[Eeglablist] Help with "missing" event markers in BDF files?
A.Pickering at gold.ac.uk
Tue Feb 2 11:44:37 PST 2021
We have a frustrating problem with event markers in BDF files recorded on a Biosemi system by our collaborators in Australia. Unfortunately, it will take a certain amount of explanation in this email.
The Australians have previously used Brain Vision Analyser to analyse these files completely successfully, and we have published some of the findings. However, the Aussie lab is shut and they don't currently have remote access to BVA or the data files. So, they are limited in their ability to help now.
In the UK my RA, Tom, and I are trying a further analysis on the Australian data files using EEGlab. I provide a link for an example Aussie BDF file, 1.BDF. There is the same issue with all 100 datafiles we have.
The example Aussie file can be downloaded from this folder (225 MB)
In this paradigm (which was programmed using a task software system called Eprime) there are 360 trials with two key critical events per trial, denoted S1 and S2. The trials types (1 to 4) are in the ratio: 144 144 36 36. On a small proportion of trials the event markers did not record (for unknown reasons) and so, in this file for example, we have 333 trials with markers in the file. However, on every trial there are 2 markers and the events are coded 1 to 4 to reflect the above trial type; the same number code was used for the onset of both S1 and S2 within the same trial. We know that 2 event markers per trial can be seen using Brain Vision Analyser -- at least for those trials on which event markers were recorded (although, as noted above, my Australian colleagues cannot look at them again right now).
When we now import these files into EEGlab we get only 1 marker event per trial, in the right proportions but with a small amount of missing data as explained.
For this Australian example file we get the following event codes and frequencies:-
Annoyingly, we cannot tell whether the single event which EEGlab extracts per trial marks the onset of S1 or S2 (which are about 850 msec apart; there is a randomly varying ITI). We have published the results for the ERPs to S2 and now want to analyse the same ERPs to S1. We can find no easy way to determine for certain which of the two events EEG lab is extracting for each trial from the file (we assume, but do not know, that it is systematically extracting the same event [S1 or S2] on each trial).
We ran the same paradigm in the UK several years before moving data collection to Australia. We used the same setup (eprime task, Biosemi BDF file formats) and an example BDF file from that study, 1250162.BDF, can also be found at the above link. It is 348 MB
The Uk version of the task was trivially different in that there were 30 practice trials first and then 480 main trials (in the ratio 192, 192, 48, 48). With these older UK BDF files, EEG lab works perfectly, and extracts all the event markers (2 per trial in pairs 850 msec part, with the same code). We can analyse these data for the ERPs to S1. Also, we can extract event markers for every trial in the UK data. We wish that the Aussie files behaved as well with EEGlab!
The markers recovered by EEGlab from this example UK (old) file are as follows (numeric code, frequency and description):
61539 1 (start of recording marker)
61441 384 (192 trials x 2 per trial, for trial type 1)
61442 384 (192 trials x 2, for trial type 2)
61443 96 (48 trials x 2, for trial type 3)
61444 96 (48 trials x 2, for trial type 4)
61541 24 (12 practice trials x 2, of type 1)
61542 24 (12 practice trials x 2, of type 2)
61543 6 (12 practice trials x 2, of type 3)
61544 6 (12 practice trials x 2, of type 4)
Any help with what is going on with the Australian file would be HUGELY appreciated. The long-winded way to do this would be to look at the Aussie example file in another EEG analysis system and assuming that it, like BVA, can see both events per trial, then document the latencies of S1 and S2 event markers for the 333 trials where event markers occurred. IF we had this info, we could look at the latencies extracted by EEG lab and work out which event (S1 or S2) was being extracted by EEGlab on each trial. If it was always the same event (S1 or S2) for each trial, then we could analyse all 100 files with confidence. We are unable to do this ourselves at the moment. Maybe there is another, simpler way?
If anyone can tell us what is going on, and how to work around it, we would be eternally grateful. 🙂
Our set up is using the latest version of EEGlab (v2019.1) in Matlab 2019 and we get the same result whether we analyse on Windows 10 or mac systems.
Alan and Tom
Professor Alan Pickering, BA PhD
Try out my YouTube Channel (mostly statistics and data analysis)
Department of Psychology
Goldsmiths, University of London
email: a.pickering at gold.ac.uk
alternative email: alan.pickering at cognivaterehab.com
More information about the eeglablist