Fluid registration is widely used in medical imaging to track anatomical changes, to correct image distortions,
and to integrate multi-modality data. Fluid mappings guarantee that the template image deforms smoothly into
the target, without tearing or folding, even when large deformations are required for accurate matching.
Here we implemented an intensity-based fluid registration algorithm, accelerated by using a filter designed
by Bro-Nielsen and Gramkow. We validated the algorithm on 2D and 3D geometric phantoms using the mean
square difference between the final registered image and target as a measure of the accuracy of the registration.
In tests on phantom images with different levels of overlap, varying amounts of Gaussian noise, and different
intensity gradients, the fluid method outperformed a more commonly used elastic registration method, both in
terms of accuracy and in avoiding topological errors during deformation. We also studied the effect of varying
the viscosity coefficients in the viscous fluid equation, to optimize registration accuracy. Finally, we applied the
fluid registration algorithm to a dataset of 2D binary corpus callosum images and 3D volumetric brain MRIs
from 14 healthy individuals to assess its accuracy and robustness.