Addressing the path-length-dependency confound in white matter tract segmentation

Matthew George Liptrot, Karam Sidaros, Tim B. Dyrby

20 Citationer (Scopus)
6732 Downloads (Pure)

Abstract

We derive the Iterative Confidence Enhancement of Tractography (ICE-T) framework to address the problem of path-length dependency (PLD), the streamline dispersivity confound inherent to probabilistic tractography methods. We show that PLD can arise as a non-linear effect, compounded by tissue complexity, and therefore cannot be handled using linear correction methods. ICE-T is an easy-to-implement framework that acts as a wrapper around most probabilistic streamline tractography methods, iteratively growing the tractography seed regions. Tract networks segmented with ICE-T can subsequently be delineated with a global threshold, even from a single-voxel seed. We investigated ICE-T performance using ex vivo pig-brain datasets where true positives were known via in vivo tracers, and applied the derived ICE-T parameters to a human in vivo dataset. We examined the parameter space of ICE-T: the number of streamlines emitted per voxel, and a threshold applied at each iteration. As few as 20 streamlines per seed-voxel, and a robust range of ICE-T thresholds, were shown to sufficiently segment the desired tract network. Outside this range, the tract network either approximated the complete white-matter compartment (too low threshold) or failed to propagate through complex regions (too high threshold). The parameters were shown to be generalizable across seed regions. With ICE-T, the degree of both near-seed flare due to false positives, and of distal false negatives, are decreased when compared with thresholded probabilistic tractography without ICE-T. Since ICE-T only addresses PLD, the degree of remaining false-positives and false-negatives will consequently be mainly attributable to the particular tractography method employed. Given the benefits offered by ICE-T, we would suggest that future studies consider this or a similar approach when using tractography to provide tract segmentations for tract based analysis, or for brain network analysis.

OriginalsprogEngelsk
Artikelnummere96247
TidsskriftPLoS ONE
Vol/bind9
Udgave nummer5
Antal sider11
ISSN1932-6203
DOI
StatusUdgivet - 5 maj 2014

Emneord

  • Det Natur- og Biovidenskabelige Fakultet

Fingeraftryk

Dyk ned i forskningsemnerne om 'Addressing the path-length-dependency confound in white matter tract segmentation'. Sammen danner de et unikt fingeraftryk.

Citationsformater