HI Eugene, thanks for you answer. I followed the instruction to calculate the delay, including masking with the contrast of interest. I'm not that familiar with the underlying math, so I still have some questions about the interpretation:
Q1) As I understand from the website the results are in 'units of % of TRs'. So let's say that I have a TR of 2 second and two clusters that I'm interested in.
Cluster A that has a delay of around 50 and another Cluster B with around 100.
That means that Cluster A has a delay of 1 second, while Cluster B of 2 seconds. Correct?
Yes.
Q2) Do negative values indicate a shift forward in time? So basically are nonsense (at least in a simple stimuluation experiment), since how can the activation occur before the actual stimulus?
Not necessarily. Remember that the standard model assumes that the blood flow response takes around 6 seconds to peak, but this may vary across regions. You may also get activity associated with anticipation. Because of the slow, uncertain HRF, it is often not straightfoward to reach strong conclusions about the timing of neural activity from fMRI data. Calhoun's 2004 paper on tds (Neuroimage) is worth looking at to get a feeling for some of the subtleties of interpreting models incorporating tds.
Q3) I read that with the temporal derivative one can capture delays of around 1 TR. How should I interpret then the extreme values, e.g. 160868 in my case? Would you set a cut-off at 100% for reasonable values?
Extreme values will be associated the temporal derivative fitting to signal independently of the main regressor (the time shift value is large because the fit to the main regressor is small). This could be related to noise or temporal responses of a form where a non-zero t.d. fit reduces variance but the main regressor does not. It may be worth using FEATquery or MRIcron to look at the responses in your cluster in more detail.
Q4) In case I'm interested in longer delays - what method could I use?
I haven't looked at the literature recently, but you might use multiple response evs with different onset times, or analyse simple parameters derived from individual trials, e.g. time to peak.
Eugene
My apologies for so many questions.
Thanks, Michael
On 29-May-09, at 4:53 AM, Eugene Duff wrote:
Hi Michael,
Have a look at the temporal derivative section of the FEAT practical:
Also, you don't actually need to define the t.d.s separately. Their parameter estimates can be found in the stats directory, following the EV they are associated with. If you want to generate contrasts using the t.d. EVs, select "Real Evs" at the top of the Contrasts tab in the setup.
Cheers,
Eugene
2009/5/29 Michael Scheel
<[log in to unmask]>
Hi, i'm interested in the temporal difference in activiation, so I thought I define the temporal derivative as a real EV and then look at the Parameter Estimate for this EV.
Now I have values for this pe of between -1500 and 2200. Can someone explain how to interpret these. Is there a better method to look at the temporal shift?
Thanks, Michael
--
Eugene Duff
FMRIB Centre,
University of Oxford
John Radcliffe Hospital, Headington OX3 9DU Oxford UK
Ph: +44 (0) 1865 222 739 Fax: +44 (0) 1865 222 717
--
--
Eugene Duff
FMRIB Centre,
University of Oxford
John Radcliffe Hospital, Headington OX3 9DU Oxford UK
Ph: +44 (0) 1865 222 739 Fax: +44 (0) 1865 222 717
--
