UnthresholdedEffectMaps - MRC CBU Imaging Wiki

Revision 1 as of 2007-02-24 13:37:52

Clear message
location: UnthresholdedEffectMaps

Using unthresholded effect size maps

We often face the situation where a particular effect in our imaging data does not survive strict familywise error control, but for various reasons does seem plausible enough to report. In this situation, and indeed in general, one attractive option is to report the entire effect size map of your analysis.

If you want to be able to report the areas that were significant at corrected thresholds also, this is easy to do with a combined map like this one: attachment:contin_eg.png.

The argument for unthresholded maps is twofold: unthresholded maps give better data with which to localize function, and are more useful for metaanalysis.

Unthresholded maps are useful for localizing function

The thresholded maps we are all used to can be seriously misleading for localizing function:

Jernigan TL, Gamst AC, Fennema-Notestine C, Ostergaard AL. More "mapping" in brain mapping: statistical comparison of effects. Hum Brain Mapp. 2003 Jun;19(2):90-5

The argument in this paper is very simple.

Let's say you only present a thresholded SPM map for task X, and you only found area A activated above threshold. You then say: area A is involved in task X. However, as we all know, just because something isn't significant, doesn't mean it isn't there. So, the fact that A _is_ significant, and the rest of the brain isn't, is perfectly compatible with the whole brain being actually activated, and A being just above threshold due to noise. So, when we say A is activated by X, we are really saying, maybe the whole brain is activated by X, and A got above threshold. We can't even say A is _particulary_ activated by X, unless we test the level of activation in A directly against the rest of the brain. So, A is activated by X, on its own, has very little value for localizing function. Therefore, the classic thresholded SPM has very little value for localizing function.

So, in order to localize function, you need some estimate of what is, and what is _not_ activated. The simplest way of starting to assess this, is a continuous map.

With a thresholded map, when you see a significantly activated area, you can only conclude something like "area A was significantly activated, but of course that isn't to say the whole brain wasn't activated about the same amount, who knows?". To get round this, you would have to compare activation in brain areas directly, and this is extremely rare. The continuous map on its own does not provide you with such a test, but at least it allows you a preliminary comparison, and makes the problem much clearer.

The unthresholded map is more useful for meta-analysis

In my experience, continuous maps give a much clearer picture of the quality of the data.