A binary entropy measure to assess nonrigid registration algorithms

Abstract

Assessment of normal and abnormal anatomical variability requires a coordinate system enabling inter-subject comparison.W e present a binary minimum entropy criterion to assess affine and nonrigid transformations bringing a group of subject scans into alignment.Th is measure is a data-driven measure allowing the identification of an intrinsic coordinate system of a particular group of subjects. We assessed two statistical atlases derived from magnetic resonance imaging of newborn infants with gestational age ranging from 24 to 40 weeks.O ver this age range major structural changes occur in the human brain and existing atlases are inadequate to capture the resulting anatomical variability.Th e binary entropy measure we propose allows an objective choice between competing registration algorithms to be made.