[FieldTrip] Problems with TMS-EEG dataset

Tue Feb 27 10:11:28 CET 2024

Dear Stefan,

It looks as if you got quite far already, but indeed the residual of the TMS artifact is still quite large. I must say that I don’t quite understand what you did in your step (2), but I assume that this was dealing “only” with the very early stuff after the TMS pulse. Clearly - in your data - quite a few electrodes have longer lasting effects (that actually last up until the onset of the next pulse), that might not be of neural origin. A fundamental problem in your experimental design is the fact that the TMS pulses are synchronous with the events of interest (i.e. the picture onset). Any heuristic that uses temporal information about the TMS-pulses runs the risk of throwing out the baby with the bathwater. Yet, one thing you might try (as an alternative to your step (4)) is an approach that is documented here: https://www.fieldtriptoolbox.org/example/use_denoising_source_separation_dss_to_remove_ecg_artifacts/<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.fieldtriptoolbox.org%2Fexample%2Fuse_denoising_source_separation_dss_to_remove_ecg_artifacts%2F&data=05%7C02%7Cfieldtrip%40science.ru.nl%7C21e81d7cfa2f4eca224408dc377414ce%7C084578d9400d4a5aa7c7e76ca47af400%7C1%7C0%7C638446218901934392%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=Wfl5QKXpB%2BNcs%2B7F5n%2B69UFJCPQRD42pWrypr7pLv7A%3D&reserved=0>

The idea would be to use the onsets of the TMS-pulses to inform the dss-algorithm (rather than the locations of the ECG’s QRS peaks) for the unmixing estimation. Specifically, you would execute the ‘DSS component rejection’ step (after adjusting the cfg.dss.wdim to be at least smaller than your number of electrodes, and after specificying ‘params.artifact’ in a way that the code can interpret it. Perhaps you would need to run the example code step by step (using the associated data) with the matlab debugger switched on to see how you can adjust the code to your needs.

No guarantee, but possibly the dss-algorithm - with a careful selection of components to be rejected - might be able to strike a good balance between reducing the artifact, and keeping the relevant signal in the data.
Obviously, it is the hope that - once you start looking at difference waves - any residual artifact (which is supposed to be condition independent) subtracts out.

If all of this doesn’t work well, the only thing to do would be to discard electrodes that are close to the stimulation site, at the risk of them actually being the interesting ones.

Good luck and keep up the good work,
Jan-Mathijs

On 22 Feb 2024, at 17:12, Stefan van Haren via fieldtrip <fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>> wrote:

Dear community,
My name is Stefan and I am finishing my master thesis at the University Medical Centre Groningen on a study with EEG-TMS data during language tests. Here the subject names a picture on a screen while we administer a train of 5 TMS pulses with a frequency of 5Hz. The TMS pulses are administered at t=0ms and the picture is also shown at t=0ms. This means that the final TMS pulse is administered at t=800ms. The biggest challenge remaining is the fact that I want to analyse the ERP (mainly the P300 component) which is located between the second (200ms) and third (400 ms) TMS pulse. This means that the data in between the TMS artefact must have a sufficient quality. I have compiled a processing pipeline to process the data;

  1.  Divide into trails.
  2.  Remove TMS artefact (by removal and interpolation)
  3.  Filter the data (40 Hz low pass, 1Hz high pass)
  4.  Remove any bad channels and trials.
  5.  ICA

I can provide the script that I use but it is quite large and might not be needed for my question. When following this pipeline, the data is processed quite nicely, but still some ‘problems’ remain, See the plot below where the first image is the averaged raw data and the second image is the averaged filtered data.
<image.png><image.png>

As you can see, there are still problems with a lot of the channels, for example, the most negative orange channel, which is visible in both plots. This plot seems to tend to rise but seems to be ‘reset’ after each TMS pulse (at 0.20, 0.40, 0.60, 0.80 sec). This decrease is caused by the interpolation of the TMS artefact. Does anyone have any idea what this might be, and how I can solve/prevent this problem? It is present in a lot of channels and at this point, I cannot reliably calculate an ERP. I also noticed that this effect is mostly present in the channels that are close to the TMS stimulation sites. Any help would be appreciated greatly.
Best,
Stefan
_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fmailman.science.ru.nl%2Fmailman%2Flistinfo%2Ffieldtrip&data=05%7C02%7Cfieldtrip%40science.ru.nl%7C21e81d7cfa2f4eca224408dc377414ce%7C084578d9400d4a5aa7c7e76ca47af400%7C1%7C0%7C638446218902090655%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=2CTYAD1Ohxf5L8gc0T2oictLnNBhpUmYHRaz71eKkdk%3D&reserved=0>
https://doi.org/10.1371/journal.pcbi.1002202

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20240227/3997992e/attachment.htm>