We have been playing around with modeling the EDA data collected during our emotion regulation fMRI paradigm. We are presenting IAPS pictures that range from negative to positive in valence with some neutral pictures mixed in for about 10 seconds (followed by a 3 second subjective rating period and a 6-9 second ISI). We have been trying to understand our data (because it really isn't behaving as we would have expected) and finally decided to just plot the raw data and noticed that the SCR was starting before the trial onset. Upon adding 6 seconds of the proceeding ISI to each stimulus onset, we noticed that there are actually two SCR peaks occuring, one in that ISI window (about 5-6 seconds after the ISI period has begun) indicating an anticipation phenomenon occuring after the previous trial has ended but the next hasn't started yet (perhaps indicating a "what kind of picture is about to show up" phenonmenon) and then an additional little peak 4-5 seconds after the stimulus onset, presumably a reaction to the picture.
We are wondering how to model both of these SCRs since they are contiguous and potentially overlapping. I noticed a post on the forum regarding stimulus presentations of 5 seconds and you suggested running the DCM model twice, breaking the trial in half. With our data as an example, would we run the model once using the onsets of the stimulus and then run the model again using the ISI onsets, or run a model with both onsets? Does this mean “model as two stimulus classes” (i.e., 50 stimuli instead of 25 stimuli), or does this mean, “run the analyses two times, first with 25 stimuli sets that model the events themselves, then again with 25 regressors that model the anticipation period”?
I also noticed in the manual under the section about the linearity prinicple of the peripheral model that non-linearities may occur if the time interval between two SN firing bursts is veryshort (presumably what we are seeing) and that it is possible to model non-linearities in a linear model using Volterra kernels like in fMRI. But it also says this isn't done with SCR models. We were wondering why this is.
Thank you in advance!
Stephanie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
please apologise the late reply, we've been quite busy in the past week or so.
If you can identify pre-trial anticipatory SCR in your grand averages, it does make sense to model them. You would run the DCM once (not twice) but define event windows before a trial as well. That is, each "trial" would consists of some time window during the ISI and all the events during the actual trial. To not bias the inversion, I would suggest defining the time windows for responses during the ISI on a grand average of all trials, not split up by conditions. Note you need to take into account the peak latency of the SCR - i.e. if you are observing a peak at 5 seconds before the next trial, then the actual sympathetic burst should be modelled around 4 s earlier, to account for the delay of the sudomotor system. (In all PsPMs the timing is in terms of the "central" event of interest, not the ensuing body change.)
Volterra kernels is a nice idea but the implementation is not straightforward and nobody has actually done this yet. We have paper hopefully coming out soon in Psychophysiology that uses intraneural stimulation to better characterises the limits for when non-linearities occur, but could not get enough data to really model these non-linearities, so this remains a challenge.
Hope this helps
Dominik
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
No worries at all! Thank you so much for the helpful reply! I was already thinking that the best way would be to set up the timing files with two events per trial.
One additional question, because the ISI and the next stimulus are contiguous and it appears based on our grand mean plots that the anticipatory peak is occurring one scond (or maybe even less) before the next stimulus onset, so the anticipatory peak and the SCR to the stimuli possibly overlap, do we need to do anything with the default inversion options? We have just been leaving them as is, but because of the potentitally overlapping nature of the responses, we are concerned that we might need to change those default options, but we aren't entirely sure if or how we would need to change them.
Thank you again in advance for your help!
Stephanie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
PsPM is set up to deal with overlapping SCR, as long as they are far enough apart to still be represented by a linear system. You just need to model the "central" (external/psychological/neural) input into the system.
Dominik
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi Dominik,
We have been playing around with modeling the EDA data collected during our emotion regulation fMRI paradigm. We are presenting IAPS pictures that range from negative to positive in valence with some neutral pictures mixed in for about 10 seconds (followed by a 3 second subjective rating period and a 6-9 second ISI). We have been trying to understand our data (because it really isn't behaving as we would have expected) and finally decided to just plot the raw data and noticed that the SCR was starting before the trial onset. Upon adding 6 seconds of the proceeding ISI to each stimulus onset, we noticed that there are actually two SCR peaks occuring, one in that ISI window (about 5-6 seconds after the ISI period has begun) indicating an anticipation phenomenon occuring after the previous trial has ended but the next hasn't started yet (perhaps indicating a "what kind of picture is about to show up" phenonmenon) and then an additional little peak 4-5 seconds after the stimulus onset, presumably a reaction to the picture.
We are wondering how to model both of these SCRs since they are contiguous and potentially overlapping. I noticed a post on the forum regarding stimulus presentations of 5 seconds and you suggested running the DCM model twice, breaking the trial in half. With our data as an example, would we run the model once using the onsets of the stimulus and then run the model again using the ISI onsets, or run a model with both onsets? Does this mean “model as two stimulus classes” (i.e., 50 stimuli instead of 25 stimuli), or does this mean, “run the analyses two times, first with 25 stimuli sets that model the events themselves, then again with 25 regressors that model the anticipation period”?
I also noticed in the manual under the section about the linearity prinicple of the peripheral model that non-linearities may occur if the time interval between two SN firing bursts is veryshort (presumably what we are seeing) and that it is possible to model non-linearities in a linear model using Volterra kernels like in fMRI. But it also says this isn't done with SCR models. We were wondering why this is.
Thank you in advance!
Stephanie
Hi Stephanie
please apologise the late reply, we've been quite busy in the past week or so.
If you can identify pre-trial anticipatory SCR in your grand averages, it does make sense to model them. You would run the DCM once (not twice) but define event windows before a trial as well. That is, each "trial" would consists of some time window during the ISI and all the events during the actual trial. To not bias the inversion, I would suggest defining the time windows for responses during the ISI on a grand average of all trials, not split up by conditions. Note you need to take into account the peak latency of the SCR - i.e. if you are observing a peak at 5 seconds before the next trial, then the actual sympathetic burst should be modelled around 4 s earlier, to account for the delay of the sudomotor system. (In all PsPMs the timing is in terms of the "central" event of interest, not the ensuing body change.)
Volterra kernels is a nice idea but the implementation is not straightforward and nobody has actually done this yet. We have paper hopefully coming out soon in Psychophysiology that uses intraneural stimulation to better characterises the limits for when non-linearities occur, but could not get enough data to really model these non-linearities, so this remains a challenge.
Hope this helps
Dominik
Hi Dominik,
No worries at all! Thank you so much for the helpful reply! I was already thinking that the best way would be to set up the timing files with two events per trial.
One additional question, because the ISI and the next stimulus are contiguous and it appears based on our grand mean plots that the anticipatory peak is occurring one scond (or maybe even less) before the next stimulus onset, so the anticipatory peak and the SCR to the stimuli possibly overlap, do we need to do anything with the default inversion options? We have just been leaving them as is, but because of the potentitally overlapping nature of the responses, we are concerned that we might need to change those default options, but we aren't entirely sure if or how we would need to change them.
Thank you again in advance for your help!
Stephanie
Hi Stephanie
PsPM is set up to deal with overlapping SCR, as long as they are far enough apart to still be represented by a linear system. You just need to model the "central" (external/psychological/neural) input into the system.
Dominik