I can’t find any one particular source, but I have heard it mentioned in passing in a number of presentations and lectures, that an optimal block length for a blocked design should be around 20s, with block lengths of over 40s becoming problematic.
I was curious what this estimate relies on. As far as I can tell, the optimal block length should be long enough to approach the maximal signal saturation, yet short enough to allow highpass filtering of drift and other low-frequency noise.
I have never heard of any metric which can balance these two limitations against each other, though, so I don’t know hoe exactly this would converge on 20s.
Perhaps it is something simpler and some widely used pipelines have a hard-coded highpass filter of 1/40Hz?
I believe this has to do with the frequency characteristics of the hemodynamic response. For example, see this paper: http://www.sciencedirect.com/science/article/pii/S1053811997902639. It might also have to do with the linearity of the signal: http://www.sciencedirect.com/science/article/pii/S1053811912000997 (note non-linearity relatively affecting short blocks more).
Simply looking at the signal in Fourier domain suggest 16 sec is actually optimal – you reach high signal and for a given session length have maximum number of blocks see my website http://www.sbirc.ed.ac.uk/cyril/fMRI3.html ‘the detection power increases with high frequency alternation because i) it depends on the number of events/blocks and ii) the noise in the BOLD time course which occurs mainly at low frequencies. Blocks with durations longer than the hemodynamic response reach a compromise between signal strength and noise (optimal 16s)’ - it comes from work of Rik Henson on designs
So baisically your guideline is the main frequency component of the HRF function (which for the “canonic” HRF used in humans just so happens to be 16s), yes?
So ideally, if I determine my hemodinamic impulse response function via a FIR set in pilot experiment, a better estimate would be the main frequency component of that, right?
And I guess the way you address the upper bounds of block length is by saying “well, we want many block repetitions, and a highpass filter as stringent as possible, so we’ll just go with the shortest feasible blocks”. This argument sounds simple enough, but it is very qualitative, wouldn’t you say? Certainly there are benefits associated with longer blocks - such as making sure that the signal returns to baseline.
For instance with a 16s block and a 20s HRF (with the main frequency component of 16s), if you would stimulate for 8s you wouldn’t expect the signal to retorn to baseline before the next block starts, no? you would have to wait at least 20s after the end of the 8s stimulation block, no?
This wiki entry by Rik Henson is illuminating: http://imaging.mrc-cbu.cam.ac.uk/imaging/DesignEfficiency
The content is adapted from an old FIL paper but to be honest the wiki entry is easier to follow.
You mention baseline. Careful about what you are trying to optimise. Efficency for contrasts between conditions will be different from efficiency for the contrast between condition and baseline.
so yes the idea is that you have max number of block given the duration of the hrf
regarding Johan’s comment on efficiency - this is quite different because the optimisation is not about how long should your block be but how should they be ordered (but both both cases are impacted by your filter)