Multi-classifier framework for atlas-based image segmentation

Three different systematic approaches to generate multiple classifiers in atlas-based biomedical image segmentation are compared. Different atlases, as well as different parametrization of the registration algorithm, lead to different atlas-based classifiers. The classifier outputs are combined and compared to a manual ground truth segmentation. Classifier combination consistently improved classification accuracy with the biggest improvement from multiple atlases. We conclude that multi-classifier techniques have a natural application to atlas-based segmentation and increase classification accuracy in real-world segmentation problems.