Biomedical Engineering Reference
In-Depth Information
especially important for cases of brain swelling, where the deformed brain would
otherwise overlap the skull and skin in the craniotomy area.
4 Discussions and Conclusions
In this paper, we present an algorithm for transforming preoperative images
using the deformation field obtained from biomechanical models. We outline the
complexity of warping brain images using the deformation field predicted by a
biomechanical model. Compared with the transformations normally used in image-
based registration methods, this algorithm is more computationally expensive,
although it needs to be applied only once. The first two steps of the algorithm can
be performed preoperatively, reducing the intraoperative computation time to
half. Even so, the current implementation is not fast enough for intra-operative
image updates, especially for large preoperative images.
Another strategy for reducing the computation time is to update only the part of
the image important for the surgeon, such as the area containing the craniotomy, the
tumor and the nearby tissue. In fact, this is the area where most of the deformation
takes place. As the transformation time increases linearly with the number of voxels
handled, partial updating can significantly reduce the computation time.
The proposed algorithm is very well suited for parallel implementation. Every
voxel in the transformed image can be processed independently, using the same
algorithm, making it suitable for parallel computing using a SIMD (Single Instruc-
tion Multiple Data) or SIMT (Single Instruction Multiple Threads) architecture.
Therefore, we consider using a graphics processing unit (GPU), which has a SIMT
architecture, to perform image warping within the time constrains of surgery. The
main challenge in the implementation will be the creation of an efficient parallel
algorithm for 3D scattered data interpolation using a general tetrahedral mesh (with
non-convex boundary and distorted tetrahedral elements). Such algorithm does not
require a 3D Delauney tessellation to be constructed, simplifying the implementa-
tion on parallel hardware.
Acknowledgments The financial support of the Australian Research Council (Grant No.
DP1092893) and NHMRC (Grant No. 1006031) is gratefully acknowledged.
References
1. Nakaji, P., Speltzer, R.F.: The marriage of technique, technology, and judgement. Innov. Surg.
Approach 51, 177-185 (2004)
2. Warfield, S.K., Talos, F., Tei, A., Bharatha, A., Nabavi, A., Ferrant, M., Black, P.M., Jolesz, F.
A., Kikinis, R.: Real-time registration of volumetric brain MRI by biomechanical simulation of
deformation during image guided surgery. Comput. Vis. Sci. 5, 3-11 (2002)
