Dear Mike and James,
> Hi. I posted a message the other day about a proposed fMRI experiment,
> but was not clear in my description of the proposed design.
>
> The study involves using sparse imaging, that is to say, for each 12
> second TR, 3-4 word stimuli will be presented auditorily during the
> initial 7.6 seconds of the TR during when there is no scanner noise,
> this then will be followed by the 4.4 second volume acquisition, during
> which no stimuli will be presented. This is a kind of halfway house
> between block and event related designs. The question is, for such a
> design, how many volumes would be necessary per condition. I would
> like it to be 24, with that I could run the ideal number of
> conditions. If it must be more, then I will need to lose a condition.
> We will be using a 3T magnet.
> Fron James:
> I think what they mean by sparse imaging here is acquiring a single
> image (of acquisition time say 4s) every 12 seconds, enabling a quiet
> period of 8 seconds between images in which to present words with
> perfect clarity. This reduces by a third the number of data points
> acquired for each trial. Fortunately, the image will be acquired at
> the time of maximal hrf response to the words (six seconds after the
> midpoint of the word presentation period) so power will hopefully be
> reduced by less than two-thirds in practice.
Ahh. Now I understand - thank you to both of you. I think 24
replications would be sufficent if the signal is sufficently high
relative to noise. The Chalfont group,in the UK, have used a similar
sparse acquisition protocol to detect complex parital seizure related
hemodynamic changes. They usually require about 32 observations to
obtain reliable results. Assuming you had at least 2 conditions (with
48 (=24x2 activation conditions and 48 baseline - giving 19.2 mintute
sessions) this would put you in the right ballpark.
> When you say that event related responses may be obtained from as few
> as 16 trials, with a TR of 2-4 seconds, you are preseumably saying
> that at least 60 to 80 data points must be obtained in all during the
> (~12s) hrf responses to those trials. If sparse imaging results in
> just one data point per trial, then a minimum of 60 to 80 trials would
> be needed.
Not really. With such a long inter-scan interval each data point can
be considered as an independent obervation (ignoring serial
correlations due to drift). As such 24 samples for each condition
should be sufficeint to make an inference (remember in PET one
typically uses 6x12 = 48 scans for at least two conditions).
With best wishes - Karl
|