An overview of classifier combination methods for atlas-based segmentation
Yolanda H. Noorda · Utrecht University Repository (Utrecht University) · 2010
In this paper, an overview will be given of several classifier combination methods, which can be used for atlas-based segmentations. Atlas-based segmentation using multiple atlases needs some kind of label fusion to combine the labels for each voxel yielded by the atlases. This label fusion process falls within a general classifier combination problem. Classifier combination methods based on rules derived from Bayesian probability theory as well as label fusion algorithms can be used to combine the segmentations. Methods of validating segmentations will be presented. Finally, results in practice will be discussed. It can be shown that certain combination methods are more beneficial in particular circumstances than others.