PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Towards gaze-independent visual brain-computer interfaces
Matthias S Treder, Nico M Schmidt and Benjamin Blankertz
Front. Comput. Neurosci. Conference Abstract: Bernstein Conference on Computational Neuroscience 2010.

Abstract

BACKGROUND The designated goal of a brain-computer interface (BCI) is to establish a direct link from the brainto a machine, thereby circumventing motor behavior. A popular approach to BCIs is the use of the eventrelated potential (ERP). There is now ample evidence that conventional brain-computer interfaces based on event-related potentials can be operated e ciently only when targets ffi are fixated with the eye [1,2]. Critically, this limits the scope of visual spellers to the scope of eyetrackers, that is, it is basically a means to recover eye gaze. Furthermore, users with impaired control of eye movements cannot use conventional visual BCIs in an efficient way. To overcome these limitations, the aim of the present study was to develop visual spellers that are highly accurate, fast-paced, and independent of eye gaze. Three different variants of a visual speller based on the ERP Hex-o-Spell [1] were tested in an online experiment with 13 healthy participants. Participants could use both covert spatial attention and feature attention (i.e., attention to color and form) to attend to the targets. METHOD AND RESULTS For each kind of speller, participants traversed a calibration phase (on basis of a classifier was trained), a copy-spelling phase (wherein they spelled a pre-defined sentence), and a free-spelling phase (wherein they spelled a self-invented phrase). For classification, we used linear discriminant analysis with shrinkage of the covariance matrix [3]. For all participants, high-accuracy BCI control was achieved, see Figure 1. Selecting one out of thirty symbols (chance level 3.3%) for the three spellers yielded mean accuracies of 90.43%, 87.95%, and 97.03%, respectively. CONCLUSION The present results are a proof of concept: It is possible to realize fast-paced, high-accuracy visual spellers that have a large vocabulary and that are independent of eye gaze. The most significant implication is that the scope of visual spellers is broadened from simply recovering eye gaze to recovering attentional focus independent of eye gaze.

EPrint Type:Article
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Brain Computer Interfaces
ID Code:7936
Deposited By:Benjamin Blankertz
Deposited On:17 March 2011