On the surface you are right, you could infer the A (the stimulus) causes B (evoked signal) as the temporal requirement is there as you present the stimulus.

However, there are many steps that must happen between A and B, which is perhaps why people don't want to say A causes B. 

Furthermore, most researchers don't want to infer that A causes B, but that A cause neural activity in the region where you observe B. This is more problematic because you are inferring something that hasn't been observed and is complicated by the many intermediate steps. 

Also, people have been trying to say that if B1 occurs for B2, that neural activity in B1 occurs first even though we don't have the temporal resolution to determine that is true.

Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Postdoctoral Research Fellow, GRECC, Bedford VA
Website: http://www.martinos.org/~mclaren
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.

On Mon, Feb 17, 2014 at 6:36 AM, Francesco Puccettone <[log in to unmask]> wrote:

Hi Anderson,

Thanks very much for your answers! In case anyone else would like to have a go at question 5), I'd be very grateful.

Francesco

On 16 February 2014 16:51, Anderson M. Winkler <[log in to unmask]> wrote:

Hi Francesco,

Please, see below:

1) It seems that in behavioural analyses, the importance of effects is quantified&compared based on the "effect size" statistic, whereas in MRI this is based on "how significant" they are, i.e. on how small the p-value (or, equivalently, how large the t-value) is. Would it not make sense for activation maps to be colour-indexed by e.g. cohen's d instead, rather than taking a higher t-value to mean "stronger activation" when it really means "more reliable activation"?

I don't agree with the first statement, that in behavioural as a rule people would favour effect sizes over significance testing, and in fMRI it would incidentally be just the opposite. But regardless, nothing prevents anyone from reporting effect size maps, percent signal change, and/or Cohen's d. I doubt any reviewer would ever complain, and at worst, you can always put these extra maps as a supplemental material. Effect size and significance tests are no replacement for each other. Instead, they complement each other.

2) Some studies find a long list of regions for contrast A>B, but none when the conditions are reversed in the contrast (B>A). Does that really mean that in the entire brain there isn't any region where the BOLD signal is significantly higher for B than for A? This would seem to imply a violation of some sort of conservation law: assuming a more-or-less constant amount of cerebral bloodflow is available during both conditions, then surely if there are regions where this bloodflow is distributed differently between conditions, then this should come out of both contrasts, not just one.
This question is, I guess, equivalent to "why isn't the result of a B>A contrast simply the opposite of A>B?"

Simply reversing the contrast for 1st level fMRI doesn't produce necessarily a contrast as meaningful as the original one, and may not be accurate at all, as there isn't an "upside-down HRF", and to model the opposite response requires careful considerations about what the hypothesis and shape of the response would be. It's different than at higher-levels, when groups of subjects are compared and the reverse contrast is almost always meaningful.

Also, there is ongoing reseach on the so called negative BOLD, and its origins may well differ from those of the "positive" BOLD -- another reason to suspect that simply negating the contrast may not bring correct answers.

Regarding the idea of a conservation law, the brain isn't a closed system: it's very well connected to the rest of the body and receives a good share of the cardiac output (near 15%). And the body itself isn't a closed system either. And in any rate, BOLD-fMRI doesn't measure either flow or volume directly, but a complex local interaction of these and the consumption of O2, and how these influence the MRI signal. It's hard to hypothesise that there would be a conservation law for something as the BOLD signal.

3) Is the relation between the concepts of "activation" and "deactivation" similar to the relation between a "A>B" contrast and a "B>A" contrast? in other words, if "deactivations" are found in a "A>B" contrast, are they equivalent to "activations" in a "B>A" contrast?

This may well happen in block designs just by reversing the baseline condition, but for the reasons above, it's not a general rule. Plus, in fMRI, the conditions can be quite complex, including the "baseline" itself.

4) In order to obtain the activation statistical map in a functional analysis, when would each of these methods be applied (either instead of, or in addition to the other two methods)?
    A. correlation between the time series of the HRF-convoluted BOLD and that of the behavioural stimulus that was used
    B. 1-sample t-test between the BOLD and a resting baseline
    C. 2-sample t-test, i.e. contrast, between condition A and condition B
Is it the case that each method is only suited to one particular type of question, or is there in fact overlap between them?

Correlations, t-tests, etc, are all particular cases of a more general framework -- the general linear model (GLM). So, in principle, choosing between one or another is irrelevant, and they would all give the same result for testing the same hypothesis. However, as convolving the stimulus function with the HRF is, nearly always, biologically more plausible and statistically more powerful, (A) takes precedence over (C) -- the last could in practice be used only in simple block-designs, and even so, modelling the HRF and do it all with the GLM is in general a better idea and makes better use of the collected data. I don't understand your option (B), but regardless, it won't be in general better than GLM+HRF.

5) One could say that fMRI experiments can prove A correlates with B (A~B), where A is the event (e.g. stimulus) and B is the evoked BOLD signal. However, in general, for two events A and B, if A~B and if temporal precedence and lack of 3rd variable can be proved, i.e. if it can be shown that A consistently happens before B and there is no obvious C factor that is a cause for both A and B, then this allows one to infer "A=>B" from "A~B". Those two extra conditions are often met in fMRI, so why it that it's wrong to interpret fMRI results causally?

There are others in the list that have considered these causality issues in more depth than I, so I won't try answer this one. Sorry.

All the best,

Anderson