Measuring white matter (made up of mostly axons) made possible based on the observation that water diffuses faster along axons than in cell bodies (but also diffuses faster in ventricles) (Basser et al. 1994 JMR). Diffusion is a random process without any coherent motion
In fMRI, we talk about k-space imaging (spatial frequency matrix based on the readout gradient), in DWI we talk about q-space imaging (based on the diffusion gradient...q is for quanta?).
Usually spin echo echo planar images (SE-EPIs) used with a pair of strong diffusion-weighting gradients straddling the 180° refocusing pulse. The diffusion-weighting applied is represented by the diffusion attenuation factor or simply the “b”-value. diffusion-weighted sequence becomes
M(b) = M0 · exp(−b · ADC),
where M0 is the spin-echo signal (including proton density and/or any T1/T2 relaxation) in the absence of any diffusion-weighting.
Lots of details on DTI acquisition sequence parameters and how that impacts DTI on Radiology Key's Survivor's Guide to DTI Acquisition.
Of course, Chris Rorden also has helpful pointers in his lab documentation on DTI with some scanner-specific info, as does the no-longer-supported community-support effort NA-MIC's wiki page on DTI acquisition (with lots of scanner-specific details, tho dated to 2009).
DTI = also look at the velocity and direction of diffusion. Based on observation that water diffusion is in all directions in GM (isotropic) but along axons in WM (anisotropic). To model this, create "tensors" (3 vectors to describe 3-D space -- just as MRI scanners have 3 spatial encoding coils to produce smaller/weaker magnetic fields that change the main B0 in 3 different axes that's used to create gradients to spatially localize signals received; sphere is when all vectors have the same length)....relate to diffusion orientation distribution (probability distribution) and eigenspace? Need to sample different directions to create the representative tensors (30-100 directions).
DTI is similar to a DWI acquisition but needs
at least 6 DWIs/diffusion directions, sampled across the space
in order to determine the 6 unknowns in 3D tensors:
at least one reference image: the B0 (AKA T2w) image, non-diffusion-weighted sequence (i.e. with no diffusion encoding/diffusion-encoding gradients turned off)
Probs in q-space imaging with crossing fibers and bends within a voxel unless use large field gradients and sample a lot. Alternative solution: high angular resolution diffusion imaging (HARDI) or q-ball imaging
EPIs, where the time to acquire a complete 2D slice is relatively long, are susceptible to spatial distortions arising from field inhomogeneities. The minimum possible echo time (TE) is usually adopted for SE EPIs. The rapid switching
Directional scans are susceptibility to eddy currents.
DWI is based on many model assumptions
and reconstruction needs to be able to distinguish signal changes due to actual diffusion and due to eddy currents -- for this reason, the more diffusion encoding directions that are sampled on a sphere (bvecs in FSL-speak, angular sampling), during image acquisition, the better
Need to correct for the susceptibility distortions (often using B0 images which are not susceptibility to eddy currents). e.g. FSL topup and eddy, respectively. Check out eddy for animated gifs of before and after correction.
Modulation - keep original volume the same by adjusting volume of spatially normalized volumes by the amount it has changed during the nonlinear spatially transformations.
Modulate or covary with overall brain volume, volume of GM/WM? The controversy....
Eckert et al. 2006
The difficulty in estimating ICV/eTIV on a T1 image is differentiating between skull and CSF, which are both dark on a T1. The more robust method is to use another image modality, but with a T1 image, the following approaches are possible.
SPM 8 seems to overestimate ICV compared to FreeSurfer (REF!). MORE INFO
There is no subroutine in FSL to do this.
FreeSurfer estimates ICV based on the relationship between the ICV and the linear transform to MNI305 space (estimating a scaling factor determined by generating the determinant of the atlas transform in <subject>/mri/transforms/talairach.xfm output by FreeSurfer's recon-all autorecon1 routine) -- basically, atlas normalization using appropriate template images provides an automated method for head size correction that is equivalent to manual TIV correction (Buckner et al. 2004). Therefore, ICV is estimated without needing to segment the brain. Only the first 5 processing steps of recon-all need to be done (motion correction, intensity bias correction, Talairach transform computation, intensity normalization 1, skull stripping) and mri_segstats can be used to estimate ICV based on talairach.xfm. For each T1 image $t1, this code snippet will create the directory/subdirectory $subjID/$subjID in $subjDir (I don't know what's up with the subject ID subfolder in the subject ID folder) and populate it with more subfolders with files output from recon-all:
recon-all -s $subjID \
-i $t1 \
-sd $subjDir \
-autorecon1 \
-parallel >> $log
Then loop through each created $subjID folder (FreeSurfer will automatically search through $subjID/$subjID), call mri_segstats to calculate TIV. The following code extracts the eTIV value itself and the determinant value with sed, which you can print to some file $f_eTIV with the subject ID:
eTIVdet=$( mri_segstats --subject $subjID \
--sd $subjDir \
--etiv-only \
| sed -n 's/.*(eTIV) = \([0-9].*\) mm^3.*det: \([0-9].*\) )/\1,\2/p' )
printf "%s,%s\n" $subjID $eTIVdet >> $f_eTIV