Simultaneous Alignment and Central Tendency Estimation for Brain Atlas Construction

Two key questions for investigators seeking to compare images across subjects are how to define a common coordinate system and how best to relate each subject to that coordinate system. We have developed a generalized expectation-maximization (EM) algorithm which iterates between estimating the most typical anatomy, assessing the typicality of each individual subject’s anatomy, and registering each individual subject to the typical anatomy estimate. Unlike methods which seek a coordinate system unbiased from the average, we weight the contribution of each tissue class from each subject based on the estimated typicality of that subject’s features. We demonstrate our algorithm using segmented brain images from a group of 14 adults and 14 preterm infants.