Biomedical Engineering Reference
In-Depth Information
withdrawal of the required volume. Finally the withdrawn blood
was gradually replaced back into the arterial system at a rate
slower than withdrawal. MR images were obtained 5 minutes after
replacement was complete.
2.3. Data Analysis
and Statistics
Low-pass filtered resting BOLD signal time courses from voxels
in the sensorimotor cortex were cross-correlated with every voxel
time course in the brain. Previously, it was observed that a sig-
nificant temporal correlation was obtained with voxels from the
sensorimotor and its associated cortex in humans
(25, 26)
.Very
few voxels outside the sensorimotor cortex were reported to have
significant correlation in humans. In the present study using the
rat model, the sensorimotor cortex, caudate putamen, hippocam-
pus and thalamus were defined as regions of interest according
to the stereotaxic rat brain atlas
(32)
. A gaussian low-pass filter
with a cutoff at 0.1 Hz was applied to all voxel time courses
(33)
.
This reduced respiratory and cardiac signals and any correspond-
ing aliased signals.
2.4. Frequency
Estimation
The frequencies present in the voxel time-series data were calcu-
lated using Welch's averaged periodogram method
(34)
. Briefly,
each data set was divided into eight sections, with 50% overlap
between adjacent blocks. A Hamming window was then used
for each section and the power spectrum computed. After the
power spectrum was calculated from each segment, the power
spectra were averaged for the eight segments. This was done
on a voxel-wise basis for the entire brain. After the power
spectrum was calculated, the frequency with the largest ampli-
tude was identified as the dominant frequency noted for all the
voxels.
2.5. Temporal
Correlation
The temporal correlation between voxels was assessed by two
different techniques. The first technique used cross-correlation
analysis between six seed pixels from the center of region cho-
sen from the function anatomy and was correlated with every
voxel in that slice. Since we are only interested in the tempo-
ral correlation due to slow periodic spontaneous oscillations, a
finite impulse response filter
(35)
was used to filter the high-
frequency components from each of the data sets (as described
above). Because of the short data size (110 time points), filter
parameters were adjusted to minimize the generation of arti-
factual frequencies (sidelobes). The correlation coefficient was
then calculated using the formula
r
X
∗
Y
(
X
Y), where
X
and
Y represent time course pixels from two voxels X and Y, and
* represents the dot product. All voxels that passed a corre-
lation of 0.3 were considered significant and locations were
noted.
=
/
