Change PsPM data files to BIDS format

[dominikbach]

Feature Description

In light of future developments, data representation should become platform-independent.

Ideally, in BIDS format.

Since, BIDS operates on the data set level, one could create a 1-subject data set, and bundle this in a tar archive.

[gjheij]

I have played around with this for eye-tracking because I wanted to run a modeling step on the preprocessed data for which I needed BIDS input (see https://gitlab.com/bachlab/methods/PupilBench/-/blob/main/matlab/benchmark_core.m?ref_type=heads#L656). That looks something like this:

PupilFear/Corpus/derivatives/pspm/
├── dataset_description.json
├── participants.json
├── participants.tsv
├── sub-001
│   └── ses-01
│       └── beh
│           ├── sub-001_ses-01_task-fearconditioning_desc-preproc_events.json
│           ├── sub-001_ses-01_task-fearconditioning_desc-preproc_events.tsv
│           ├── sub-001_ses-01_task-fearconditioning_desc-preproc_physio.json
│           └── sub-001_ses-01_task-fearconditioning_desc-preproc_physio.tsv.gz

Where _physio.tsv.gz contains:

In [4]: df = read_physio_tsv_headerless("/mnt/z/PupilFear/Corpus/derivatives/pspm/sub-001/ses-01/beh/sub-001_ses-01_task-fearconditioning_desc-preproc_physio.tsv.gz")

In [5]: df
Out[5]:
        timestamp  pupil_size
0           0.000    5.256403
1           0.001    5.256498
2           0.002    5.256594
3           0.003    5.256690
4           0.004    5.256787
...           ...         ...
592097    592.097    3.339109
592098    592.098    3.339109
592099    592.099    3.339110

and _events.tsv:

In [7]: ev = pd.read_csv("/mnt/z/PupilFear/Corpus/derivatives/pspm/sub-001/ses-01/beh/sub-001_ses-01_task-fearconditioning_desc-preproc_events.tsv", sep="\t")

In [8]: ev
Out[8]:
      onset  duration trial_type  value
0     5.000         0       CSpr      1
1     8.500         0        USp      2
2    16.054         0       CSpu      3
3    30.088         0        CSm      4
4    33.588         0        USm      5

If all channels are the same shape after processing, the could all go in the compressed tsv file. To maintain some level of BIDS, I would probably favor splitting sessions, tasks, and runs. Modalities can be combined if data shape/sampling rate are equal, otherwise these may have to be separated too.

Ideally, this should then also be compatible with the BIDS importer. This may require tinkering with how the channels are read in.

[dominikbach]

See recent BEPs for some data modalities.

EMG
bids-standard/bids-specification#1371