If you have difficulty viewing files that end '.gz',
which are gzip compressed, then you may be able to find
tools to uncompress them at the gzip
web site.
If you have difficulty viewing files that are in PostScript, (ending
'.ps' or '.ps.gz'), then you may be able to
find tools to view them at
the gsview
web site.
We have attempted to provide automatically generated PDF copies of documents for which only PostScript versions have previously been available. These are clearly marked in the database - due to the nature of the automatic conversion process, they are likely to be badly aliased when viewed at default resolution on screen by acroread.