Distributed-Memory Large Deformation Diffeomorphic 3D Image Registration
SessionInverse Problems and Quantum Circuits
Session ChairGeorge Biros
Event Type
Paper
Algorithms
Applications
Effective Application of HPC
Introductory
TimeThursday, November 17th1:30pm - 2pm
Location355-BC
DescriptionWe present a parallel distributed-memory algorithm for large deformation diffeomorphic registration of volumetric images that produce large isochoric deformations. Finding the optimal deformation mapping requires the solution of a highly nonlinear problem that involves pseudodifferential operators, biharmonic operators, and pure advection operators. We use a preconditioned, inexact, Gauss-Newton-Krylov solver. Our algorithm integrates several components: a spectral discretization in space, a semi-Lagrangian formulation in time, analytic adjoints, different regularization functionals including volume-preserving ones, a spectral preconditioner, a highly optimized distributed FFT, and a cubic interpolation scheme for the semi-Lagrangian time-stepping. We demonstrate the scalability of our algorithm on images with resolution of up to 1024^3 on Maverick and Stampede systems. The challenging problem in the medical field is to solve the registration problem for moderate size of 256^3. We are able to compute registration for the size of 256^3 in less than five seconds on 64 x86 nodes of Maverick.
