Trying to design your first neuroimaging study? Here are some thoughts you might want to consider (see also at the end some very useful links plus a PDF of my slides from my presentation at the DUKE-NUS Neuroimaging Short Course (2012)):
· Make sure you have concrete independent variables and concrete predictions. Assume you had collected the data and imagine some ideal results. How would you interpret them?
· The brain records more than what we think! Therefore:
o Aim for orthogonal designs; make sure that you do not actually measure more than one dimensions that are collinear.
o Watch out for low-level differences: visual effects, reaction time, eye movements.
o Try to identify the underlying cognitive processes that might overshadow your main process of interest (attention is a usual suspect).
· If you are using event-related design then jitter appropriately; randomize presentation of stimuli.
· Look for appropriate control tasks.
· Allow for at least 20-25 measurements per condition
· Make sure you have a balanced number of trials per condition.
· Go simple! Do not try to answer everything in one experiment
· Do the task yourself. Watch out for boredom, especially for tasks that study ‘higher’ cognitive functions
· Test – test – test! It is great if you have behavioral differences!
· Interpretation: Watch out for reverse inferences and other conceptual errors.
Also, here are some algorithms to help you optimize your design:
1. Design Magic – Multicollinearity assessment for fMRI designs SPM99 SPM2
Summary: Allows you to assess the multicollinearity in your fMRI-design by calculating the amount of factor variance that is also accounted for by the other factors in the design (expressed in R^2). Essentially, the higher R^2 is, the lower the t-values will be. Also can calculate (and create) the most optimal high-pass filter for your design.
Author: Matthijs Vink
2. GA – Genetic Algorithm for Experimental Design SPM99 SPM2 SPM5 SPM8
Summary: Optimizes event related fMRI designs by statistical and psychological criterion. (Matlab Signal Processing Toolbox required).
Author: Tor Wager