Title: Target Selection Method in Atlas Construction Based on Multidimensional Scaling

Probabilistic atlases provide pivotal information for medical image segmentation and registration. Typically an atlas has been built on a common target image space which other training images are mapped onto. This introduces bias towards to the chosen target. Here we present a method to choose a target image which has the least bias considering all training images. Our method chooses a target image that is the closest to the mean geometry of the training images determined by bending energy. Our approach is based on forming a distance matrix based on bending energies of all pair-wise registrations and performing multidimensional scaling (MDS) on the distance matrix.