Correspondence evaluation in local shape analysis and structural subdivision

Regional volumetric and local shape analysis has become of increasing interest to the neuroimaging community due to the potential to locate morphological changes. In this paper we compare three common correspondence methods applied to two studies of hippocampal shape in schizophrenia: correspondence via deformable registration, spherical harmonics (SPHARM) and Minimum Description Length (MDL) optimization. These correspondence methods are evaluated in respect to local statistical shape analysis and structural subdivision analysis. Results show a non-negligible influence of the choice of correspondence especially in studies with low numbers of subjects. The differences are especially striking in the structural subdivision analysis and hints at a possible source for the diverging findings in many subdivision studies. Our comparative study is not meant to be exhaustive, but rather raises awareness of the issue and shows that assessing the validity of the correspondence is an important step.