This work presents a detailed investigation of two
multiple-templates based fusion approaches for automated segmentation of
structures in the brain MR images: (i) fusion based on direct pairwise
registrations between each template and the target image, and (ii) fusion based
on an intermediate groupwise template, requiring only a single onsite
registration. The key finding from these evaluations is that, if computational time for automated segmentations is a
major concern, then groupwise-template based registration followed by fusion is
an optimal choice; if time is not a major constraint, then multiple pairwise
registrations followed by fusion provides more accurate segmentations.
