A data-driven approach to discovering common brain anatomy

An atlas defines a common coordinate system to enable the comparison of data from different subjects. Key in the development of a brain atlas are the identification of a common coordinate system and the definition of a procedure for aligning an individual brain to the common coordinate system. The algorithms used for atlas construction to date have not sought to characterize residual anatomical variability after registration, and have sought to assign equal weight to each subject, rather than assessing their degree of commonality. Our new algorithm defines a common coordinate system by estimating the typical anatomical distribution of brain structures and characterizes the quality of alignment of each subject within the common coordinate system. Residual anatomical variability is quantitatively described by the extent to which the brain structures of an individual fail to match the typical anatomy. We have applied this to cohorts of 14 adult and 11 newborn brain segmentations and demonstrated the ability of the algorithm to distinguish groups of subjects in a clinically relevant application.