pvestimation.rst

Partial volume estimation

Surface-based partial volumes are calculated as the volume of intersection between a voxel grid and a closed surface that is contained within it. Some voxels will partially intersect the voxel; the interior fractions is the partial volume. The voxel grid of interest is referred to as the reference grid.

Partial volume estimation is performed on a surface-by-surface basis. For a structure defined by a single surface (eg thalamus), the function toblerone.pvestimation.structure will return a single array containing values in the range [0,1] where 1 denotes that a voxel is fully contained within the surface.

For a structure defined by multiple surfaces (cortex), partial volumes can be estimated by performing estimation for each individual surface in turn and then combining the results. The function toblerone.pvestimation.cortex does this with the pial and white surfaces, and then takes the voxel-wise difference between the two as grey matter. Currently the cortex is the only mulit-surface structure supported, though others could be processed by using toblerone.pvestimation.structure on the individual surfaces and combining the results as appropriate.

PV estimation functions

Toblerone estimates PVs across the brain using surface segmentations (for example, those from FreeSurfer and FSL FIRST). The following functions are available in toblerone.pvestimation:

• toblerone.pvestimation.cortex: estimate PVs for one/both hemispheres of the cortex
• toblerone.pvestimation.structure: estimate PVs for a structure delineated by a single surface, eg thalamus
• toblerone.pvestimation.complete: estimate PVs across the whole brain, including subcortical structures identified by FSL FIRST

The latter function can be used for a direct replacement for conventional PV estimation tools such as FSL FIRST as it provides complete whole-brain PV estimates for GM, WM and non-brain tissue.

Example usage

Partial volume estimation for the cortex via FreeSurfer:

import toblerone as tob 

fsdir = '/path/to/freesurfer_subject_dir'
ref = '/some/reference_image.nii.gz'

# Identity transform for registration of surfaces to reference image  
pvs = tob.pvestimation.cortex(fsdir=fsdir, ref=ref, struct2ref='I')

Partial volume estimation for a structure with a single surface:

import toblerone as tob 

# NB it is assumed that the surface is in world-mm coordinates, 
# if not use surface.transform() to get it so. 
surf = '/path/to/surface.surf.gii'
ref = '/some/reference_image.nii.gz'

# Identity transform for registration of surfaces to reference image 
pvs = tob.pvestimation.structure(surf=surf, ref=ref, struct2ref='I')

Partial volume estimation for the whole brain via FreeSurfer and fsl_anat:

import toblerone as tob 

fsdir = '/path/to/freesurfer_subject_dir'
fslanat = '/path/to/fsl_anat_dir'
ref = '/some/reference_image.nii.gz'

# Identity transform for registration of surfaces to reference image  
pvs = tob.pvestimation.complete(fsdir=fsdir, fslanat=fslanat, ref=ref, struct2ref='I')

Note also that toblerone can be used at the terminal. Type toblerone -h for help.

PVEc with oxasl

Toblerone has been integrated with oxasl to provide PV correction via the spatial Variational Bayesian (spatial-VB) inference. See oxasl and oxasl_surfpvc.