[Zurück]

S. Zambal:
"3D Active Appearance Models for Segmentation of Cardiac MRI Data";
Betreuer/in(nen): J. Hladuvka, K. Bühler, E. Gröller; Institut für Computergraphik und Algorithmen, 2005.

Segmentation of volumetric medical data is extremely time-consuming if done manually. This is the reason why currently great efforts are being made to develop algorithms for automatic segmentation. Model based techniques represent one very promising approach. A model representing the object of interest is matched with unknown data. During the matching process the model´s shape and additional properties are varied in order to iteratively improve the match. As soon as the model fits sufficiently well to the data, the properties of the model can be mapped to the data and so a segmentation is derived. Recently the segmentation of cardiac magnetic resonance images (MRI) has been of great interest. In this work we outline some of the methods proposed to solve the problem of cardiac segmentation. We review Active Appearance Models (AAMs) which are a special type of deformable models. AAMs rule changes in shape and texture using statistical information obtained from a data base of representative examples. We describe the theory behind AAMs with special focus on 3D AAMs. These are applicable to volumetric medical image data. Our implementation of 3D AAMs is outlined and the results obtained for 3D segmentation of the left cardiac ventricle are presented.