16.11.2012 Views

Brain–Computer Interfaces - Index of

Brain–Computer Interfaces - Index of

Brain–Computer Interfaces - Index of

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Digital Signal Processing and Machine Learning 327<br />

Step 2 For the qth round in a new trial, feature vectors Fek,q (k = 1, ··· , 12,<br />

q = 1, ··· , 10) are extracted. Applying the SVM to these feature vectors,<br />

120 scores denoted as sk,q are obtained. These scores are the values <strong>of</strong> the<br />

objective function <strong>of</strong> the SVM.<br />

Step 3 P300 detection. Sum the scores for all the rows or columns k = 1, ··· ,12<br />

across all the q rounds using ssk = �10 q=1 sk,q. Therow(k = 1, ··· , 6) and<br />

column (k = 7, ··· , 12) with the respective maximum summed scores are<br />

then identified. The decision is then made on the character at the intersection<br />

<strong>of</strong> the identified row and column. This character forms the output <strong>of</strong> the BCI<br />

speller.<br />

Using this BCI speller, the 10 users were able to write the test phrase <strong>of</strong> 42<br />

characters with an average accuracy <strong>of</strong> 99% [54].<br />

8 Summary<br />

This chapter discusses signal processing and machine learning methods that are<br />

commonly used in BCIs. However, there are many methods that are not discussed.<br />

Furthermore, signal processing <strong>of</strong>ten involves several stages and there is no best<br />

method for each stage. Nevertheless, the most suitable method and parameters<br />

depend on the brain patterns used in BCIs (such as P300, SSVEP, or ERD/ERS),<br />

the quantity and quality <strong>of</strong> the training data acquired, the training time, and various<br />

user factors.<br />

Acknowledgments The authors are grateful to G. Townsend, B. Graimann, and G. Pfurtscheller<br />

for their permission to use Figures 7 and 8 in this chapter. The authors are also grateful to anonymous<br />

reviewers and the editors B. Graimann, G. Pfurtscheller, B. Allison <strong>of</strong> this book for their<br />

contributions to this chapter. Yuanqing Li’s work was partially supported by National Natural<br />

Science Foundation <strong>of</strong> China under Grant 60825306, Natural Science Foundation <strong>of</strong> Guangdong<br />

Province, China under Grant 9251064101000012. Kai Keng Ang and Cuntai Guan’s work were<br />

supported by the Science and Engineering Research Council <strong>of</strong> A ∗ STAR (Agency for Science,<br />

Technology and Research), Singapore.<br />

References<br />

1. B. Rockstroh, T. Elbert, A. Canavan, W. Lutzenberger, and N. Birbaumer, Eds., Slow cortical<br />

potentials and behavior, 2nd ed. Urban and Schwarzenberg, Baltimore, MD, (1989).<br />

2. N. Birbaumer, Slow cortical potentials: their origin, meaning and clinical use. In G.J.M.<br />

van Boxtel and K.B.E. Böcker, (Eds.) Brain and behavior:past, present, and future, Tilburg<br />

University Press, Tilburg, pp. 25–39, (1997).<br />

3. E. Donchin, ‘Surprise!...surprise?’ Psychophysiology, 18(5) 493–513, (1981).<br />

4. L.A. Farwell and E. Donchin, Talking <strong>of</strong>f the top <strong>of</strong> your head: Toward a mental prosthesis<br />

utilizing event-related brain potentials. Electroencephalogr Clin Neurophysiol, 70(6) 510–<br />

523, (1998).<br />

5. T.F. Collura, Real-time filtering for the estimation <strong>of</strong> steady-state visual evoked brain<br />

potentials. IEEE Trans Biomed Eng, 37(6), 650–652, (1990).

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!