Wednesday, 1 December 2010, 17 ct. - H1, SFS
Dr. Christoph Brune works in the Workgroup Imaging at the Institute for Computional and Applied Mathematics, University of Münster.
Title: 4D Imaging and Registration using Optimal Transport and Sparsity Concepts
Abstract: This talk contributes to the field of time-dependent image processing and inverse problems. Such problems arise in a wide variety of applications, particularly in biomedical applications, such as dynamic positron emission tomography, life-cell fluorescence microscopy or fast magnetic resonance imaging sequences.
The first goal of this talk is to present models, analysis and algorithms for accurate 3D static reconstruction in biomedical imaging based on total variation and Bregman distances. Standard imaging methods do not incorporate time dependent information or dynamics, e.g. heart beat or breathing in tomography, or cell motion in microscopy. This can lead to insufficient accuracy particularly at object boundaries, e.g. at cardiac walls.
Aiming this deficiency of simultaneous registration is the second goal of this talk. We present models, analysis, and efficient algorithms for 4D reconstructions in biomedical imaging using optimal transport and sparsity concepts. We discuss constraint optimization models combining reconstruction techniques known from inverse problems, spatio-temporal regularization and mass conservation.
The numerical realization is based on multigrid preconditioned Newton-SQP methods with filtered line-search. In the case of nonlinear spatio-temporal regu- larization we present efficient operator splitting techniques with preconditioning facilitating parallelization on GPUs. Dynamic large-scale biomedical data illustrate the performance of our methods.
Contact: Lars Ruthotto
