Highest BCI session of the world in the Himalayas


Christoph Guger, Robert Leeb, Herbert Ramoser, Hannes Stiebitzhofer, Michael Gruber, Nepal 2003


Measure biosignals | BCI Paradigm | Feature Extraction and Classification Error



Measure biosignals reliably even outside in the Himalayas

During an Austrian expedition to Chulu Far West (6419 m) g.MOBIlab was used to perform BCI sessions from almost sea level up to highcamp at 5600 m. The expedition started in Besi Sahar at an altitude of 700 m near Annapurna I. The team gained each day a height between 300 and 600 m and settled basecamp at 4800 m. After one night in BC the highcamp was established in 5600 m on the Chulu glacier. At 3 p.m. in the morning the team started to climb Chulu Far West (two left pictures on top) and reached the summit at 11 a.m. The BCI session used for the following analysis was recorded in the highcamp at 7 p.m.

 


BCI Paradigm

g.MOBIlab was used to record 2 EEG channels over the sensorimotor areas C3 and C4 and 1 ECG channel of 2 expedition members. The persons performed a self-paced finger movement every 10 seconds. A 4th channel was used to indicate the on-set and off-set of the movement and was recorded by an external switch connected to g.MOBIlab. The data were first triggered into epochs 4 seconds prior and 4 seconds after movement and resulted in 56 trials.

Then ERDmaps from 5 to 30 Hz with a bandwidth of 2 Hz were calculated. Red colors indicate an amplitude attenuation (event-related desynchronization) and blue colors an amplitude enhancement (event-related synchronization). The data show a clear alpha and beta power desynchronization on the contralateral channel. The desychnronization starts approximately 1 second prior to movement onset (planning of movement; movement onset was at second 4 in the right picture) and ends about 1 second after movement onset.


Feature Extraction and Classification

A Hjorth parameter estimation algorithm of g.BSanalyze was used to calculate the Activity, Mobility and Complexity of the 2 EEG channels. All three Hjorth features were classified with a 10 times 10 fold cross validation of a linear discriminant analysis. The time course of the classification error is shown in the right picture. The classification error reaches its minimum of about 11 % at second 3.5 which is 0.5 seconds before movement onset.


Classification Error in %

The recording was performed with g.MOBIlab, the analysis with g.BSanalyze. Both products are registered trade-
marks of GUGER TECHNOLOGIES OEG.

Photos made by Christoph Guger, Robert Leeb, Herbert Ramoser, Hannes Stiebitzhofer, Michael Gruber, Nepal 2003.


Up