|
Dear Darren, Stefan and Aaron,
thank you for your comments and sorry for my delayed answer.
We think we have an explanation for the strange behaviour in our data (re
mail Tue, 25 Jan 2000). We have found that:
- There are global signal changes in our time series when we consider the
whole image intensity, i.e. brain + noise. These signal changes are of the order
of 1% above baseline noise and they have the form of a waveform pretty much in
phase with the house presentation stimuli. These signal changes were the reason for our aberrant analysis (whole
brain activity during house presentation) since we were using global
intensity scaling. This is what you Stefan suspected! If I do the statistics without global
scaling I get decent results: mainly V1 activation during the presentation of
houses and faces.
A question about SPM: how do I normalise the whole time series to a single
volume in the time series? The default option (Global intensity normalization:
scale/none) doesn't seem to allow for this.
- Now, where are these global signal changes coming from?
A more careful look at the time courses of different ROIs in the raw image
data shows that the signal changes are coming from the noise (i.e. areas
outside the brain) and not from the brain signal.
A comment: it seems that for the global intensity scaling SPM99b is not
using the brain signal only (by means of a mask). Therefore, large signal
changes in an image artifact outside the brain may affect your statistics.
- We are convinced that there was no significant motion during this scan
(i.e., large translations/rotations), in particular no strong motion in phase
with the house presentation stimuli. This is suggested by at least three different analysis:
a) The motion correction analysis gives very small realignment parameters
(less than 1mm and 1 degree). A power spectrum analysis of these parameters
shows that the peak frequency is centred at zero, and at the house presentation
frequency the intensity is less than 25% of the maximum.
b) A view of the data in a movie fashion shows no motion larger than 1
pixel (=3mm).
c) A simple t-test on the difference images of the raw image data does not
show the typical 'brain edge activation' usually seen due to movement in phase
with stimuli.
So, where are the noise fluctuations in the images coming from? [We used a
1.5T GE scanner, birdcage head coil, single-shot T2*-weighted spiral, TE=50ms,
TR=2500ms, 64x64, FOV=20cm, interleaved]
We think that they come from a small-motion artifact, which in spiral
images is spread in the overall image. We think that, for some reason, the
subject got particularly affected by the houses and had a kind of
tremulous motion during their presentation. This apparently did not happen
(or happened to a smaller degree) during the face presentation.
As a result, the noise level in our images got larger during the houses
than during the face stimuli.
Any comments are welcome!
Cheers,
Jorge
-------------------------------------------------------------------
Jorge Jovicich, Dr. rer. nat.
Dept. of Neurology | Computation and Neural Systems
Harbor-UCLA Medical Center | California Institute of Technology
1124 W Carson St. B-4 | 1200 E California Blv., BI 139-74
Torrance, CA 90502, USA | Pasadena, CA 91125, USA
310-222-5656/7 | 626-395-2882
310-222-5658 (fax) | 626-796-8876 (fax)
[log in to unmask] | [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
