Image Processing Reference
In-Depth Information
17.2.1
R
EALIGNMENT
Changes in signal intensity over time, from any one voxel, can arise from head
motion and this represents a serious confound for fMRI studies. Despite
physical restraints on head movement, subjects can still show displacements
of upto several millimeters. Realignment involves estimating the six parame-
ters of an affine “rigid-body” transformation that minimize the (sum of
squared) differences between each successive scan and a reference scan (usu-
ally the first or the average of all scans in the time series), and applying the
transformation by resampling the data using trilinear, sinc, or spline interpo-
lation. Estimation of the affine transformation is usually effected with a first-
order approximation of the Taylor expansion of the effect of movement on
signal intensity using the spatial derivatives of the images (see the following
subsection). This allows for a simple iterative least square solution that cor-
responds to a Gauss-Newton search (4). Even if this realignment were perfect,
other movement-related signals (see the following text) could still persist. This
calls for a further step in which the data are adjusted for residual movement-
related effects.
17.2.2
A
M
-R
E
MRI
DJUSTING
FOR
OVEMENT
ELATED
FFECTS
IN
F
In extreme cases, as much as 90% of the variance in fMRI time series can be
accounted for by the effects of movement even after realignment (5). Causes
of these movement-related components are due to movement effects that cannot
be modeled using a linear affine model. These nonlinear effects include: (1)
subject movement between slice acquisition, (2) interpolation artifacts (6), (3)
nonlinear distortion due to magnetic field inhomogeneities (7), and (4) spin-
excitation history effects (5). The latter can be pronounced if the repetition time
(TR) approaches T
, making the current signal a function of movement history.
These multiple effects render the movement-related signal (
1
y
) a nonlinear
function of displacement (
).
By assuming a sensible form for this function, its parameters can be estimated
using the observed time series and the estimated movement parameters
x
) in the
n
th and previous scans
y
=
f
(
x
,
x
,
…
n
n
n
−
1
from
the realignment procedure. The estimated movement-related signal is then
simply subtracted from the original data. This adjustment can be carried out as
a preprocessing step or embodied in model estimation during the GLM analysis.
The form for ƒ(
x
), proposed in (5), was a nonlinear autoregression model that
used polynomial expansions to second order. This model was motivated by the
spin-excitation history effects and allowed displacement in previous scans to
explain the current movement-related signal. However, it is also a reasonable
model for many other sources of movement-related confounds. Generally, for
TRs of several seconds, interpolation artifacts predominate (6) and first-order
terms, comprising an expansion of the current displacement in terms of periodic
basis functions, are sufficient.
x
Search WWH ::
Custom Search