Investigating anisotropic diffusion for the registration of abdominal MR images

Sliding boundaries are a challenge when registering abdominal images, especially for MRI where the boundaries are not clearly visible. Various automatic methods have been proposed, but evaluation has been insufficient. In this study we assess the performance of several anisotropic diffusion based methods for registering abdominal MR images undergoing large respiratory motion. Evaluation was based on manually selected landmarks inside and outside of the liver. Symmetric iLogDemons reduced the mean error from 12.1 mm to 3.8 mm. Anisotropic diffusion, based on the velocity field and the modified curvature diffusion equation, reduced this further to 3.4 mm.