Brain Invaders bi2014a

Description

This cohort is the Brain Invaders bi2014a dataset — “Brain Invaders calibration-less P300-based BCI using dry EEG electrodes” — collected at GIPSA-lab, Grenoble, France, in 2014 and publicly released in 2019 (Korczowski et al., 2019). It contains scalp EEG from subjects playing Brain Invaders, a visual P300-based brain-computer interface (BCI) videogame built on the OpenViBE platform that uses the classic oddball paradigm: a grid of 36 symbols (1 Target, 35 Non-Target) is flashed pseudo-randomly, and the attended target flashes evoke a P300 event-related potential that an online Riemannian-geometry classifier decodes without a separate calibration phase. The recording emphasizes active dry electrodes (a then-novel headset technology) to test whether calibration-less P300 BCI remains viable with the lower signal quality of dry sensors. The released sample is dominated by naive BCI users (57 of the provided subjects had never used a BCI before), making it a useful general-population P300 cohort rather than a trained-operator set.

Citation

• Paper DOI: TODO(confirm)
• Dataset DOI: 10.5281/zenodo.3266223
• Reference: Korczowski, L., Ostaschenko, E., Andreev, A., Cattan, G., Rodrigues, P. L. C., Gautheret, V., & Congedo, M. (2019). Brain Invaders calibration-less P300-based BCI using dry EEG electrodes Dataset (bi2014a). GIPSA-lab research report, HAL hal-02171575.

Catalog citation block

Korczowski, L., Ostaschenko, E., Andreev, A., Cattan, G., Rodrigues, P. L. C., Gautheret, V., & Congedo, M. (2019). Brain Invaders calibration-less P300-based BCI using dry EEG electrodes Dataset (bi2014a) (Research Report). GIPSA-lab; IHMTEK. HAL: hal-02171575. Dataset DOI: 10.5281/zenodo.3266223. Released under CC-BY-4.0.

Download and access

Field	Value
Official source	https://doi.org/10.5281/zenodo.3266223
Access class	open
Size	TODO(confirm)
Format	mat, csv
BIDS	false
Mirrors	moabb.datasets.BI2014a; https://github.com/plcrodrigues/py.BI.EEG.2014a-GIPSA

Source & access

• Repository / DOI: Zenodo — 10.5281/zenodo.3266223 (record published 2019-07-02). Files are provided in both MATLAB .mat and .csv formats.
• Data descriptor: HAL technical report hal-02171575 (Korczowski et al., 2019) gives the full experimental description.
• Reference code: Python loader at github.com/plcrodrigues/py.BI.EEG.2014a-GIPSA; also exposed as moabb.datasets.BI2014a in the MOABB benchmark suite.
• License: Creative Commons Attribution 4.0 International (CC-BY-4.0).

Data structure

Each subject is one recording. The Zenodo release ships both .mat (MATLAB) and .csv per-subject files. In the CSV form, the continuous EEG sits alongside trial markers in dedicated columns — an Event column (flash/stimulus markers) and a Target column (1 = the flashed item was the attended target, 0 = non-target) — and the channel layout is read from an accompanying Header.csv rather than inferred from a fixed montage. The dataset is not distributed in BIDS layout and there is no datalad/git-annex packaging; it is a flat collection of per-subject files. The reference MOABB/py.BI.EEG.2014a-GIPSA loaders parse these into MNE Raw objects with the standard_1010 montage attached.

License

• Name: CC-BY-4.0
• Snippets on this site: allowed — the site may ship short derived snippets and precomputed products from this dataset, labelled as derivatives (re-referenced, filtered, cropped) with citation, license and DOI.

