Dear Leonhard:

> -----Original Message-----
> From: Leonhard Schilbach [mailto:[log in to unmask]] 
> Sent: Saturday, June 16, 2007 4:36 AM
> To: d gitelman
> Subject: Re: [SPM] hrf peak
> 
> Hi Darren,
> 
> I have a question related to your suggestion of using an FIR 
> basis set.
> Do I understand it correctly that when using this, the 
> impulse response function is not convolved with the hrf in 
> the usual way (i.e. when using the canonical hrf set)?

The stimulus onsets are convolved with an FIR basis set rather than an hrf.
The FIR basis is mathematically the same as selective averaging but it does
not require one to line up the trial onsets and the TR. Basically each
column of the design estimates a time bin occurring after a stimulus onset.
If you then examine the FIR parameter estimatets using an F-test you will
likely see something that looks like an HRF not because you imposed a
synthetic hrf but because that was the shape of the bold response. 

> But is the delay (neural activity -> BOLD change) still 
> modelled? 

Yes. You will see a peak somewhere between 2 and 6 bins after a stimulus
onset.

In other
> words: when specifying FIR time bins in my analysis, am I 
> looking at BOLD change within these exact time bins or at 
> signal change which occurs later and is, thus, (temporally) 
> associated with the time bins?

Not sure what you mean. The estimate is relative to each time bin, but you
will see the delay intrinsic in the BOLD response because each time bin is
an independent column in the design.

The one caveat with FIR models is that when examining contrasts you have to
use F-tests as that allows including all the time bins. (This can be done
either in SPM at the second level or one can put the data in a statistics
program like SPSS and perform an ANOVA with one of the factors being time,
which corresponds to the parameter estimates from each time bin.)

Regards,
Darren

> Best,
> 
> Leonhard
> 
> 
> 
> Am 16.06.2007 3:21 Uhr schrieb "d gitelman" unter
> <[log in to unmask]>:
> 
> >  Hi Kathleen
> > 
> > As you note the HRF + TD will account for some shifts in 
> the timing of 
> > the HRF. You could explicitly test for this by examining 
> the hrf and 
> > TD responses using F-tests at the 2nd level. You could also include 
> > the dispersion derivative in the analyses to account for additional 
> > variance in the canonical hrf.
> > 
> > Another way of analyzing the data would be by using an FIR 
> basis set. 
> > This would eschew any specific form for the HRF and allow you to 
> > assess for changes in peri-stimulus BOLD responses based on 
> your FIR time bins.
> > 
> > 
> > Darren
> > 
> >> -----Original Message-----
> >> From: SPM (Statistical Parametric Mapping) 
> >> [mailto:[log in to unmask]] On Behalf Of Kathleen W. Smith
> >> Sent: Friday, June 15, 2007 8:02 PM
> >> To: [log in to unmask]
> >> Subject: [SPM] hrf peak
> >> 
> >> Dear Dr. Friston, Dr. Worsley, and other SPM experts,
> >>  
> >> One of my committee members is setting me a very interesting
> >> challenge:
> >>  
> >> "...my primary concern is confirming that your analysis is not 
> >> compromised by using an incorrect hrf peak. We spent a lot of work 
> >> deciding on the correct oxygenation peak delay to use in our 
> >> analyses, and perhaps now (many years later) 6s is 
> considered stable 
> >> across studies. I've never seen that statement made, so my 
> concern is 
> >> that each study may need a different hrf peak.
> >> This is critical since all your analyses are based on the 
> assumption 
> >> that oxygenation will peak at 6s, and if it doesn't (or more 
> >> troubling, if the peak differs by condition, which I 
> sincerely hope 
> >> it doesn't and I don't expect), your analyses are severely 
> >> compromised. All that said, you can placate my concerns by 
> choosing 
> >> one set of predictions (your choice, but choose the most critical 
> >> hypothesis), and showing that the observed hrf peak is the 
> same for 
> >> all conditions, and that the peak is close to 6s."
> >>  
> >>  
> >> Having read Kalina Christoff's procedures on identifying 
> individual 
> >> hrf's, I don't believe that I can establish the hrf for 
> individual Ps 
> >> retroactively from my existing data.
> >> Christoff indicates that such a study would be designed so that no 
> >> stimulus would evoke a new hrf until the previous one had run its 
> >> course.
> >>  
> >> I know that the canonical hrf peaks at six seconds and 
> that this is 
> >> grounded in theory.
> >>  
> >> I have been analysing my data using SPM2, using hrf+time 
> derivative.
> >>  
> >> I wonder whether there is an indirect way to address this 
> question. 
> >> Would any of you mind pointing me in the right direction?
> >>  
> >> Many thanks!
> >>  
> >> Kathleen Smith
> >> 
> 
>