16.11.2012 Views

Brain–Computer Interfaces - Index of

Brain–Computer Interfaces - Index of

Brain–Computer Interfaces - Index of

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

A Simple, Spectral-Change Based, Electrocorticographic Brain–Computer Interface 245<br />

common-mode phenomena. One must be careful that there are not any electrodes<br />

that are broken, or have extremely large contamination, or every electrode will be<br />

contaminated by the re-referencing process. Local re-referencing may also be performed,<br />

such as subtracting the average <strong>of</strong> nearest-neighbors (Laplacian), which<br />

ensures that the potential changes seen in any electrode are spatially localized. One<br />

may also re-reference in a pair-wise fashion, producing bipolar channels which are<br />

extremely local, but phenomena cannot be tied to a specific electrode from the pair.<br />

This re-referencing can also be interpreted as applying a spatial filter. Please see<br />

Chapter “Digital Signal Processing and Machine Learning” for details about spatial<br />

filters.<br />

In order to appropriately understand both the experimental context and the<br />

connection between the structure <strong>of</strong> the brain and signal processing findings, it<br />

is necessary to co-register electrode locations to the brain surface. The simplest<br />

method is to use x-rays, and plot data to template cortices (as illustrated in Fig. 2).<br />

A clinical schematic will typically be obtained from the surgeon. The position <strong>of</strong><br />

each electrode may then be correlated with potential recordings from each amplifier<br />

channel. Different diagnostic imaging may be obtained from the course <strong>of</strong> the clinical<br />

care, or through specially obtained high-fidelity experimental imaging. The level<br />

and quality <strong>of</strong> this may be highly variable across time and institutions, from x-ray<br />

only, to high-fidelity pre-operative magnetic resonance imaging (MRI) and postoperative<br />

fine-cut computed tomography (CT). The clinical schemata and diagnostic<br />

imaging may be used in concert to estimate electrode positions, recreate cortical<br />

locations, and plot activity and analyses. The most simple method for doing this,<br />

using x-rays, is the freely-available LOC package [16], although there is the promise<br />

<strong>of</strong> more sophisticated methodology for doing this, when higher fidelity diagnostic<br />

imaging is obtained.<br />

Choosing a sampling frequency is important – there is <strong>of</strong>ten a trade-<strong>of</strong>f between<br />

signal fidelity and the practical issues <strong>of</strong> manageable data sizes and hardware limitations.<br />

Whatever sampling rate is chosen, one should be sure to have high signal<br />

fidelity up to at least 150 Hz. This means that the sampling rate should be above<br />

300 Hz, because <strong>of</strong> a law called the Nyquist Law (aka Nyquist Theorem), which<br />

says that you must record data at (at least) twice the frequency <strong>of</strong> the highest wave<br />

you wish to measure. The sampling rate may have to be higher if the amplifiers used<br />

have built in filtering properties. The reason for this is that there is a behavioral split<br />

in the power spectrum (see Fig. 3) which can be as high as 60 Hz [17]. In order to<br />

capture the spatially focal high frequency change, one must have large bandwidth<br />

above this behavioral split. Some characteristic properties <strong>of</strong> motor and imagery<br />

associated spectra are shown in Fig. 3. There is a decrease in the power at lower<br />

frequencies with activity, and an increase in the power at higher frequencies [6, 18–<br />

20]. The intersection in the spectrum is dubbed the “primary junction” (J0). A recent<br />

study involving hand and tongue movement [17] found that, for hand movement,<br />

J0 = 48+/−9 Hz (mean+/− SD) (range 32−57 Hz), and, for tongue movement,<br />

J0 = 40+/−8 Hz (range 26–48 Hz). Rather than this indicating two phenomena, a<br />

“desynchronization” at low frequencies, and a “synchronization” at high frequencies,<br />

as some have proposed [19, 21], this might instead reflect the superposition

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

Saved successfully!

Ooh no, something went wrong!