mne-bids: Possible error when anonymizing EDF files

Describe the bug

Using write_raw_bids, I converted an EDF file using v0.6 while anonymizing it. However, I tried opening the same file in EDFBrowser, and I got this error:

Error, file is marked as EDF+ but startdate field does not match with startdate in
recordfield:
"12.02.85" <-> "12-FEB-2012".
 File is not a valid EDF or BDF file.

This suggests that the anonymization didn’t work. In addition, it doesn’t seem to be a valid file in the context of edfbrowser suggesting that further errors might occur in the future.

Steps to reproduce

Convert dataset attached here: https://www.dropbox.com/sh/w0wd0icoo9bib2j/AAB0MVgibmrTUeSRPNGmeIaya?dl=0

Expected results

The file after doing anonymization in write_raw_bids should still be accessible by EDFBrowser.

Additional information

A related issue is just telling write_raw_bids to convert said file to BrainVision. #659

About this issue

Original URL
State: closed
Created 4 years ago
Comments: 15 (15 by maintainers)

Most upvoted comments

sappelhoff on Dec 29, 2020

The 1925 date requirement keeps coming up all the time. It’s a problem in fif, now edf. But HIPPA doesn’t require it. Folks should really reconsider this …

jasmainak on Dec 29, 2020

@sappelhoff Don’t know much about the BrainVision format’s internals, but the main points in favour of EDF+/BDF+ I can think of are:

Single file per recording
Broad software compatibility (well, maybe not so much…)
Some potentially useful metadata in the header that might get lost with a format conversion
Avoids extra processing/conversion of data when not explicitly necessary: IIRC BrainVision stores values in float32, whereas EDF and BDF store values as int16 and int24 with conversion factors (respectively). That said, MNE converts all EDF/BDF data to float32 upon import anyway, so it might not make any difference.
Discontinuous recordings and text annotations? Not sure what BrainVision supports here or how the conversion process works, but if it doesn’t support those, that would be a potentially unexpected loss of data

That’s the best case I can think of for anonymizing them without a format conversion, but I also see the benefits of always converting to BrainVision. As much effort as I put into the EDF anonymization PR, I’m perfectly happy to see it dropped by the wayside if it’s not the best possible solution for end-users!

EDIT: I think if we can find any other big EDF utilities that do the same date-matching check as EDFBrowser, that would negate any ‘broad compatibility’ benefit that the format has, and clearly make BrainVision conversion a better choice. If it’s just an EDFBrowser quirk, I think it would be easiest to file an issue or PR there and remove the sanity check.

a-hurst on Dec 24, 2020

but then set a random date < 1920 in the scans.tsv?

Shouldn’t be random, but shifted by a user-specified duration 😃 But otherwise yes, why not. This is once again a deviation from BIDS, as the file content deviates from the sidecar metadata, but for me and a few others this seems to be okay 😉 we should codify this though, i.e. amend BIDS asap…

hoechenberger on Dec 27, 2020

in brief: It’s not possible to reasonably anonymize EDF/EDF+/BDF files to a year prior to 1985. 😦 I am starting to be in favor of going back to auto-conversion to BrainVision if anonymization to prior 1985 is wanted. And I am starting to develop a severe dislike for EDF … Austin, can you tell me one nice thing about EDF which makes it good in my eyes again? 😃

sappelhoff on Dec 24, 2020