Meeting Banner
Abstract #2078

Higher Order Variational Denoising for Diffusion Tensor Imaging

Florian Knoll1, Tuomo Valkonen2, Kristian Bredies3, Rudolf Stollberger1

1Institute of Medical Engineering, Graz University of Technology, Graz, Austria; 2Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, United Kingdom; 3Department of Mathematics and Scientific Computing, University of Graz, Graz, Austria

High resolution diffusion weighted imaging and diffusion tensor imaging with isotropic voxels are desirable for a large number of applications. Unfortunately, acquisitions of such data sets are challenging, due to the notoriously low SNR of DWI. In addition, even with fast sequences like single shot EPI, measurement times often become prohibitively long because of the large number of diffusion encoding directions that have to be acquired. In this work we introduce two higher-order variational denoising approaches to reconstruct DTI data with isotropic voxels from a single average which are based on Total Generalised Variation.

Keywords

absolutely acceleration acquisition acquisitions addition additional afterwards applications approaches audience available averaging become bottom calculations challenging channel choice chosen clinical coil column comparable computing considered contains convex correction defined denote described desirable diffusion directly displayed displays dual eddy edges employ encoding engineering enhancement ensured equation error even fast features fidelity field fine form function functional gain generalized gold gradient gradients ground head healthy implementation index individual individually institute introduce investigated isotropic knoll leads like long loss maps math mathematics matrix measuring medical middle minimization minimize model movements namely neck noisy norm noted notoriously numerical occurred often operates optimal partial physics pock practice preserving preset primal prior problems prohibitively property reconstruction reduce reduced reduction regularization represents researchers resolution restoring saddle seeking sequentially sets sharp shot slice slices slight slightly smith smooth solution solve space structural system tanner target technology tensor tensors theoretical third truth unfortunately unlike users variation variational volunteer white write