> This leads to several serious limitations of fMRI, which are
> particularly problematic when someone tries to treat an fMRI image as a
> picture of "nervous system activity":
>> 1. fMRI is a map of changes in blood flow: nothing more, nothing less.
> It is conjectured, but not definitively proven, that active tissue in
> the CNS can modulate local blood flow to produce changes in density on
> an MRI scan. fMRI advocates would say that the blood flow changes are
> produced by whatever processes contribute to the mental activity being
> performed, but that's rather a circular argument. There's no proven
> causal chain between, say, a neuron firing more actively and a change in
> the amount of oxygenated blood flowing through the nearest capillary.
Neurovascular coupling has been demonstrated quite thoroughly by Grinvald's
group (among others), I believe. I have Villringer and Dirnagl (1995)
written down. May be time to review it.
>> 2. As far as I know, the mechanism by which blood flow is altered is not
> understood. It is likely to be strongly mediated by astrocytes, whose
> processes wrap brain capillaries in continuous sheaths, and hence would
> be controlled by glial cells, rather than neurons. Hence it seems
> likely that the needs of glia are more relevant to fMRI than the
> activity of neurons directly.
I think it's been shown that action potential propagation and synaptic
transmission require the bulk of the ATP available and are therefore
responsible for the hemodynamic signal. This doesn't address blood flow
specifically, though. However, recent evidence has demonstrated that the
fMRI signal can also exhibit the earlier oxymetry signal, and that the two
appear to be correlated. Regardless, the concern does become whether or not
one can pull out inhibitory responses from the overall signal, if synaptic
transmission is indeed a major component.
> 3. There is no absolute scale for what constitutes "active" levels of
> difference on the fMRI image. To a large extent they are set ad-hoc
> based on the levels observed in individuals. Hence it is difficult to
> compare between individuals and between studies. Normalised
> co-ordinates such as the Talairach system can help adjust for the
> structural differences in individual brains, but there's no way to
> adjust for the real differences in regional function between
> individuals, which have been demonstrated anatomically and
> physiologically but don't show up on MRI.
The possible exception would be found in the rare cases when intraoperative
techniques such as microstimulation/recording and optical imaging can be
used to verify things that fMRI doesn't have the spatial resolution to
> 4. Nobody has any idea if blood flow is modulated by the *type* of
> processing in a given region - aside from the ignorant people who
> believe we only use 10% of our brains, most neuroscientists would agree
> that even during "resting", there is sustained and large scale activity
> in the nervous system, and we are looking for fluctuations on a much
> smaller scale, which may or may not be relevant to what the region is
> doing. For instance, active suppression of activity in a region is
> almost certainly more metabolically expensive than normal processing, so
> a strongly suppressed area would show up as more "active" on an fMRI
> scan. Histological stains which show markers of high metabolic need are
> usually labelling inhibitory neurons, which tells us who the major
> "blood consumers" are in the brain.
>> 5. Similarly, it is possible that changes in patterns of activity,
> rather than the overall amount of activity, could be far more relevant
> in processing tasks and stimuli. The idea of synchronisation has been
> widely proposed as a way that processes could be "bound" together in
> consciousness, but that only involves altering the timing of spikes, not
> the overall number. Presumably that wouldn't show up on fMRI at all,
> since the tissue's metabolic demands would be unchanged.
Even in most electrophysiological studies those ideas aren't examined.
People play with rate coding more often than not (and perhaps more often
than they should). fMRI, however, could provide some insights into
spontaneous activity. Optical imaging has revealed highly structured maps
(corresponding to orientation maps) in visual cortex in the absence of
stimulation. Surely fMRI is also showing something - we just don't know
what it is yet.
I share your sentiment about the shortcomings of the technique though. As
for the original comment/question about the brain surface, the columnar
architecture of the cortex has lulled many folks into the idea that knowing
what's going on in one layer is sufficient. Maybe that's true; with the
kind of resolution we're talking about, a single voxel is often
incorporating the entire depth of the cortex. But other methods that
examine hemodynamic responses, like optical imaging, maybe it's not such a