Statistically-constrained Robust Diffeomorphic RegistrationMS28

We present a novel framework to construct statistical deformation models for diffeomorphisms, aiming to improve the robustness of deformable registration in the presence of pathologies. We model the high-dimensional velocity field as a collection of local velocity fields. For each local field, we learn a low-dimensional representation using principal component analysis. Dependencies across local transformations are captured through canonical correlation analysis. We showcase the improved robustness of the proposed method using simulated brain lesion images as well as real brain images with pathologies.

This presentation is part of Minisymposium “MS28 - Diffeomorphic Image Registration: Numerics, Applications, and Theory (2 parts)
organized by: Andreas Mang (Department of Mathematics, University of Houston) , George Biros (Institute for Computational Engineering and Sciences, University of Texas at Austin) .

Aristeidis Sotiras (University of Pennsylvania)
Ke Zeng (University of Pennsylvania)
Christos Davatzikos (University of Pennsylvania)
image registration