Hello everyone!
This is the Yinaoyun scientific research circle, I am sister Miaojun~
When we conduct EEG analysis, the first thing we need to do is a big preprocessing action. What is the role of preprocessing in EEG analysis? Let's see what the big cows have to say.
What role does preprocessing play in EEG data analysis?
After sorting out the previous research, it is found that the researchers mainly focus on the following aspects:
1. Is preprocessing a sufficient and necessary (sufficient) condition for EEG data analysis?
2. Is preprocessing a sufficient but unnecessary condition for EEG data analysis?
3. Is preprocessing a necessary but not sufficient condition for EEG data analysis?
4. Is preprocessing an insufficient and unnecessary condition for EEG data analysis?
Recently, Arnaud Delorme (Note: the developer of EEGLAB) submitted the article "EEG is better left alone" on the bioRxiv preprint platform.
The article access link is as follows:
https://doi.org/10.1101/2022.12.03.518987
This article conducts a cross-data analysis platform (EEGLAB, FieldTrip, MNE, and Brainstorm) on the influence of some methods adopted in the preprocessing link of EEG data analysis (such as: interpolation bad guide, re-referencing, ICA, etc.) on EEG results. fully discussed.
In short, the results of this preprint show that, except for high-pass filtering and interpolation of bad leads, automatic data correction has no effect or significantly reduces the percentage of important electrode points.
Re-referencing and baseline correction methods may have a negative impact on the stability of the data results.
Rejecting bad data segments or trials does not compensate for the loss of statistical power.
ICA analysis rejects oculograph artifacts, EMG artifacts and does not reliably improve data performance.
In addition, when data analysis is performed across the pipeline of analysis platforms, better expected results have not been obtained.
Let us understand the detailed analysis process together!
01
high pass filter
For the Go/No-go dataset and the Oddball dataset, filtering resulted in an increase of about 50% in the percentage of electrode points, as shown in the figure below. For the Face Dataset, filtering did not bring significant improvement, which may be due to the use of 0.1Hz high-pass filtering during data collection.
02
Remove linear noise
Bad lead interpolation may affect the percentage of electrode points, notch filtering has no significant effect on the percentage of data. But the interpolation algorithms in cleanline and Zapline-plus even have a negative impact on the performance of Face Dataset.
Furthermore, offline removal of linear noise may not be a critical step in preprocessing EEG data, except for some specific EEG data analysis metrics or methods.
03
heavy reference
As shown, we found that re-referencing did not increase the percentage of important channels in all three datasets. Also, re-referencing methods such as median, mean, REST, and PREP may cause some negative effects.
04
baseline
As shown, subtracting baseline activity had no or negative impact on data quality, especially when the baseline was shorter than 500 ms. Additionally, subtracting mean baseline activity should be omitted in ERP analysis if data are high-pass filtered at 0.5 Hz or above.
05Analysis
platform pipeline
In the analysis pipeline, no analysis pipeline has been found to have a clear advantage.
06
postscript
The results and viewpoints of this preprint have sparked intense discussions among scholars. Driven by open scientific research, the open sharing of massive EEG datasets, data analysis links and corresponding analysis materials are quietly changing the current The EEG data analysis process also allows us to think about what kind of EEG data processing and analysis is a more scientific and objective analysis that can make the results robust.
In addition, we also need to take a cautious look at the results and viewpoints in this preprint, and in the follow-up EEG research, especially in the data collection and processing, more detailed records and report related parameter settings, so as to better in the Stepping forward in the tide of open science.
The above is the main content of this sharing, see you next time!
Reference:
Delorme, A. (2022). EEG is better left alone. bioRxiv.
https://doi.org/10.1101/2022.12.03.518987
Reprint | Wandering Heart Ball
Author | Nian Jingqing
Typesetting | Right Right
Proofreading |