Data set IVb ‹motor imagery, uncued classifier application›

Data set provided by Fraunhofer FIRST, Intelligent Data Analysis Group (Klaus-Robert Müller, Benjamin Blankertz), and Campus Benjamin Franklin of the Charité - University Medicine Berlin, Department of Neurology, Neurophysics Group (Gabriel Curio)

Correspondence to Benjamin Blankertz ⟨⟩

The Thrill

Most demonstrations of algorithms on BCI data are just evaluating classification of EEG trials, i.e., windowed EEG signals for fixed length, where each trial corresponds to a specific mental state. But in BCI applications with continuous feedback one is faced with the problem that the classifier has to be applied continuously to the incoming EEG without having cues of when the subject is switching her/his intention. This data set poses the challenge of applying a classifier to continuous EEG for which no cue information is given.
Another issue that is addressed in this data set is that the test data contains periods in which the user has no control intention. During those intervals the classifier is supposed to return 0 (no affiliation to one of the target classes). Our experience is that is does not help to include a relax class in the training measurement, because the mental state of relax during an initial recording (without feedback) is substatially different from the mental state of having no control intention during feedback. During feedback the user is in a rather active state, and despite of having no actual control intention the user may very well have strong considerations concerning her/his feedback application. When one can design a classifier that positively recognizes the two learned classes (here left and foot motor imagery) it should return zero for other mental states as relax or periods of absense of control intention.

Experimental Setup

This data set was recorded from one healthy subject. He sat in a comfortable chair with arms resting on armrests. This data set contains only data from the 7 initial sessions without feedback. The first 3 sessions are given with labels as training set. Visual cues (letter presentation) indicated for 3.5 seconds which of the following 3 motor imageries the subject should perform: (L) left hand, (F) right foot, (Z) tongue (=Zunge in german). The presentation of target cues were intermitted by periods of random length, 1.75 to 2.25 seconds, in which the subject could relax. Continous EEG signals of sessions 4 to 7 are given without any cue information (neither target class nor timing) as test set. In these sessions the target classes left, foot and relax were ordered by acoustic stimuli for periods for varying length between 1.5 and 8 seconds. Intermitting periods were given as above.

Format of the Data

Given are continuous signals of 118 EEG channels and, for the training data, markers that indicate the time points of 210 cues and the corresponding target classes. Only cues for the classes left and foot are provided for the competition (since tongue imagery was not performed in the test sessions).

Data are provided in Matlab format (*.mat) containing variables:

As alternative, data is also provided in zipped ASC II format:

Requirements and Evaluation

Please provide an ASC II file (named 'result_IVb.txt') containing classifier outputs (real number between -1 and 1) for each sample point of the test signals, one value per line. (I.e., the file should have 76566 lines.) The submissions are evaluated in view of a one dimensional cursor control application with range from -1 to 1. The mental state left is used to position the cursor at -1, and the mental state foot is used to position the cursor near 1. In the absense of those mental states (left and foot) the cursor should be at position 0. The latter requirement is checked at intervals in which the subject was instructed to relax. The output during intermitting periods (in which the subject is not in a defined mental state) are ignored. Note that it is unknown to the competitors at what intervals the subject is in a defined mental state. Competitiors submit classifier outputs for all time points. The evaluation function calculates the squared error with respect to the target vector that is -1 for class left, 1 for foot, and 0 for relax, averaged across all time points for which the subject is in a defined mental state (these 'active areas' are delayed for 500ms compared to the stimuli to account for the reaction time).
You also have to provide a description of the used algorithm (ASC II, HTML or PDF format) for publication at the results web page.

Technical Information

The recording was made using BrainAmp amplifiers and a 128 channel Ag/AgCl electrode cap from ECI. 118 EEG channels were measured at positions of the extended international 10/20-system. Signals were band-pass filtered between 0.05 and 200 Hz and then digitized at 1000 Hz with 16 bit (0.1 uV) accuracy. We provide also a version of the data that is downsampled at 100 Hz (by picking each 10th sample) that we typically use for analysis.


Note that the above reference describes an older experimental setup. A new paper analyzing data sets similar to the one provided in this competition and presenting feedback results will appear soon.

[ BCI Competition III ]