[FieldTrip] ITC and harmonic artefacts?

Wed May 10 12:00:44 CEST 2023

Yes, I think the harmonics part I now completely get :) Very last, easiest
question from me ever. What can be cited on the claim I have observed a few
times, that ITC best uses a minimum of 100 trials to avoid false positives?

Thanks in advance!

Ivo

Am Mi., 10. Mai 2023 um 11:58 Uhr schrieb Schoffelen, J.M. (Jan Mathijs)
via fieldtrip <fieldtrip at science.ru.nl>:

> Ah, you want to compute ITC time resolved? Isn’t it then trivial to get
> what you call  ‘harmonics’ in the ITC? This is just due to the ISI, and the
> steady state response being non sinusoidal in nature.
>
> JM
>
>
> On 10 May 2023, at 11:44, Ivaylo Iotchev <ivaylo.iotchev at gmail.com> wrote:
>
> By defining trials I mean where the trials start, not how long they are.
> If there is phase-alignment relative to word-beginning a trial starting at
> the word beginning and one starting in the middle of the word will not be
> aligned in phase with each other, this is why I think starting point
> matters. The length of course should not be too short :)
>
> I am roughly speaking using a version of this How can I compute
> inter-trial coherence? - FieldTrip toolbox
> <https://urldefense.com/v3/__https://www.fieldtriptoolbox.org/faq/itc/__;!!HJOPV4FYYWzcc1jazlU!4tklQRla8xdpPiI-P2OykbDFQDU5Ib33v0KStRMbTPpcBGFJb2NfKQgVFDb8NAn1X4AUvT2v14b2Qjox20_A7TGBG3z7vF8rJOJKs5Q$> with
> some changes adapted to the data.
>
> Bests,
>
> Ivo
>
> Am Mi., 10. Mai 2023 um 11:25 Uhr schrieb Schoffelen, J.M. (Jan Mathijs) <
> janmathijs.schoffelen at donders.ru.nl>:
>
>> Hi Ivo,
>>
>> The additional info provieded is even confusing me further.
>>
>> If you are interested in studying entrainment, there are a few avenues to
>> take:
>>
>> - the time domain avenue, using temporal response functions, a.k.a. time
>> shifted regression, where the regressors are a series of time-shifted stick
>> functions.
>> - the time domain avenue, using averaging across trials first, followed
>> by a spectral transformation to assess the power spectrum of the steady
>> state response.
>> - frequency domain avenue, using spectral transformation of long snippets
>> of data, keep the phase information, and then either
>>    * average across trials, and then take the absolute value, which is I
>> think you aim to do with what you call ITC and which is roughly equivalent
>> to the previous avenue, but then in the frequency domain,
>> or
>>    * compute coherence/plv etc between the neural signals’ FFTs and the
>> FFT of an appropriately defined stimulus signal (can be a stick function,
>> or an audio envelope). this approach is robust against trial-by-trial
>> fluctuations of the exact timing of the stimuli.
>>
>> The starting point of what you call a ’trial’ is irrelevant when you take
>> the frequency domain route, as long as your epochs are sufficiently long in
>> order to be able to capture the frequencies of interest. I don’t understand
>> why you would need/want to define the trials relative to syllable onset,
>> you need to define long trials, to get sufficient spectral resolution.
>>
>> Good luck,
>> Jan-Mathijs
>>
>>
>> On 10 May 2023, at 10:58, Ivaylo Iotchev <ivaylo.iotchev at gmail.com>
>> wrote:
>>
>> I am sorry Jan Mathijs, and thank you for bearing with me!
>>
>> I mainly wanted to make sure that harmonics can meddle with ITC, but I
>> also found an answer to another sub-question which might be of interest to
>> other yet inexperienced with this type of analysis.
>>
>> As you suggested, some more information first... we are looking at
>> responses to speech streams, wherein word boundaries occur at a rate of 1.3
>> Hz and syllables follow each other at a rate of 4 Hz... the reason a 1.3 Hz
>> response is not visible when trials are defined relative to syllable start
>> is that every word consists of 3 syllables, meaning that two out of three
>> trials cannot be expected to show frequency alignment with the recurrence
>> of word boundaries (since only every 3rd syllable marks the beginning of a
>> new word). Hence I discovered the maybe trivial to some users rule, that
>> you have to select your starting points relative to the question at hand,
>> i.e. when looking for word-level entrainment all trials should be defined
>> relative to the word-beginnings, otherwise there will be a lot of trials
>> that introduce noise.
>>
>> I hope I make more sense now :) Thank you dearly!
>>
>> Ivo
>>
>> Am Mi., 10. Mai 2023 um 08:25 Uhr schrieb Schoffelen, J.M. (Jan Mathijs)
>> via fieldtrip <fieldtrip at science.ru.nl>:
>>
>>> Hi Ivo,
>>>
>>> I understand less than half of what you are writing. If you don’t
>>> provide any additional details about what (and how) you are trying to
