Presented by Roland Göcke at Graphics & Web'96, Dresden, Germany, 6-8 December 1996
In medicine patients are often imaged with more than one tomographic radiological imaging modality (e.g. CT, MR) for the purpose of improved diagnosis or treatment planning. Medical diagnosis benefits from the complementarity of the information in images of different modalities. It can be difficult for a clinician to mentally combine all the image information accurately because of variations in patients orientation or differences in resolution or contrast of the images. Therefore, an image registration technique is sought that transfers all the image information into a common coordinate system. The aim is to present the images in a way that makes it easier for the clinician to fuse the image information to find similarities and differences.
Rigid-body registration of 3D medical images was used to align two 3D scans of a patient that were taken at different times with the same modality (single-modality) or generated on different medical imgaing modalities (multi-modality). Voxel-similarity-based matching criteria are measures of misregistration that are functions of the attributes (e.g. grey-values, gradients, textures) of all pairs of common voxels from the images to be registered at a given position of misregistration. Such approaches purely based on voxel similarity can be fully automated.
Two voxel-similarity-based measures, proposed by A.Collignon et al, were investigated as matching criteria, the entropy and the relative entropy of the joint probability distribution of the grey-values of all common voxel pairs in two images. The developed algorithm recursively evaluates the similarity measure at the current position of misregistration, tries to optimize the measure and calculates the rigid-body transformation parameters until the images are registered. Several optimization techniques were investigated for their usability in the optimization process.
The algorithm is completely automatic, and does not need any user-interactivity or pre- segmentation. It works for both complete and truncated images. The algorithm is based on the original grey-values only and is therefore more robust than the surface-based registration algorithms.
Download (without figure) (199kB, PDF)
[Homepage] [Research] [Publications]