Sampling Requirements for Highly Convoluted Brain Surfaces with FMRI

1
Sampling Requirements for Highly Convoluted Brain
Surfaces with FMRI
Ziad S. Saad, Jakub Otwinowski, Robert W.
Cox Scientific and Statistical Computing Core.
National Institute of Mental Health, National
Institutes of Health, Department of Health and
Human Services, USA
Introduction
Methods Results
Surface-based brain imaging analysis is used for
detailed mapping of brain activation patterns and
changes in cerebral gray matter. Due to the
highly convoluted nature of the cortical surface,
the topology of activation along the cortical
sheet can be obscured by the volumetric grid used
to sample brain activation. These sampling
errors are likely to occur at brain locations
such as A and B (Figure 1) where the shortest
path along the surface (SAB) is much longer than
the distance in R3 (RAB) SAB / RAB gtgt 1. In such
instances brain activity from functionally
distinct areas will be sampled with one voxel (a
sort of aliasing). We present 1- A method to
automatically delineate brain locations
susceptible to aliasing 2- Illustrate the effect
of aliasing on measured brain activity 3-
Propose methods to avoid or account for aliasing
A
B
Figure 6 To gain a practical sense of the
effects of aliasing we simulated retinotopic
polar angle data and sampled it with a volumetric
grid 1- Draw polar angle map in visual cortex
(A) 2- Create time series at each node as a
function of phase and add some noise 3- Sample
timeseries from pial surface with voxel grid of
4x4x4 mm 4- Compute response delays in voxel time
series 5- Map delay data back onto pial surface
(B)
Volumetric Grid and Surface Topology
Figure 7 A- Normalized changes in the
distribution of phase data caused by sampling.
Note the underrepresentation, relative to the
original distribution, of certain phases. B-
Average Sab/Rab ratio (see Figure 1) per phase
value. Note how decreases in certain phase
representation coincided with phases occurring at
nodes with high Sab/Rab ratios. (extreme bins
centered at 1 and 180 should be ignored because
of edge artifacts of sampling). This suggests
that horizontal and vertical meridia, typically
located in regions of high ratios, are not as
well sampled as horizontal meridia.
A
B
Figure 1 Volumetric sampling obscures the
topology of activation. The two points A and B,
though distant on the cortical surface, are
juxtaposed in the FMRI grid (4mm voxel size).
Volume-based interpolation will
disproportionately alter the topography of
activation at points such as A and B from the
topology at other points at less crucial
locations.
Phase (degrees)
B
A
Methods Results
A
Brain Locations Susceptible to Aliasing
B
Phase (degrees)
Figures 8 and 9 Respective equivalents of
Figures 6 and 7 using gray matter as a source of
signal as opposed to the pial surface. Note the
moderate reduction in sampling errors.
Conclusions
We have developed methods for automatically
delineating brain regions susceptible to sampling
errors. Variance from sampling errors that are
unavoidable at a given voxel size varies across
brain regions and can be reduced with selective
filtering at aliased areas. Sampling errors can
distort the true pattern of activation as
illustrated with the simulation using the
retinotopy data set.
Software Implementation
Horton, J. C. and W. F. Hoyt (1991). "The
representation of the visual field in human
striate cortex. A revision of the classic Holmes
map." Arch Ophthalmol 109(6) 816-24. Cox, R. W.
and J. S. Hyde (1997). "Software tools for
analysis and visualization of fMRI data." NMR in
Biomedicine 10(4-5) 171-178. Reprint
Requests ziad_at_nih.gov
The proposed methods have been implemented and
included with the distribution of AFNI and SUMA
http//afni.nimh.nih.gov See also Poster
PT-49 by P.C. Christidis et al. Poster
PT-50 by G. Chen
Figure 2 Highest SAB/RAB ratio along the pial
surface. The same data are shown on three
versions of the surface pial, white matter and
inflated, to expose data in buried sulci while
facilitating orientation. As expected, areas most
susceptible are located on the lips of sulci.
Note that occipital and temporal lobes are more
susceptible to mapping errors (aliasing) than the
parietal and frontal lobes.