Non rigid registration in neuroimaging: from a retrospective evaluation of global approaches to new advances in local and hybrid methods

摘要

Within the scope of three-dimensional brain imaging and brain functional mapping, we present a retrospective evaluation framework of inter-individual fusion scheme to register brain cortical anatomy and functional activations. The evaluation methodology relies on global and local criteria such as gray/white matter segmentation map overlap, correlation ratio of the mean Lvv intensities (i.e. 3D measure of the curvature of the cortex from the MRI volumes), distance and shape variation of cortical sulcal landmarks (segmented using the "active ribbon" approach). Visual and global measures seem to promote the non-linear global registration methods while local measures, based on sulci, did not show any significant differences between all global methods. As a perspective, new approaches are introduced to account for these local cortical deformations by means of either introducing sparse constraints in global registration methods or by introducing a local non-linear registration framework based on active shape models of local cortical features.