[Eeglablist] average reference and connectivity

[Roy Cox]

hi all,

I've been recently made aware of this simulation study where different
dipoles with different phase relations are modeled and power and phase
estimates are then obtained using different montages. ( Tenke and Kayser,
2015, http://www.sciencedirect.com/science/article/pii/S0167876015001907)

it appears that CSD is the overall winner, and the average reference gives
some spurious results quite in line with my worries as I first expressed
them.

Best,

Roy


On Mon, Jun 22, 2015 at 5:23 PM, Roy Cox <roycox.roycox at gmail.com> wrote:

> Thanks all:
>
> Dezhong: definitely a relevant paper, thanks!
>
> Pal: I like the CSD approach and I've used it in the past. I never
> considered that differences in impedance between left/right mastoids leads
> to a virtual reference that "floats" in between somewhere. Very good to
> know.
>
> Mathis: I'm aware of methods that get rid of zero-phase lag connectivity,
> and I've used PLI in the past. But as I see it, average reference would
> result in opposite phases for distant areas (frontal peaks matched by
> posterior troughs). I'm not sure how these methods deal with anti-phase
> connectivity. Plus I would imagine phase differences are never going to be
> exactly pi so you'd still pick up connectivity there..
>
> Best,
>
> Roy
>
>
>
> On Fri, Jun 19, 2015 at 5:20 AM, Mathis Kaiser <mathis.kaiser at charite.de>
> wrote:
>
>>  Hi Roy,
>>
>> not claiming to be an expert on matters of averaging, but it might be
>> useful to think about the connectivity measure you're applying:
>> while spurious connectivity is an issue when using the PLV, it should not
>> be present for measures that are based on the imaginary part of coherency
>> (and therefore don't take into account zero-phase lag synchronization, see
>> Nolte et al. 2004). Proposed improvements on the ImC include the PLI (Stam
>> et al. 2007) and wPLI (Vinck et al. 2011).
>>
>> Cheers,
>> Mathis
>>
>>
>> On 17.06.2015 19:47, Roy Cox wrote:
>>
>> Hi all,
>>
>>  I would like to get some expert opinions on the use of the average
>> reference when investigating phase-based connectivity (e.g., PLV, PLI,
>> etc), and a potential problem when using this approach.
>>
>>  While no referencing scheme is optimal, it is often argued that the
>> average reference offers "the best" solution given a sufficient amount of
>> electrodes (we use 60). The reference can be interpreted as the "height"
>> from which the topographical landscape of voltage amplitudes is viewed.
>> While any perspective is valid, average referencing places the viewpoint at
>> the average of all electrodes, which is declared zero. Importantly, this is
>> done on a sample-by-sample basis, meaning that the average is always zero.
>>
>>  Enter widespread synchronous oscillations. I've often noticed that when
>> strong in-phase alpha activity is present over posterior cortex, the
>> average reference results in equally strong and anti-phase alpha
>> oscillations over anterior regions (with a small region in between where
>> alpha activity is relatively absent). Similarly, during deep sleep there
>> are very strong frontal slow oscillations that are inverted in polarity
>> over posterior regions.
>>
>>  Now, any phase-based metric will return beautiful long-range
>> (anti-phase) connectivity, which is entirely (or at least largely) an
>> artefact of the referencing.
>>
>>  When using average mastoids as a reference (also not perfect - I know),
>> it is evident that there is no phase reversal from anterior to posterior
>> areas: alpha activity is (visually) absent from frontal regions, and sleep
>> oscillations are synchronous (in-phase) across most of cortex (but, by
>> necessity, relatively small close to the sites used for referencing).
>>
>>  In sum, while I'm sure the average reference is valid on a
>> sample-by-sample basis, problems seem to arise when time enters the
>> equation and widespread large negativities have to be matched by widespread
>> positivities to keep the average zero.
>>
>>  I imagine amplitude-envelope correlations could also suffer from
>> spurious, average reference-induced, oscillations.
>>
>>  This is all terribly hand-wavy and non-mathematical, so it would be
>> great if someone could comment on this to support or disprove my reasoning.
>>
>>  Roy
>>
>>
>> _______________________________________________
>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
>> For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu
>>
>>
>>
>> _______________________________________________
>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>> To unsubscribe, send an empty email to
>> eeglablist-unsubscribe at sccn.ucsd.edu
>> For digest mode, send an email with the subject "set digest mime" to
>> eeglablist-request at sccn.ucsd.edu
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20150626/0e4e7466/attachment.html>