| --- |
| license: cc0-1.0 |
| tags: |
| - eeg |
| - sleep-staging |
| - in-ear-eeg |
| - scalp-eeg |
| - hdf5 |
| --- |
| |
| # EESM23-Processed |
|
|
| Processed HDF5 export of the [EESM23 OpenNeuro dataset](https://doi.org/10.18112/openneuro.ds005178.v1.0.0). It contains paired in-ear EEG and scalp EEG sleep-staging samples from the scored `ses-001` and `ses-002` recordings of 10 subjects. |
|
|
| > **Update:** this export now stores one **30-second epoch per sample** (previously seven 4-second windows). This matches [`EESM19-Processed`](https://huggingface.co/datasets/Zachary1150/EESM19-Processed) and the canonical sleep-staging unit. See "Preprocessing" below. |
|
|
| ## Preprocessing |
|
|
| Generated with `Ear-EEG-FM-Benchmark/dataset/preprocess_eesm23.py` using schema/eegfm version `0.5.0`: |
|
|
| - 0.1–100 Hz band-pass and 50 Hz notch filtering on each continuous recording |
| - no re-referencing, resampling, or channel renaming |
| - **each labeled 30-second AASM scoring event is stored as one 30-second window** (one sample = one epoch = one label = one prediction — the canonical sleep-staging unit) |
| - classes: `Wake`, `N1`, `N2`, `N3`, `REM`; `Artefact` events are dropped |
| - real sensor/data-loss NaN/Inf samples are preserved and recorded in overall and per-channel quality fields |
| - the ear-EEG and scalp outputs are strictly row-aligned; a session is excluded from both when either modality is missing/unreadable, and a sample is retained only when it exists in both modalities |
| - recording bounds are checked per 30-second window (the full epoch must lie inside the recording) |
| - all signal values are stored as `float32` microvolts at 250 Hz (7,500 samples per window) |
|
|
| Brief device data-loss gaps are interpolated before filtering to prevent FIR-kernel contamination, after which the original NaN positions are restored before window selection. |
|
|
| ### Why 30-second windows (and how to get 4-second windows) |
|
|
| The 30-second epoch is the unit at which sleep is scored (AASM) and evaluated. Every |
| reference EEG foundation model that does sleep (BENDR, EEGPT, CBraMod, REVE, and the |
| EEGPT comparison implementations of LaBraM/BIOT) ingests the full 30-second epoch |
| downstream and emits one stage prediction per epoch, rather than splitting one label |
| into independently-scored short windows. Storing 30 s is also a strict superset: the |
| benchmark loader (`dataset/loader.py`) accepts an `epoch_sec` argument and crops a |
| shorter window from each stored sample at load time, so a 4-second (or any ≤30 s) view |
| is available without re-exporting; the reverse is not possible. |
|
|
| ## Files |
|
|
| | File | Channels | Shape `(N, C, T)` | Size | |
| |---|---|---:|---:| |
| | `eesm23-in-ear-eeg.h5` | RB, RT, LB, LT | `(15,523, 4, 7500)` | 1.74 GiB | |
| | `eesm23-scalp-eeg.h5` | M1, F3, C3, O1, M2, F4, C4, O2 | `(15,523, 8, 7500)` | 3.48 GiB | |
|
|
| `T = 7500` is one 30-second epoch at 250 Hz. `N` is the number of scored epochs, one sample per 30-second AASM epoch. |
|
|
| Label distribution (identical for both modalities — the samples are row-aligned): |
|
|
| | Wake | N1 | N2 | N3 | REM | |
| |---:|---:|---:|---:|---:| |
| | 1,375 | 1,292 | 7,328 | 2,494 | 3,034 | |
|
|
| ## Retained and discarded epochs |
|
|
| The source scoring files label 17,074 30-second epochs across the 20 sessions (10 |
| subjects × `ses-001`/`ses-002`): 16,558 carry one of the five retained sleep-stage |
| labels and 516 are labeled `Artefact`. Each retained epoch becomes one 30-second |
| sample, so the final files contain **15,523** strictly paired samples per modality. |
| The exclusions from the 16,558 five-class candidates are: |
|
|
| | Reason | Epochs | |
| |---|---:| |
| | Entire `sub-006/ses-002` pair: truncated/unreadable PSG | 1,031 | |
| | 30-second epoch out of bounds or with no cross-modality partner | 4 | |
| | **Total excluded** | **1,035** | |
|
|
| (The 516 `Artefact` epochs are outside the five-class task and are never candidates.) |
|
|
| `sub-006/ses-002` is dropped because its scalp PSG file is truncated/unreadable |
| (`OSError: could not read bytes`). No sample is excluded because it contains NaN/Inf: |
| the files retain 2,230 in-ear and 1,913 scalp samples with at least one non-finite |
| value; their indices remain paired even when quality differs between modalities. |
|
|
| ## HDF5 schema (v0.5) |
|
|
| ```text |
| /data (N, C, 7500) float32 |
| /durations (N,) int64 |
| /nan_fraction (N,) float32 |
| /channel_nan_fraction (N, C) float32 |
| /labels (N,) int64 |
| /sample_id (N,) int64 |
| /subject (N,) string |
| /session (N,) string |
| /task (N,) string |
| /acquisition (N,) string |
| /run (N,) string |
| /recording_id (N,) string |
| /trial_id (N,) int64 |
| /event_id (N,) int64 |
| /split_group_id (N,) int64 |
| /window_start_sample (N,) int64 |
| /window_stop_sample (N,) int64 |
| /ch_names (C,) string |
| ``` |
|
|
| `event_id` and `split_group_id` both identify the source 30-second scoring row; here each sample *is* one epoch, so there is one sample per split group. For sequence-model sleep staging that needs adjacent-epoch context, group by `subject`+`session` and order by `window_start_sample`. |
|
|
| Important attributes include `sfreq`, `class_names`, `unit`, `eegfm_version`, `preprocess_config_json`, `split_group_kind`, and `window_reference`. |
|
|
