Spatial normalization of brain images and beyond

Abstract

The deformable atlas paradigm has been at the core of computational anatomy during the last two decades. Spatial normalization is the variant endowing the atlas with a coordinate system used for voxel-based aggregation of images across subjects and studies. This framework has largely contributed to the success of brain mapping. Brain spatial normalization, however, is still ill-posed because of the complexity of the human brain architecture and the lack of architectural landmarks in standard morphological MRI. Multi-atlas strategies have been developed during the last decade to overcome some difficulties in the context of segmentation. A new generation of registration algorithms embedding architectural features inferred for instance from diffusion or functional MRI is on the verge to improve the architectural value of spatial normalization. A better understanding of the architectural meaning of the cortical folding pattern will lead to use some sulci as complementary constraints. Improving the architectural compliance of spatial normalization may impose to relax the diffeomorphic constraint usually underlying atlas warping. A two-level strategy could be designed: in each region, a dictionary of templates of incompatible folding patterns would be collected and matched in a way or another using rare architectural information, while individual subjects would be aligned using diffeomorphisms to the closest template. Manifold learning could help to aggregate subjects according to their morphology. Connectivity-based strategies could emerge as an alternative to deformation-based alignment leading to match the connectomes of the subjects rather than images.

Topographic organization of the cerebral cortex and brain cartography

“…In turn, a brain atlas consisting of a multivariate feature vector per location also requires few assumptions other than the possibility of sufficient registration between individuals. We note, however, that the latter assumption may be critical (and may provide an upper bound on resolution), given that any registration needs itself to be driven by some set of recorded features and the choice of those features that drive the image registration may in turn bias the ensuing multivariate description (Robinson et al, 2014; Mangin et al, 2016). Moreover, the proposed concept has, at this stage, no real power for labeling a particular location as a particular cortical area or for providing a biologically meaningful data compression, i.e., the problem of “where are the cortical modules” remains to be addressed as outlined below.…”

A population level atlas ofMus musculuscraniofacial skeleton and automated image‐based shape analysis

“…2015; Mangin et al. ). Although the difference can be attributed to groups leading the research (radiologists or biomedical imaging experts in clinical studies, geneticists or evolutionary biologists in non‐human organisms), there is now ample experience in neuroimaging that can be translated to the model organisms in general, as has been done in the analysis of murine MRI (Johnson et al.…”

Three-Dimensional Probabilistic Maps of Mesial Temporal Lobe Structures in Children and Adolescents’ Brains

“…Conversely, it could rather relate to the remaining variability after normalization due to a poor alignment of sulci since we used a normalization process not optimized for that purpose. Overall, the normalization process remains a key or even limiting step when it comes to group analysis of sulcal morphology and surface location ( Lancaster et al, 2010 ; Lerch et al, 2017 ), and some teams try to avoid it ( Mangin et al, 2016 ). We selected a very classical group normalization process as a first attempt to measure the impact of RS-CS proper morphotype on MLT cortices localization to stay close to the routine functional MRI procedure.…”

“…We selected a very classical group normalization process as a first attempt to measure the impact of RS-CS proper morphotype on MLT cortices localization to stay close to the routine functional MRI procedure. Further work is needed to investigate whether the two sulcal conformations can be distinguished in terms of spatial location or have a significant impact on MLT cortices location by using other normalization procedures meant to preserve sulcal characteristics, such as a DARTEL normalization, or HIP-HOP model-driven harmonic parametrization of the cortical surface: ( Auzias et al, 2013 ; Mangin et al, 2016 ). Machine learning could also be used to investigate whether sulcal conformation can be predicted from extrema location and other morphometric features.…”