Elastic registration in 3D volume data

The automatical comparison of images of mouse embryos is an important step for fusion of image contents, analysis and transfer of segmentation. Given the images of two different embryos from the same developmental stage, one could define them to be equal up to an elastic mathematical transformation. Elastic registration is then the estimation of this transformation. In this diploma thesis we have evaluated and optimized leading techniques of the field of elastic registration to perform inter-subject registration of embryonal mice. Motivated by our findings, we developed a new algorithm to extract a topological shape descriptor the medial axis. The medial axis provides a common data-driven coordinate frame for elastic registration. We show the feasibility of extracting this feature from our data and illustrate it’s potential particularly for registration, but also for segmentation. Finally we validate two successful results, that show the feasibility of elastic registration on this complex data. We conclude with a discussion and many ideas for further research.