1
$\begingroup$

I have a dataset in which we expect to evoke P300 peaks in an oddball setting. My part of the project considers building a machine learning model to classify whether a participant saw a relevant stimulus (eliciting a P300) or a distractor. The model works, so that's good. However, I wanted to include a figure showcasing the data for one subject.

  • Is it OK to display the evoked response (so averaged across all trials), but also averaged across all electrodes? This would mean I have essentially only two curves on the plot, one for "target" and one for "distractor".
  • Another option is to only use the occipital electrodes to make the plot, as we expect the subject to have visual activation.
$\endgroup$
3
  • $\begingroup$ Welcome. So I understand it is solely for informative purposes? The model apparently used single-epoch data from many scalp electrodes at once? In that case I reckon it doesn't matter too much what you show in your sample data; whatever looks best. $\endgroup$
    – AliceD
    Commented Sep 21, 2023 at 6:49
  • $\begingroup$ Indeed, it is solely for informative purposes. The model uses all electrodes at once as you said. Thanks! $\endgroup$
    – Smamorti7
    Commented Sep 22, 2023 at 7:02
  • 1
    $\begingroup$ In addition to the good suggestions by AliceD, you might also check how well your procedure would perform if you first averaged across channels, the way you're planning to display the data. :) $\endgroup$
    – Bryan Krause
    Commented Sep 22, 2023 at 13:32

1 Answer 1

1
$\begingroup$

From what I understand from the Q and comments is the following scenario:

  • The graph is solely for informative purposes only;
  • The model uses single-epoch data from many scalp electrodes at once;
  • Single-epoch waveforms are to noisy too be of any value in a graph;

Thus, the data in your recordings that were fed in your AI don't lend themselves for visualization, because of the low signal-to-noise ratio. Hence you need averaging of epochs to make sense out of it for us human readers. In this situation I reckon it doesn't matter too much what you show in your sample data; whatever looks best. Assuming you'll bunch up multiple epochs and multiple electrodes from one or a few locations, make sure to:

  • Check whether your ensemble-averaged waveform shows a feasible P300 representative for the P300 values that are being fed into the model;
  • Mention which electrodes were averaged, with how many epochs and under what conditions so as to make sure the reader understands where the waveforms come from.
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.