P300 speller tested with several 100 persons in 2010
Fig. 1: BCI components
g.tec tested the P300 based speller Intendix® with hundreds of people at conferences, exhibitions and social events in 2010. The performance in terms of accuracy and spelling speed was impressively high, although most of the tests were done under noisy conditions and with new BCI users.
intendiX® is designed to be installed and operated by caregivers or the patient’s family at home. The system is based on visually evoked EEG potentials (VEP/P300) and enables the user to sequentially select characters from a keyboard-like matrix on the screen just by paying attention to the target for several seconds (see Fig. 1).
In a previous study (Guger et al. 2009) we examined the overall accuracy of our P300 speller for healthy subjects. After five minutes of training the subjects were asked to spell 5 characters either with the row/column (RC) speller or with the single character (SC) speller. 72.8% (N=81) were able to spell with 100 % accuracy in the RC paradigm and 55.3 % (N=38) spelled with 100 % accuracy in the SC paradigm. Less than 3% of the subjects did not spell any character correctly.
C. Guger, S. Daban., E. Sellers, C. Holzner, G. Krausz, R. Carabalona, F. Gramatica, G. Edlinger (2009). "How many people are able to control a P300-based-brain-computer interface (BCI)?" Neuroscience Letters 462: 94 - 98. How many people are able to control a P300-based brain–computer interface (BCI)?
Based on these results we invited people at conferences and exhibitions to test intendiX® . Most of the volunteers never used a BCI system before but were interested or even skeptic.
P300 speller performance for healthy subject under noisy conditions
Successful Intendix® testing (>10 letters
The system consisted of an EEG cap (g.GAMMAcap), active electrodes (either g.LADYbird or g. BUTTERfly), the g.GAMMAbox supplier box connected to the main amplifier (either g.USBamp or g.MOBIlab+) and the P300 software. Experiments were made at more than 30 exhibitions and conferences, among them there are: Arab Health 2010, Dubai, AE (> 20 volunteers); CeBit, Hanover, Germany (>30 volunteers); CNS 2010, Montreal, Canada (> 20 volunteers); ICT, Brussels, Belgium (>30 volunteers); CNS 2010, San Antonio, TX, USA (>20 volunteers); ICCN 2010, Kobe, Japan (>15 volunteers); SfN 2010, San Diego, CA, USA (>30 volunteers). The subjects were trained within 5 min using only 5 characters. Afterwards subjects had to copy spell 5 characters to measure the performance.
After one year of testing the BCI system we can summarize that the copy spelling accuracy after training is greater than 90%. In most volunteers the accuracy is 100% with 15 flashes per row and column. About 20 % of people are still able to spell without any mistake with only 2 flashes per row and column. This is an extremely good result for only 5 training characters. An increase of training characters to e.g. 42 letters increases the speed significantly, but was avoided at these exhibitions. Some subjects were even able to spell with only 1 flash without mistake. Interesting is also that people report to be able to follow conversations or even to participate in conversations with exhibition visitors during spelling.
Why is perfect technology important?
The experiments showed that the P300 BCI system is extremely robust and works with very high accuracy. Of special importance are perfect EEG signals to have the whole EEG spectrum even with low frequencies available. The g.tec system achieves that with amplifier technology inside the EEG electrodes and with very high over-sampling rates combined with 24 Bit resolution. The P300 response results in very small amplitudes of just a few µV and therefore a high quality recording is needed. Artifacts in the training data and also in real-time mode would destroy the performance of the system. This is of special importance for BCI usage with patients which is of course more difficult. If there is any doubt that the system is not perfectly configured than it will be difficult to interpret spelling results correctly.
Important is also that experiments are used in real-time that allow an objective measurement of accuracy. The copy spelling task is easy to understand and gives a clear accuracy measure (number of characters correctly spelled). This is a crucial property of BCI systems in contrast to other systems that sometimes present tasks which do not allow to measure accuracy.
Information about the intendiX could be found here.