Dear Anusha,
>I am trying to analyse fmri data collected using sparse temporal sampling
>technique using SPM2.
>
>The experimental design as follows:
>
>For a TR of 12 secs a partial volume of brain is acquired for first 1.5
>sec and the scanner is silent for 10.5 sec. This pattern was repeated for
>about 8 mins acquirinng in total 40 partial volumes. The task was a
>listening task where the auditory stimuli was presented during the silent
>period alternated with silent base line. Basically it is an ABAB design
>with partial volumes acquired at the begining and at the end of a block.
>
>I have analsyed this data set using spm99 with TR=12sec and boxcar basis
>function convolved with heamodynamic response function. The results looked
>very good with very strong activation in the auditory regions.
>
>Howevr, I am unable get the same results with spm2. I dont get the
>auditory regions activated, even at very low significant level. I have
>tried with TR = 12 sec, basis function as heamodynamic response function
>wih time derivatives and 0 sec duritaiion for the trials. I have also
>tried 12 sec duration for the trial as I thought this may produce a basis
>function similar to a boxcar function.
>
>What is the correct method to analyse this data ? what are the correct
>inputs to spm2 to reproduce spm99 results ?
That's an interesting question. I think the problem is in the convolution
with the HRF, with such a long TR. First, with this TR there is no
information
in your data to estimate efficiently the temporal derivative. Therefore, you
should just use the canonical HRF. Second the default bin sizes for the
convolution
are 1/16 of the TR. This is getting a bit long for a 12 sec. TR. Could you
change the defaults to 1/64 of a TR? If this does not work I would trick SPM2
into not convolving at all; You can do this by using a single basis function
which is 1 for one time bin. e.g. a Fourier set with one component and a
very small duration (temporal support).
If this does not work, use PET models and simply construct the ABAB regressors
by hand (serial correlations in fMRI can be discounted with a 12 sec TR).
If you do this, do not forget to put the low frequency confounds (i.e. drift
terms) in the design matrix.
I hope this helps - Karl
|