This toolbox is for the optimisation of experimental designs for fMRI. Minimising the variance of the design matrix will help detect or estimate (depending on the outcome of interest) the effect researchers are looking for. The genetic algorithm for experimental designs was introduced by Wager and Nichols (2002) and further improved by Kao, Mandal, Lazar and Stufken (2009). We implemented these methods in a userfriendly web-application and introduced some improvements and allows more flexibility for the experimental setup.
Goal of the optimisation
Let's say a researcher wants to design an experiment with 3 possible conditions: A (seeing happy faces), B (seeing sad faces) and C(seeing angry faces). The first goal of the optimisation is to provide an optimal order of the conditions.
For example: compare the designs with stimulus order AAAABBBBCCCC and AABBCCBBCCAA. With the first design, you'll have an optimal BOLD response to distinguish the different stimuli, but it will be hard to distinguish the design from scanner drift. Moreover, the first design is very predictable. In this case, it might be advisable to use the second example. This is an example of how we optimise the stimulus order.
A second example is the exact timing of the stimuli. For example: a whole brain scan is taken every 2 seconds (TR=2s), and the stimuli are also presented every 2 seconds (ITI=2s). We assume that the HRF (the fMRI signal after activation) is a continuous function that peaks somewhere around 4 seconds and then has a slow decay. However, with this design, you will measure the exact same timepoint on the HRF. You won't be able to estimate the whole HRF. Moreover, a whole brain scan is taken in slices. For example: the temporal lob (or part of it) will be measured 0.5s after stimulus presentation and the parietal lobe will be measured after 1s. This is clearly unwanted.A layman's solution is to set the ITI different from the TR. But an even better solution is to spread the stimuli at specific timepoints that will enable to estimate the hrf at many different time points, with a variable ITI. This toolbox searches for optimal stimulus timing.
Once the design optimisation is run, the output will be a file for each stimulus type (or condition) with stimulus onsets. As such, it captures both the stimulus order and the exact timing.
Next: Experimental Outcomes