Probabilistic tractography for complex fiber orientations with automatic model selection

Conference Paper (2018)
Authors

Edwin Versteeg (Student TU Delft)

Frans Vos (Amsterdam UMC, Student TU Delft)

Gert Kwakkel (Amsterdam UMC)

Frans C.T. Van Der Helm (TU Delft - Biomechatronics & Human-Machine Control)

Joor Arkesteijn (ImPhys/Quantitative Imaging)

O.G. Filatova (TU Delft - Biomechatronics & Human-Machine Control)

Research Group
Biomechatronics & Human-Machine Control
To reference this document use:
https://doi.org/10.1007/978-3-319-73839-0_9
More Info
expand_more
Publication Year
2018
Language
English
Research Group
Biomechatronics & Human-Machine Control
Pages (from-to)
117-128
ISBN (print)
978-3-319-73838-3
ISBN (electronic)
978-3-319-73839-0
DOI:
https://doi.org/10.1007/978-3-319-73839-0_9

Abstract

Fiber tractography aims to reconstruct white matter (WM) connections in the brain. Challenges in these reconstructions include estimation of the fiber orientations in regions with multiple fiber populations, and the uncertainty in the fiber orientations as a result of noise. In this work, we use a range of multi-tensor models to cope with crossing fibers. The uncertainty in fiber orientation is captured using the Cramér-Rao lower bound. Furthermore, model selection is performed based on model complexity and goodness of fit. The performance of the framework on the fibercup phantom and human data was compared to the open source diffusion MRI toolkit Camino for a range of SNRs. Performance was quantified by using the Tractometer measures in the fibercup phantom and by comparing streamline counts of lateral projections of the corpus callosum (CC) in the human data. On the phantom data, the comparison showed that our method performs similar to Camino in crossing fiber regions, whilst performing better in a region with kissing fibers (median angular error of 0.73∘ vs 2.7∘, valid connections of 57% vs 21% when seed is in the corresponding region of interest). Furthermore, the amount of counts in the lateral projections was found to be higher using our method (19–89% increase depending on a subject). Altogether, our method outperforms the reference method on both phantom and human data allowing for in-vivo probabilistic multi fiber tractography with an objective model selection procedure.

No files available

Metadata only record. There are no files for this record.