Diffeomorphic registration of images with variable contrast enhancement
Journal of Biomedical Imaging - Special issue on modern mathematics in biomedical imaging
| Hi-index | 0.00 |
We present an exploration of the interplay between the extent of organ motion and contrast-enhancement on image registration success. A model of a DCE MRI of the liver incorporating an isotropic elastic non-rigid deformation is used to simulate both breathing and breath-hold data, a volume-preserving modification for tumour regions is included. Contrast enhancement is simulated by applying a pharmacokinetic model. For each simulated dataset, a direct fluid registration of each image to the first in the dataset is compared to a contrast-enhancement guided method known as Progressive Principal Component Registration (PPCR). Analysis of the correction to the deformation fields, tumour volume change and dispersion of joint image histograms are used to show the importance of motion type on PPCR success and of enhancement level on direct fluid registration success. For breathing motion, PPCR registers groups of images to separate locations, but maintains enhancing tumour volume. This is not the case for direct registration with volume changes of up to 7%. For inconsistent breath-hold depth, PPCR out-performs direct registration, particularly for large enhancement levels. Analysis of the joint histograms suggests that the generation of target images using PPCR reduces dispersion due to contrast enhancement. Since this distinction is not made using direct registration, it is unable to register images when large enhancement intensity changes are present.