¥ ÒNoiseÓ in FMRI is caused by several factors, not completely characterized
H MR thermal
noise (well understood, unremovable)
H Cardiac and
respiratory cycles (partly understood)
o In principle, could measure these
sources of noise separately and then try
to regress them out
å RETROICOR program underway (R Birn & M Smith of
FIM/NIMH)
H Scanner
fluctuations (e.g., thermal drift of hardware)
H Small
subject head movements (10-100 mm)
H Very low
frequency fluctuations (periods longer than 100 s)
¥ Data analysis should try to remove what can be removed and allow for the statistical effects of what canÕt be
removed
H ÒSerial
correlationÓ in the noise time series affects the t- and F-statistics
calculated by 3dDeconvolve
H At present,
nothing is done to correct for this effect (by us)