Douwe John Horsthuis 2022-12-20
The pop_clean_raw function to clean data in EEGLAB lets you use several settings that determin how much data will be deleted. This script is created so one can run it and find out how much data can be deleted without losing too much participants. Since it is a part of the EEG pipeline we will only focus on what the script does and why it might be useful to run it before you clean your data.
If you have questions or suggestions please reach out to douwehorsthuis@gmail.com
Disclaimer, do check out the EEGLAB tutorials etc. there might be reasons why data behaves as it they do that should be adressed instead of the cleaning functions
Pop_clean_raw is the suggested cleaning method for data by the EEGLAB creators. This function allows you to delete bad channels based on flatlines, noise, and correlation between channels. It also allows you to deleted bad portions of the data where many channels get noisy without deleting the channels. Lastly it is possible to reconstruct noisy data, so it becomes usable, this is not something that we use. **If you have experience with the Artifact Subspace Reconstruction (ASR) algorithm, and feel like this should be a part of this script somehow, please let us know either in the issues part or email me at douwehorsthuis@gmail.com
The script will either load the data, and pre-process it
(downsample/high and low pass filter/adding channel info), or if you
already have your data ready for the cleaning part, you can just load
that data. It then runs the pop_clean_rawdata function on a loop.
Every Loop the ChannelCriterion and the BurstCriterion get changed.
And stores how many channels and how much % data got deleted. After this
is done for all the participants you have a variable called quality.
This will be used to create figures that show how much the different
settings impacted the data.
Of course everything can be changed, but the main things that are easy
and quick are and maybe required:
subject_list This is where you store all the Unique ID numbers of your
participants.
home_path This is where you set the path to the main folder of the
data. All data is expected to be in individual subject folders.
Burst_criteria This is requires a start point a step size and end
point of what you want to use for the threshold for how much data get’s
deleted. The higher the number the more std the data is allowed to be
different, so high number equals less deleted data.
Channel_criteria This is requires a start point a step size and end
point of what you want to use for the threshold for how much the
channels can be correlated, the higher the number equals less channels
getting rejected.
Here are 3 examples, where for the first 2 we used a 1Hz filter and for
the last one we used a 0.1Hz filter.
We used a 1Hz highpass and a 45Hz lowpass filter and use relatively
strict burst criteria
We used a 1Hz highpass and a 45Hz lowpass filter and use a more liberal
burst criteria
We used a 0.1Hz highpass and a 45Hz lowpass filter and use a more
liberal burst criteria
It is clean, and probably expected that lowering the highpass filter means that more data will be deleted, because of this is you need different settings.
Next there are 3 more plot that show the same data in a different way.
We used a 1Hz highpass and a 45Hz lowpass filter and use relatively
strict burst criteria
We used a 1Hz highpass and a 45Hz lowpass filter and use a more liberal
burst criteria
We used a 0.1Hz highpass and a 45Hz lowpass filter and use a more
liberal burst criteria