Meeting Banner
Abstract #3581

Outlier Detection for High B-Value Diffusion Data

Kerstin Pannek1, David Raffelt2, Christopher Bell1, Jane Mathias3, Stephen Rose1

1The University of Queensland, Brisbane, Queensland, Australia; 2Brain Research Institute, Australia; 3University of Adelaide, Australia

Diffusion weighted images are prone to artefacts caused by physiological noise. Existing model based approaches for voxel-wise identification of such artefacts rely on the diffusion tensor model, which is problematic in crossing fibre areas and at higher b-values required for high angular resolution diffusion imaging. We developed a voxel-wise identification method based on a higher order model of diffusion, and compared outlier probability maps obtained using the tensor model with those obtained using a higher order model in a cohort of 103 healthy participants.

Abstract PDF E-Poster Video

Keywords

amplitude ants bell bias blue bottom brain capable cardiac caused circle closely collinear computed considered constrained contain contrast corona corpus correction crossing cyan deconvolution decrease degree degrees described detection deviation diffusion directly distributed dropout ejection every excluded excluding fitting freedom frequencies frontal gating half harm harmonic harmonics head healthy homo identification identified identify inaccurate individual inhomogeneity institute intensities investigate isotropic iteration iterative larger limitation linear logical mainly male maps metrics model movement noise note orientation outliers output overcome package participants population populations prep previously probability process processing pron prone proposed pulsation quantitative reject rejection related reoriented required residuals rest restore robust rose sign sing slight software space spatial spread subsequent summarized susceptibility template temporal tensor tensors tore transformed umber update variability vary walker warps wind yellow