Brain Invaders 2013a

P300 dataset bi2013a from a “Brain Invaders” experiment (2013) carried-out at University of Grenoble Alpes.

Simple Python scripts for working with the dataset are available at https://github.com/plcrodrigues/BrainInvaders-2013a-Dataset

Dataset Description

This dataset concerns an experiment carried out at GIPSA-lab (University of Grenoble Alpes, CNRS, Grenoble-INP) in 2013.

Principal Investigators: Erwan Vaineau, Dr. Alexandre Barachant
Scientific Supervisor :  Dr. Marco Congedo
Technical Supervisor : Anton Andreev

The experiment uses the Brain Invaders P300-based Brain-Computer Interface [7], which uses the Open-ViBE platform for on-line EEG data acquisition and processing [1, 9]. For classification purposes the Brain Invaders implements on-line Riemannian MDM classifiers [2, 3, 4, 6]. This experiment features both a training-test (classical) mode of operation and a calibration-less mode of operation [4, 5, 6]. 

The recordings concerned 24 subjects in total. Subjects 1 to 7 participated to eight sessions, run in different days, subject 8 to 24 participated to one session. Each session consisted in two runs, one in a Non-Adaptive (classical) and one in an Adaptive (calibration-less) mode of operation. The order of the runs was randomized for each session. In both runs there was a Training (calibration) phase and an Online phase, always passed in this order. In the non-Adaptive run the data from the Training phase was used for classifying the trials on the Online phase using the training-test version of the MDM algorithm [3, 4]. In the Adaptive run, the data from the training phase was not used at all, instead the classifier was initialized with generic class geometric means and continuously adapted to the incoming data using the Riemannian method explained in [4]. Subjects were completely blind to the mode of operation and the two runs appeared to them identical.

In the Brain Invaders P300 paradigm, a repetition is composed of 12 flashes, of which 2 include the Target symbol (Target flashes) and 10 do not (non-Target flash). Please see [7] for a description of the paradigm. For this experiment, in the Training phases the number of flashes is fixed (80 Target flashes and 400 non-Target flashes). In the Online phases the number of Target and non-Target still are in a ratio 1/5, however their number is variable because the Brain Invaders works with a fixed number of game levels, however the number of repetitions needed to destroy the target (hence to proceed to the next level) depends on the user’s performance [4, 5]. In any case, since the classes are unbalanced, an appropriate score must be used for quantifying the performance of classification methods (e.g., balanced accuracy, AUC methods, etc).

This database has been used in the development of the common spatio-temporal pattern method for estimating ERPs [8].

Data were acquired with a Nexus (TMSi, The Netherlands) EEG amplifier:

• Sampling frequency: 512 samples per second
• Digital Filter: No
• Electrodes:  16 wet Silver/Silver Chloride electrodes positioned at FP1, FP2, F5, AFz, F6, T7, Cz, T8, P7, P3, Pz, P4, P8, O1, Oz, O2 according to the 10/20 international system
• Reference: left ear-lobe
• Ground: N/A

The data for Subject X is available as a subjectX.zip file. In it, there is a folder for each Session that the Subject performed. In each Session's folder, there are four .gdf files, one for each Run performed in the Session. The names of the files and the conditions associated to them are available below and also inside the meta.yml file in the .zip file.

  - filename: '1.gdf'
    experimental_condition: adaptive
    type: training
  - filename: '2.gdf'
    experimental_condition: adaptive
    type: online
  - filename: '3.gdf'
    experimental_condition: nonadaptive
    type: training
  - filename: '4.gdf'
    experimental_condition: nonadaptive
    type: online   

References

[1] Arrouët C, Congedo M, Marvie J-E, Lamarche F, Lècuyer A, Arnaldi B (2005) Open-ViBE: a 3D Platform for Real-Time Neuroscience. Journal of Neurotherapy, 9(1), 3-25. people.rennes.inria.fr/Anatole.Lecuyer/Open-ViBE.pdf)

[2] Barachant A, Bonnet S, Congedo M, Jutten C (2013) Classification of covariance matrices using a Riemannian-based kernel for BCI applications. Neurocomputing 112, 172-178. (hal.archives-ouvertes.fr/hal-00820475/document)

[3] Barachant A, Bonnet S, Congedo M, Jutten C (2012) Multi-Class Brain Computer Interface Classification by Riemannian Geometry. IEEE Transactions on Biomedical Engineering 59(4), 920-928. (hal.archives-ouvertes.fr/hal-00681328/document)

[4] Barachant A, Congedo M (2014) A Plug & Play P300 BCI using Information Geometry, arXiv:1409.0107. (https://arxiv.org/pdf/1409.0107.pdf)

[5] Congedo M, Barachant A, Andreev A (2013) A New Generation of Brain-Computer Interface Based on Riemannian Geometry. arXiv:1310.8115 (arxiv.org/ftp/arxiv/papers/1310/1310.8115.pdf)

[6] Congedo M, Barachant A, Bhatia R (2017) Riemannian Geometry for EEG-based Brain-Computer Interfaces; a Primer and a Review. Brain-Computer Interfaces, 4(3), 155-174. (hal.archives-ouvertes.fr/hal-01570120/document)

[7] Congedo M, Goyat M, Tarrin N, Ionescu G, Rivet B,Varnet L, Rivet B, Phlypo R, Jrad N, Acquadro M, Jutten C (2011) “Brain Invaders”: a prototype of an open-source P300-based video game working with the OpenViBE platform. Proc. IBCI Conf., Graz, Austria, 280-283. (hal.archives-ouvertes.fr/hal-00641412/document)

[8] Congedo M, Korczowski L, Delorme A, Lopes da Silva F. (2016) Qpatio-temporal common pattern: A companion method for ERP analysis in the time domain. Journal of Neuroscience Methods, 267, 74-88. (hal.archives-ouvertes.fr/hal-01343026/document)

[9] Renard Y, Lotte F, Gibert G, Congedo M, Maby E, Delannoy V, Bertrand O, Lécuyer A (2010) OpenViBE: An Open-Source Software Platform to Design, Test and Use Brain-Computer Interfaces in Real and Virtual Environments. PRESENCE : Teleoperators and Virtual Environments 19(1), 35-53. (hal.archives-ouvertes.fr/hal-00477153/document)