A nice 2010 Human Brain Mapping paper by Church, Petersen & Schlaggar covers a number of interpretational issues confronting modern neuroimaging. Their particular application is pediatric neuroimaging (I will also use developmental examples), but the general issues apply to nearly all fMRI studies. So here are some important things to keep in mind whenever you read an fMRI study:
1. “The Performance Burden” If neural activity is found to differ between groups or conditions, you can’t necessarily make inferences about differences in neural information processing – this could reflect behavioral differences alone. For example, if younger children perform worse than older children but also recruit a few different regions, those neural regions might be operating in exactly the same way across ages (e.g., processing errors) – there’s just more errors in the younger group! Even with similar levels of performance there’s the possibility that children of different ages are using different strategies, different amounts of mental effort, or are differentially reacting to the closed, loud, and claustrophobia-inducing space that is an fMRI scanner.
2. Movement & “Resting” State. A related issue is that younger children may find the movement restrictions imposed by neuroimaging more restrictive than older children. One influential neurophysiologist has even said that resting-state neuroimaging (when no task is being performed) itself constitutes a response inhibition task for children! Thus, even if movement correction and registration algorithms are perfectly accurate (which they’re almost certainly not, with potentially profound effects on functional connectivity analyses), there are cognitive factors to consider about the scanner environment itself that may differ between groups or conditions.
3. The Task B Problem.. One way of dealing with differences like those highlighted above is to compare a task of interest (“Task A”) with a control “Task B”, and compare how two groups differ from one another in terms of the contrast (task A – task B). Inferences can then be made on the basis of the task of interest… at least, if task B is directly compared across groups. The problem is that people frequently don’t report this. Unfortunately, even if they did, comparisons of “task B” across groups is itself a contrast of (“Task B” – baseline), where baseline is a trial with no explicit cognitive task. See “Movement & ‘Resting’ State” above.
4. Below-baseline Activity. Church et al emphasize that reductions in hemodynamic activity do not necessarily mean that a circuit is being inhibited, or even if it is being inhibited, that its function is not therefore being expressed. Here’s an example. Let’s say you’re running a task where subjects must quickly and abruptly move their eyes from one part of the visual field to another. The substantia nigra pars reticulata (SNpr) is crucial to enabling this type of eye movement, because it inhibits the expression of eye movements until they are appropriate. Thus, SNpr gets turned off when an eye movement is made. If fMRI showed a reduced response in SNpr as a result of its reduced neural activity during eye movements, one might think it’s not involved in eye movements. However, SNpr is crucially relevant for eye movements and is participating in producing the task-relevant eye movement precisely via its lack of activity. So be careful with any strong inferences being made on the basis of negative BOLD. Also keep in mind that inhibition actually results in an increased local BOLD response, at least near the cell bodies of the inhibitory interneurons.
There are undoubtedly more issues to keep in mind (two more: the above are all distinct from the voodoo correlations and dead salmon issues, which are each distinct from one another). And Church et al describe other issues I haven’t covered here.
But these 4 are certainly a very, very good start.