Matthew et al.,
> I know this is going to seem very tedious to everyone, but the
> thresholded map does not address my assertion 1, and therefore doesn't
> address localization. I feel this is a major point, as most of us
> believe that localization is the main advantage of brain imaging. To
> check - do you disagree with this? If so, why?
> Because the thresholded map is so poor for localization, we need to present
> the unthresholded map, if we want to talk about localization. In any case,
> given that it is easy to show the 'significant' areas on the continuous map, the
> thresholded map has no advantages that I can see.
With certain qualifications, I agree that localization is the main advantage of brain imaging.
Nonetheless, the thresholded map does have the distinct advantage of statistical reliability -- you
can be confident that there is greater brain activity in a specified location in the experimental
condition relative to the baseline. This does not give a complete picture of what areas are
involved in the experimental condition -- or the extent of the area involved -- but it limits the
uncertainty about whether or not the thresholded area is actually active.
The actual area that is active is likely to be somewhat larger than the thresholded map in every
instance, which raises your issue about whether the unthresholded map is preferable. Although
there are some advantages to interpreting data using unthresholded maps (which I think should be
described on Wiki or another SPM site for newcomers to the field), my general response would be
that unthresholded maps are generally NOT preferable for publications.
When I was in graduate school, I was warned against speculating too much while interpreting
experimental results. Speculation, I was told, was the realm of philosophers -- because science
was advancing so rapidly and tends to be accepted as factual, only strongly defensible
interpretations should be reported. Additional information / interpretations, I was told, would
only clutter the literature and make matters more difficult for scientists, who already have too
little time. Although I personally like looking at unthresholded maps when trying to figure out
patterns in my own data (or sometimes those of others), they invite overinterpretation. Whereas
thresholded maps likely underestimate the extent of activation, unthresholded maps undoubtedly
overestimate it. This is partly from the processing procedures -- especially normalization and
smoothing -- but also perhaps from the nature of the hemodynamic response itself. If I stick my
middle finger into ice-cold water, blood flow to my entire hand will soon be reduced. Similarly,
if exactly 1 cubic mm of cortex becomes extremely active, I expect that the increase in blood flow
may be expanded, and not purely restricted to that 1 cubic mm of cortex. If gradients of metabolic
by-products are inducing changes in blood flow, then there will be a larger area with increased
blood flow, but the maximal change in blood flow should still be directed to the mm of cortex that
became active. This is the advantage of thresholded maps -- it identifies regions with
sufficiently strong (or consistent) activation that they are unlikely to be identified falsely for
spurious reasons.
One other comment about localization and a suggestion for the field... There is a tendency to try
to localize functions to areas activated by an experiment. As pointed out by many others,
cognitive functions are carried out by networks that utilize information carried between many
areas, and not just a circumscribed region. An area activated by one class of stimuli in one set
of experiments (e.g., Broca's area preferentially activated by words) may be activated by an
entirely different class of stimuli in another set of experiments (e.g., music). To understand the
computational processes carried out by a brain area, it would be helpful to identify the many
different experiments that have identified activation in or near a circumscribed area. This type
of meta-analysis is difficult when only maxima are reported, and perhaps the unthresholded maps you
advocate could help. Personally, I think a Brain Mapping Project is needed comparable to the Human
Genome Project, attempting to identify the types of experimental conditions that are known to
activate each small parcel of the brain. Although difficult due to differences in methodologies,
such a map would certainly improve our understanding about regional brain function.
Doug Burman
|