Hybrid surface-volume segmentation for improved anatomically-constrained tractography

Abstract

The "Anatomically-Constrained Tractography (ACT)" framework enhances the biological accuracy of diffusion MRI streamlines tractography reconstruction, by constraining the propagation and termination of streamlines based on tissue segmentation from a high-resolution, high-contrast anatomical image (e.g. T1-weighted). The tissue segmentation image to be used in this context has historically been derived using a sequence of processing tools provided within the FSL software package - in particular the intensity-based segmentation of the "fast" command - as automated in the MRtrix3 software command "5ttgen fsl". This can however lead to erroneous segmentations in the presence of image noise and/or poor tissue contrast, with ACT concomitantly applying inappropriate anatomical priors in such locations. The ACT framework itself is however independent of the particular approach used to derive this image, and thus may benefit from the use of more advanced tissue segmentation algorithms. Here we demonstrate a novel processing pipeline for the derivation of a tissue segmentation image for ACT. It combines information from both surface-based reconstructions of various structures, and volume-based tissue segmentations for structures for which the former is not performed explicitly; we thus name this algorithm "Hybrid Surface-Volume Segmentation (HSVS)". Because these data are mapped back to a common image space, its utilisation does not require modification to the ACT software framework, nor does it incur any substantial computational performance penalty during streamlines propagation.