Hi, all,
I have used this optimization program to compute rapid event fMRI schedules and it does many useful things
1. It calculates the efficiency of your timings using Henson's work (the higher the efficiency, the better your script) 2. it computes the probability with which every condition follows each other (so you can know for sure that the order of the stimuli is truly random) 3. It inserts random "null time" or inter-trial time (an important factor you need to maximaze the chances that the GLM will correctly estimate the hemodynamic response). 4. it calculates your best possible design given the constraints you specify (from a set of 10000 or so randomly generated designs)
In order to use this program, you need to know how long your experiment is going to be. So it does not tell you for example, how much stimulation time you need to get good data. You need to set this up independently. The rule of thumb is that you should allow as much null (or down time or baseline) as one of your conditions. So if you have 2 conditions with 40 events in each and the stimulation time of each event is 2 seconds, you have a total of 160 seconds of stimulation time in your experiment. This means that at the minimum you should allow 80 seconds of null time (preferably more, as you will see, your efficiency increases with more null time you have). Then you will tell the program that your experiment last a total of 120 TRs (this is the total number of points). Then you tell the program how many events you have and how long they last, and it calculates the best possible design given these temporal constraints. The program can also optimize the design for a particular contrast of interests, from the conditions you set up. The efficiency calculations are in the summary file that the program produces. The program also produces the best possible schedules (you need to specify how many schedules you want to keep of all the one it generates).
I think it is a good idea to run different possible versions and see which design is best, as one can play around with the length of the experiment (the longer the better).
On a more theoretical note, some people like Henson and Friston thinks that these random designs are efficient enough to estimate hemodynamic response to individual events (Friston et al, 1999). The key to this ability relies on the random temporal structure of the events. This is why coming up with the proper timings is critical. Rapid event designs are very common in the literature, particularly in some fields . Like with any design, which design is best depends on the question you are asking, for the reasons that Tim was pointing out. If one is looking to detect an effect for the first time, block designs have better detection power. But if you are planning to do a ROI analysis, detection power is less of an issue, as you will be able to model the hemodynamic response well enough to capture activity within the ROI.
Moreover, if one has practical constraints, rapid event designs may be the only viable choice. For example, inability to block stimuli due to their predictability may force you to use such designs: Go/no- go tasks are useless in block designs because all the no-go trials would have to be blocked together, which means you do not really have no-go trials with response conflict (going vs. not going). In such cases, you have to intermix the go and the no-go trials. Situations in which you do not want your participants to figure out your conditions (something that becomes clear from the blocking of stimuli) may be another case that requires event related designs, or situations in which you want to identify activity to correct or incorrect responses. Moreover,
Thus, many factors enter in deciding which design to use. Claire and I went for a rapid design because we did not want our participants to develop strategic performance and be aware of the manipulation, but this probably comes at the cost of less detection power. We seem to be modelling the hemodynamic response pretty well, so for example, voxels in visual areas correspond exactly with the timing of our stimuli.
If people would like more help on the use of this program, feel free to contact me.
Silvia
On 15 Nov 2007, at 10:11, Antony Morland wrote:
Hi All,
I agree with all that Tim wrote. Rik Henson has published a few papers on optimizing event related designs - you can find them here http://www.mrc-cbu.cam.ac.uk/~rh01/refs.html#_meth. Looking through them might help.
Tony
-----Original Message----- From: Tim Andrews [mailto:t.andrews@psych.york.ac.uk] Sent: 15 November 2007 09:58 To: ynic-users@ynic.york.ac.uk Subject: Re: Event-Related optimum schedule
Hi Laura,
Rapid event-related fMRI has many advantages for cognitive neuroscience research. However, the main problem with this technique is that the signal is very small compared to a block design. One way to increase the signal-to-noise ratio is to increase the number of trials. This typically means using a short ISI (otherwise your subjects could be in the scanner for a long time!). It is also good to vary the ISI to avoid expectation effects (i.e the subject predicting when the next event is likely to occur). However, the problem with a short ISI is that the response to one stimulus will likely overlap with the response to the next. This is not a problem if the BOLD response is linear (i.e. the response to two successive stimuli is the same as adding the response to two independent stimuli with an appropriate temporal offset). However, a number of studies have found that there are significant non-linearities when the ISI is less than ~5sec (eg Dale and Buckner, 1997; Huetell and McCarthy, 2000). So, varying ISI can have positive and negative effects on the BOLD signal. I haven't used the programs that Claire and Silvia are using, but I assume they are trying to find an optimum balance between these effects.
Users - please feel free to correct or comment!
Tim
Laura Lee wrote:
Hi MRI-support,
I'm struggling along trying to work out how to create a 'stochastic' event-related design for fMRI. Claire Moody has passed on a program that searches for the optimum stimulus schedule (she & Silvia used it for their last project). I've read over all the bumpf but am still quite confused by all the new concepts. I think I know roughly what I want but then there are some parameters I am unsure about and don't really understand the implications of the settings. I'd be really grateful if you could give me a hand.
This is the programme I downloaded... http://surfer.nmr.mgh.harvard.edu/optseq And there's a pretty comprehensive help page here... http://surfer.nmr.mgh.harvard.edu/optseq/optseq2.help.txt
Thanks, Laura
-- Dr Tim Andrews Department of Psychology University of York York, YO10 5DD UK
Tel: 44-1904-434356 Fax: 44-1904-433181
http://www-users.york.ac.uk/~ta505/ http://www.york.ac.uk/depts/psych/www/admissions/cns/
-- ynic-users mailing list ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
-- ynic-users mailing list ynic-users@ynic.york.ac.uk https://www.ynic.york.ac.uk/mailman/listinfo/ynic-users
Silvia Gennari Department of Psychology University of York Heslington, York YO10 5DD United Kingdom