Atlas-Based Anatomic Labeling in Neurodegenerative Disease via Structure-Driven Atlas Warping

A new method is presented for the automated anatomic labeling and comparative morphometric analysis of brain magnetic resonance imaging, warping a prelabeled atlas into congruence with the subject anatomy. The strategy emphasizes anatomically meaningful atlas deformations in the presence of strong degeneration and substantial morphologic differences, for example, cases with high levels of atrophy. The atlas deformation is not driven by image intensity similarities but by continuous anatomic correspondence maps, derived from individual presegmented brain structures. Automatically generated correspondence maps provide large sets of fiducials, driving a warp with many thousand degrees of freedom. Validation included a scan-rescan study and anatomically relevant self-validation in multiple sclerosis patients with substantial cortical and subcortical degeneration. The mean coefficient of variation in the scan-rescan study was 1.4%, with no significant difference in precision between normal and neurodegenerative anatomies. The self-validation demonstrated good structural overlap, with substantial improvement over established methods, such as Talairach spatial normalization.