>>> achieve, the only thing we can do as readership of this list (about 1800
>>> people), is to call in the cavalry.
>>>
>>> I asked my pal chatGPT, and this is what I got for an answer:
>>>
>>>
>>>
>>> "It sounds like you are observing an effect in your inter-trial
>>> coherence analysis that is influenced by the distance between the
>>> time-points defining the trial-start.
>>>
>>> One possibility is that you are observing harmonics of the frequencies
>>> of interest, as you suggested. Harmonics are multiples of a given
>>> frequency, and they can produce similar patterns of activity in the brain.
>>> Therefore, it's possible that your analysis is picking up on harmonics of
>>> the frequencies you're interested in.
>>>
>>> Another possibility is that there may be some kind of artifact in your
>>> data that is causing the effect. For example, if there is some kind of
>>> noise or artifact in the data that is strongest at particular time
>>> intervals, this could cause the coherence values to be higher or lower at
>>> those intervals.
>>>
>>> To help determine the cause of the effect you're observing, you could
>>> try a few things:
>>>
>>>    - Check your data for artifacts: Look at the raw data to see if
>>>    there are any obvious sources of noise or artifact that could be causing
>>>    the effect you're observing. You could also try preprocessing the data
>>>    differently (e.g., applying different filters, using different reference
>>>    electrodes) to see if this changes the results.
>>>    - Conduct a control analysis: Run a control analysis where you
>>>    randomly shuffle the trial-start times and then compute the coherence
>>>    values. This will help determine if the effect you're observing is due to
>>>    the specific trial-start times you've chosen or if it is a more general
>>>    effect.
>>>    - Consider other possible explanations: Think about other possible
>>>    explanations for the effect you're observing. For example, it could be
>>>    related to the way the brain is processing the stimuli you're presenting,
>>>    or it could be related to some other aspect of the experimental design.
>>>
>>> Overall, it's difficult to say exactly what is causing the effect
>>> without more information, but these are some things you could try to help
>>> narrow down the possibilities."
>>>
>>>
>>>
>>> Perhaps that will help to get you going.
>>>
>>> Good luck,
>>> Jan-Mathijs
>>>
>>>
>>>
>>> On 8 May 2023, at 13:04, Ivaylo Iotchev via fieldtrip <
>>> fieldtrip at science.ru.nl> wrote:
>>>
>>> Dear community,
>>>
>>> for some reason choosing the distancing between time-points defining
>>> trial-start for my inter-trial coherence analysis seems to affect the
>>> results... high coherence is observed for each frequency which is
>>> x/distance. What is funny is that trial duration and exact timing of
>>> pre-stimulus baseline and post-stimulus end are not influencing this
>>> picture, only the distance between the points that define trial start. For
>>> example if I space my trials starting points 750 milliseconds apart, I will
>>> get increases for 1.3 Hz, 2.6 Hz, 4 Hz... eventually 8 Hz... and so on...
>>>
>>> To some extent I wouldn't be surprised if these are effects related to
>>> harmonics and I frankly do not expect a zero value for frequencies that are
>>> harmonic to the frequencies of interest, but maybe I am running up an
>>> artefact??
>>>
>>> Bests,
>>>
>>> Ivo
>>> _______________________________________________
>>> fieldtrip mailing list
>>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>
>>> https://urldefense.com/v3/__https://doi.org/10.1371/journal.pcbi.1002202__;!!HJOPV4FYYWzcc1jazlU!4SCrZ_DqN1nri9GmVju9DYmBzMHuZQY-8-u-6WziIXAseaG7LxEk2GKR3CmdM2aT2OHklcPd8qRPNc7R1KpoxDQ5v8tGddzvKu7XWQ$
>>>
>>>
>>> _______________________________________________
>>> fieldtrip mailing list
>>> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>> https://urldefense.com/v3/__https://doi.org/10.1371/journal.pcbi.1002202__;!!HJOPV4FYYWzcc1jazlU!42ilkRccxfUDGDjSsIlmOUQlK8X0uW7RDEyc7xOUwa7YKp5uFf_fRMSv9ad3epn2dC1p-toxNgHPKuKYgWUwZPtW440UIw$ 
>>> <https://urldefense.com/v3/__https://doi.org/10.1371/journal.pcbi.1002202__;!!HJOPV4FYYWzcc1jazlU!9wYerVVUCrjjSU3BjGzS3jThyJEdwfVuSJJgJ4k-U_THv4pzzh1bFvC21Mw5W7_UYEnHgEAfzgbokmlACvcW2fUMsN_JagTomV8GGuU$>
>>>
>>
>>
> _______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://urldefense.com/v3/__https://doi.org/10.1371/journal.pcbi.1002202__;!!HJOPV4FYYWzcc1jazlU!42ilkRccxfUDGDjSsIlmOUQlK8X0uW7RDEyc7xOUwa7YKp5uFf_fRMSv9ad3epn2dC1p-toxNgHPKuKYgWUwZPtW440UIw$ 
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20230510/44abe9a1/attachment.htm>