Parallelization of Coherent Point Drift for patient registration
Norwegian University of Science and Technology, Faculty of Natural Sciences, Department of Physics
Norwegian University of Science and Technology, 2019
@mastersthesis{knutsen2019parallelization,
title={Parallelization of Coherent Point Drift for patient registration},
author={Knutsen, Gjert Magne Kahrs},
year={2019},
school={NTNU}
}
Point set registration is a central part in any application where the correspondence between two data point sets is of interest, for instance patient data from medical examinations. There exists numerous different algorithms that aim at solving the registration problem, and one of which is the Coherent Point Drift (CPD) algorithm. In this thesis a fast implementation of this algorithm is developed with parallelization on both CPU and GPU, significantly reducing the time required to perform the registrations. The developed algorithm, called FAST CPD (FCPD), includes rigid, affine and elastic registration and the code will be freely available as open-source. The rigid FCPD runtimes allow for real-time registration with the CPD algorithm, and the elastic registration runtimes make it possible to register relatively large point sets within a few seconds. The elastic FCPD is applied to data used to correct for the brain-shift occurring during brain surgeries.
August 5, 2019 by hgpu
