Skip to content

Importing recordings

braino import turns recordings in any supported format into a BIDS dataset. The pipeline is the same for every format: detect → read → normalise → de-identify → write BIDS.

# One recording. The key file links subject labels to real IDs: keep it private, outside the
# dataset (BrainO refuses to put it inside).
braino import raw/patient_a.edf --subject-id "patient A" --task rest \
  --dataset study_bids --key private/key.json

# Many, from a CSV with columns source,subject_id,task[,session,run,line_freq]
braino import --mapping recordings.csv --dataset study_bids --key private/key.json

Formats

EDF/EDF+, BDF, BrainVision, FIF, EEGLAB, Neuroscan CNT, EGI MFF, GDF, Curry, Nihon Kohden and Persyst, all read through MNE-Python. XDF (Lab Streaming Layer) is planned.

Normalisation

Every format names channels and stores events differently, so import makes them alike and reports every change:

  • standard electrode names get their standard spelling (FP1 → Fp1, EEG Fz-REF → Fz);
  • channels typed EEG whose names say otherwise become EOG, ECG, EMG or miscellaneous;
  • electrode positions come from the file, or from a standard layout when the names identify one (and the BIDS files then say they are approximate);
  • events in trigger channels become annotations.

De-identification

Names, birth dates and recording dates are removed from file headers; dates are shifted per subject into 1902-1924, keeping age, time of day and intervals; identifiers, dates, e-mail addresses and phone numbers are removed from event text, and any other free text is listed for you to check. Data are always written new, never copied, because an original header still holds the personal data. The provenance record keeps only hashes of the original files, not their names.

Checking the result

braino inspect --dataset study_bids    # subjects, tasks, channels and events per recording
braino qc --dataset study_bids         # flat and noisy channels, line noise, artifacts

The BIDS files pass the official validator (tested in CI with only "recommended metadata" warnings).

Tip

Datasets that are already in BIDS (like ERP CORE) need no import: point the analysis tools at them directly.