| |
Abstract:
Although often ignored in practice, it is acknowledged that
functional Magnetic Resonance Image (fMRI) time series are
sometimes nonstationary. It is important to know about
nonstationarity because some statistical procedures assume
stationarity, and also because nonstationarity itself may indicate
some underlying physiological process which provide richer
information than conventional methods do. We applied a test of
nonstationarity developed by Kwiatkowski, Phillips, Schmidt and
Shin (1992) to fMRI data collected from a motor task and an
auditory task. This method tests if the time series contains a
random-walk component. The null hypothesis is that there is no
random-walk component and thus the time series is stationary. In
the auditory task, a subject listened to an intermittent pure
tones. In the motor task, another subject performed intermittent
fingertapping. Functional MRI data were acquired with a 1.5T magnet
every 200 msec. for 30 sec. We found that 20% and 21% of the pixels
within the brain were nonstationary for motor and auditory tasks
respectively (p=0.0025). We also found that brain regions that
yield high scores in the conventional t-test tend to be stationary:
correlations between alpha for t-test and nonstationarity test were
r=-0.63 and r=-0.74 (for areas with alpha < 0.0001 for either
t-test or nonstationarity test, which amounts to 23% and 21% of
pixels within the brain) for motor and auditory tasks,
respectively.
|
