At 01:47 04/02/2005 +0900, takawatanabe-tky wrote:
>Dear list  members.
>
>We are having some troubles in analyzing data in SPM2. The data was obtained
>in sparse temporal sampling technique.
>
>TR = 14 sec, Acquisition time is 3 sec( one whole brain set) i.e. MRI scan
>of 3 sec > no scan interval of 11 sec > MRI scan of 3 sec > ....
>
>
>1) When we adjust slice timing in preprocessing, the TR required in SPM2 is,
>in my case, 14 sec.or 3 sec.?

I think that for sparse-sampling data with a long TR, slice-timing
correction is not only unnecessary, but also to be discouraged.
Slice-timing correction is performed by interpolating between adjacent
scans in the time-series. Where each scan is 14 seconds apart, the
assumptions used in interpolation are likely to be less valid.

In some sparse-sampling experiments, the block length is equivalent to the
TR. That is, each scan follows events in a different condition. In such a
design interpolation average together scans in different conditions. This
will directly impair your ability to detect differences between conditions.

>2) How can we achieve no-hrf-convolving design in MRI design phase in SPM2?
>Previous studies said that they did not convolve hrf with data from sparse
>sampling.
>In addition, I hope someone knows the way to convolve " mean and exponential
>delay " with data in SPM2.

The way to avoid convolving is to specify an FIR model with only a single
bin and a basis set that is the same length as your TR.

>3) Does anyone know how much low cut (i.e. High pass) filter is appropriate
>in sparse temporal sampling?

I think that you can get away with no high-pass filtering. A typical
sparse-imaging design will induce changes in BOLD activation at a lower
frequency than an equivalent continuous imaging design. Also, the
time-series is sampled much less often in sparse imaging, the nyquist limit
will be correspondingly lower and therefore there will be much less
high-frequency information in the time-series. Both of these properties can
make it more difficult to remove low-frequency noise from the signal (using
the HPF) without impairing your ability to detect activitation.

My suggestion that you analyse data without applying a HPF is (I think)
unconventional for conventional fMRI. Potentially, this can introduce bias
into the statistics in your first level analysis (single-subject, or
fixed-effects analysis), since the number of independent observations will
be over-estimated. However, if you're conducting group or random-effects
analysis, this bias will not cause statistical problems in the second level
analysis.

These suggestions are based on my own experience of sparse-imaging. Perhaps
these do not have as sound a mathematical basis as some would like. I would
be interested to hear opinions from other SPM experts on these topics.

best wishes,

matt


****************************************************
  Dr Matt Davis
  MRC Cognition and Brain Sciences Unit
  15 Chaucer Road, Cambridge, CB2 2EF

  email: [log in to unmask]
    tel: 01223 273 637  (direct line)
    tel: 01223 355 294  (reception)
    fax: 01223 359 062

****************************************************