Acquisition

Field	Value
Device	g.tec g.USBamp with 16 g.Sahara active dry electrodes
Device class	research-dry
Channels	16
Sampling rate (Hz)	512
Online filters	none
Reference	right earlobe
Mains frequency (Hz)	50
Paradigms	p300-oddball-bci
Participants	64 (71 recorded; 64 released)
Population	64 healthy adults (of 71 recorded), 57 BCI-naive; demographics undocumented
Clinical groups	none
Sessions	3
Durations	up to three game sessions per subject (MOABB collapses them into one); ~198 target / ~990 non-target flashes per subject

Acquisition

EEG was recorded with a g.tec g.USBamp amplifier at a 512 Hz sampling rate (256 Hz Nyquist) using 16 active dry electrodes — gold-alloy, 8-pin dry sensors of the g.tec g.Sahara type, positioned per the international 10-10 (10/20) system. The recording reference was the right earlobe. No online digital filter was applied: the data are delivered as raw EEG with hardware event tagging (a USB digital-to-analog converter was used to synchronize stimulus markers with the EEG stream). Each subject completed up to three game sessions; the MOABB packaging collapses these into a single session per subject. The 16-channel montage, in file order, is: Fp1, Fp2, F5, AFz, F6, T7, Cz, T8, P7, P3, Pz, P4, P8, O1, Oz, O2.

Participants

The experiment recorded 71 subjects; usable EEG is provided for 64 of them (the seven dropped recordings are excluded from the public release, which is why the canonical subject count here is 64). Of the provided subjects, 57 were naive BCI users (no prior BCI experience). Per-subject demographics — exact age range/mean and sex breakdown — are not documented in the accessible Zenodo/MOABB metadata or the data-descriptor abstract, so treat the cohort as a mixed general-population adult sample of unspecified age/sex distribution. There is no clinical sub-group: all subjects are healthy volunteers.

Tasks / conditions

The single paradigm is the Brain Invaders P300 oddball game. On each repetition a 6×6 grid of 36 symbols is shown; one is the a-priori-known Target and the other 35 are Non-Targets, and groups of symbols flash pseudo-randomly while the subject mentally attends/counts the target flashes. The decoding problem is therefore a two-class ERP task (Target vs. Non-Target), with the target flash eliciting the P300. Each 1-second post-flash epoch is one trial; per subject there are roughly 990 Non-Target and 198 Target trials (about a 5:1 Non-Target-to-Target ratio, typical of oddball designs).

Used on this site

L3.2, L7.1, L7.5

Loader

Fetches only the stated subset. Recorded loader: moabb.datasets.BI2014a.

from moabb.datasets import BI2014a

ds = BI2014a()
data = ds.get_data(subjects=[1])  # one subject; dict[subject][session][run] -> mne.io.Raw

Caveats worth knowing

• Dry electrodes, no online filter, 50 Hz mains.
• 64 of 71 recorded subjects released.

Notable caveats

• Notch / power-line: no notch filter was applied at acquisition. Because recording was in Europe, expect a 50 Hz mains line and its harmonics.
• Dry-electrode signal quality: the whole point of bi2014a was dry-electrode feasibility, so signal-to-noise is lower and motion/contact artifacts are more likely than in the wet-electrode siblings (e.g. bi2013a uses wet Ag/AgCl). Expect more channel-level dropout and high-impedance artifacts.
• Subject count (71 vs 64): the descriptor reports 71 recorded subjects but the public release provides 64; do not expect to find subjects 65—71.
• Dataset family disambiguation: “Brain Invaders” denotes a series (bi2012, bi2013a, bi2014a/b, bi2015a/b). The 32-channel wet, multi-user, and flash-duration-modulation variants are different datasets with different N and montages — do not conflate their specs.
• Layout source: channel layout is read from Header.csv rather than inferred from a fixed montage.

Related datasets

• ERP CORE (Compendium of Open Resources and Experiments)
• EEG Motor Movement/Imagery Dataset (EEGMMIDB)
• LEMON (MPI-Leipzig Mind-Brain-Body)
• Dortmund Vital Study resting EEG
• SRM Resting-state EEG

# Loader recorded in the catalog: moabb.datasets.BI2014a
# MOABB fetches per-subject files on demand; restrict subjects to keep the download small.
import moabb  # noqa
# TODO(confirm) exact MOABB dataset class and subject subset