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.

102 E.W. Sellers et al.<br />

Fig. 5 Comparison <strong>of</strong><br />

regression and classification<br />

for EEG feature translation.<br />

For the two-target case, both<br />

methods require only one<br />

function. For the five-target<br />

case, the regression approach<br />

still requires only a single<br />

function, while the<br />

classification approach<br />

requires four functions. (See<br />

text for full discussion)<br />

function. In contrast, for the five-target case the classification approach requires<br />

that four functions be parameterized. With even more targets, and with variable targets,<br />

the advantage <strong>of</strong> the regression approach becomes increasingly apparent. For<br />

example, the positioning <strong>of</strong> icons in a typical mouse-based graphical user interface<br />

would require a bewildering array <strong>of</strong> classifying functions, while with the regression<br />

approach, two dimensions <strong>of</strong> cursor movement and a button selection can access<br />

multiple icons, however many there are and wherever they are.<br />

We have conducted preliminary studies that suggest that users are also able to<br />

accurately control a robotic arm in two dimensions just as they control cursor movement<br />

(34]. In another recent study [32] we trained users on a task that emulated<br />

computer mouse control. Multiple targets were presented around the periphery <strong>of</strong> a<br />

computer screen, with one designated as the correct target. The user’s task was to<br />

use EEG to move a cursor from the center <strong>of</strong> the screen to the correct target and then<br />

use an additional EEG feature to select the target. If the cursor reached an incorrect<br />

target, the user was instructed not to select it. Users were able to select or reject the<br />

target by performing or withholding hand-grasp imagery [33]. This imagery evokes<br />

a transient EEG response that can be detected. It can serve to improve overall accuracy<br />

by reducing unintended target selections. The results indicate that users could<br />

use brain signals for sequential multidimensional movement and selection. As these<br />

results illustrate, SMR-based BCI operation has the potential to be extended to a<br />

variety <strong>of</strong> applications, and the control obtained for one task can transfer directly to<br />

another task.<br />

Our current efforts toward improving SMR-based BCI operation are focused<br />

on improving accuracy and reliability and on three-dimensional control [32]. This<br />

depends on identifying and translating EEG features so that the resulting control signals<br />

are as independent, trainable, stable, and predictable as possible. With control<br />

signals possessing these traits, the mutual user and system adaptations are effective,<br />

the required training time is reduced, and overall performance is improved.

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

Saved successfully!

Ooh no, something went wrong!