Description of the calibration paradigms
See the following publication for more information:
Design considerations for long term non- invasive Brain Computer Interface training with tetraplegic CYBATHLON pilotaC (arrow-based calibration paradigm):
This is the conventional calibration paradigm used to collect data to develop the BCI decoder for different MI classes.
The pilot participated in two runs of experiment in every session in Phase I and one run per session in Phase II. In Phase I, each run consisted of a total of 10 trials per MI class and 30 trials for “rest”. In Phase II, each run consisted of 20 trials per MI class and 60 trials for “rest”. Refer to Figure 1(a) in the paper for graphical representation of the calibration paradigm. The rest trial preceded an MI trial. A fixation cross appeared at the start of the trial for 2s. This was followed by the task cue (‘REST’ or arrow for MI class) for 2s. The pilot began the task when the cue was replaced by a white cross at the centre of the screen, which remained on the screen for a total of 4s. At the end of the task period, the white cross disappeared signalling the end of the trial. A blank screen marked the period for relaxing for the pilot so he could prepare himself for the next trial. Stimcodes correspond to the beginning of each trial.
gC (game-based calibration paradigm):
The purpose of this calibration paradigm was to use a GUI that would be closer to the end application for the purpose of collecting data to develop the MI-BCI decoder.
The pilot participated in 8 runs of gC (one run being one game) in every session in Phase II. Each run in gC yielded four trials per class, resulting in a total of 16 trials per run. Refer to Figure 1(b) for a graphical description of the calibration paradigm. The pilot played the BrainDriver game, developed by the Cybathlon organizers. The pilot was asked to perform the task as per the following rules: move avatar left: left hand MI; move avatar right: right hand MI; turn on headlights: both hands MI; no command: rest. For calibration, the correct command was always fed to the game. However, the pilot was asked to play the game as if the commands were being generated due to his brain signals. The stimuli corresponding to each class were embedded in the datastream as the corresponding class appeared in the game. For training the classifier, a 4s window starting from each task appearance was used.
Phase I
Session 5: left hand, right hand, both hands, rest; aC (2 runs) Session 6: left hand, right hand, both hands, rest; aC (2 runs) Session 9: left hand, right hand, both feet, rest; aC (2 runs) Session 10: left hand, right hand, both feet, rest; aC (2 runs)
Phase II
Session 15: left hand, right hand, both feet, rest; aC (1 run), gC (8 runs)
Session 16: left hand, right hand, both feet, rest; aC (1 run), gC (8 runs) Session 17: left hand, right hand, both feet, rest; aC (1 run), gC (8 runs) Session 18: left hand, right hand, both feet, rest; aC (1 run), gC (8 runs) Session 19: left hand, right hand, both feet, rest; aC (1 run), gC (8 runs) Session 20: left hand, right hand, both feet, rest; aC (1 run), gC (8 runs) Session 21: left hand, right hand, both feet, rest; aC (1 run), gC (8 runs)
Marker Info
:‘S 3’: rest
‘S 4’: left hand MI
‘S 5’: right hand MI
‘S 6’: both hands/feet MI
EEG Hardware Setup
:64 active channels, sampling frequency 500 Hz Phases II: vEOG: Fp1, Fp2; hEOG: TP7, TP8
Directory Structure
:Cyabthlon_Data
|
--- Session 5 |
--- aC --- Session 6 |
--- aC .
. .
--- Session 15 |
--- aC --- gC --- Session 16 |
--- aC --- gC .
. .
--- Session 21 |
--- aC --- gC