Difference between revisions of "BCI-KurtusKahleFL2011"

From ESE497 Wiki
Jump to navigationJump to search
Line 5: Line 5:
  
 
== Methods ==
 
== Methods ==
There are two methods of spatial filtering being investigated.  First, those currently applied with clinical EEG caps, namely Bipolar, Laplacian, and CAR.  An additional "Half CAR" filter will be investigated.  For this filter, the reference for a given electrode will be the average of the electrodes in its respective hemisphere.  This filter may be more effective than full CAR if there is noise specific to one hemisphere, which may result from the brain signaling muscle motions to one side of the body.  The laplacian filters need to be modified for this headset.  Normally, one electrode would be taken from each of 4 sides of the electrode under investigation, to show how the signal changes with respect to the sampling location.  The Emotiv Epoc headset does not have electrodes in the necessary locations for this procedure, so for this investigation, as many nearest electrodes as possible (either two or three) will be used.  For each of these filters, R^2 values will be computed and compared.
+
There are two methods being used to investigate spatial filtering.  First, those filters currently applied with clinical EEG caps, namely Bipolar, Laplacian, and CAR, are being explored.  An additional "Half CAR" filter will also be investigated.  For this filter, the reference for a given electrode will be the average of the electrodes in its respective hemisphere.  This filter may be more effective than the standard CAR if there is noise specific to one hemisphere, which may result from the brain signaling muscle motions to one side of the body, for example.  The laplacian filter schema needs to be modified for this headset.  Normally, four electrodes –– one from each side of the electrode under investigation –– would be used to create a reference.  Such a combination shows how the signal changes with respect to the sampling location.  The Emotiv Epoc headset does not have electrodes in the necessary locations for that procedure, so for this investigation, as many nearest electrodes as possible (either two or three) will be used.  For each of these filters, R^2 values will be computed and compared.
  
 
The second method is directly optimizing for linear combinations of electrodes.  Using MATLAB, it is possible to find local maxima (and ideally the global maximum) of R^2 based on the variations in linear combination.
 
The second method is directly optimizing for linear combinations of electrodes.  Using MATLAB, it is possible to find local maxima (and ideally the global maximum) of R^2 based on the variations in linear combination.
  
Another important factor in spatial filtering is the role of the temporal alignment of the electrodes.  EEG headsets generally sample each electrode sequentially rather than all electrodes simultaneously.  This does not effect data for most applications, but in spatial filtering applications noise is produced when one waveform is subtracted from a similar waveform.  For this study, the data will be linearly interpolated to approximate the values that would be seen, were the electrodes sampled simultaneously.  The manufacturer has provided data on approximate temporal offset between electrodes, which will be used for this study.  The R^2 values for uninterpolated and interpolated data will be compared.  In future studies, the offset could be found experimentally and higher order interpolation functions could be used rather than the linear model.
+
Another important factor in spatial filtering is the temporal alignment of the electrodes.  EEG headsets generally sample each electrode sequentially rather than all electrodes simultaneously.  This does not effect data for most applications, but in spatial filtering applications noise is produced when one waveform is subtracted from a similar waveform.  For this study, the data will be linearly interpolated to approximate the values that would be seen, were the electrodes sampled simultaneously.  Emotiv has provided data on approximate temporal offset between electrodes.  Those data will be used as the offset values for this study.  The R^2 values for uninterpolated and interpolated data will be compared.  In future studies, the offset could be found experimentally and higher order interpolation functions could be used rather than the linear model.
  
 
== Still to do ==
 
== Still to do ==

Revision as of 20:25, 18 December 2011

Abstract

The purpose of this project is to determine the optimal spatial filtering techniques for a Brain-Computer Interface (BCI) using an Emotiv Epoc EEG headset. A spatial filter is type of tool used to improve the characteristics of a signal by linearly combining the outputs of multiple electrodes in a particular pattern. Those combinations produce a new signal with less noise, which gives BCI patients better control over a device, such as an orthosis.

In general, the methods used are a simple ear-reference, Bipolar filters, Common Average Reference (CAR) filters, and Laplacian filters. Bipolar filters one electrode as the reference for another electrode. CAR filters use the mean of every electrode's voltage at a given time as their zero. Laplacian filters use several near neighbor electrodes to approximate a spatial second derivative of the signal. While the Emotiv headset offers certain advantages for BCI applications, its limited number of electrodes in a limited number of locations makes generating accurate signals much more difficult, particularly for Laplacian filters. Obtaining high-quality signals given these limitations is the primary goal of this investigation.

Methods

There are two methods being used to investigate spatial filtering. First, those filters currently applied with clinical EEG caps, namely Bipolar, Laplacian, and CAR, are being explored. An additional "Half CAR" filter will also be investigated. For this filter, the reference for a given electrode will be the average of the electrodes in its respective hemisphere. This filter may be more effective than the standard CAR if there is noise specific to one hemisphere, which may result from the brain signaling muscle motions to one side of the body, for example. The laplacian filter schema needs to be modified for this headset. Normally, four electrodes –– one from each side of the electrode under investigation –– would be used to create a reference. Such a combination shows how the signal changes with respect to the sampling location. The Emotiv Epoc headset does not have electrodes in the necessary locations for that procedure, so for this investigation, as many nearest electrodes as possible (either two or three) will be used. For each of these filters, R^2 values will be computed and compared.

The second method is directly optimizing for linear combinations of electrodes. Using MATLAB, it is possible to find local maxima (and ideally the global maximum) of R^2 based on the variations in linear combination.

Another important factor in spatial filtering is the temporal alignment of the electrodes. EEG headsets generally sample each electrode sequentially rather than all electrodes simultaneously. This does not effect data for most applications, but in spatial filtering applications noise is produced when one waveform is subtracted from a similar waveform. For this study, the data will be linearly interpolated to approximate the values that would be seen, were the electrodes sampled simultaneously. Emotiv has provided data on approximate temporal offset between electrodes. Those data will be used as the offset values for this study. The R^2 values for uninterpolated and interpolated data will be compared. In future studies, the offset could be found experimentally and higher order interpolation functions could be used rather than the linear model.

Still to do

• Determine the effectiveness of each filter created for the Emotiv headset.

• Code a script to optimize possible linear combinations of electrodes for R^2.

Further reading

McFarland, D. J., McCane, L. M., David, S. V., & Wolpaw, J. R. (1997). Spatial filter selection for eeg-based communication. Electroencephalography and clinical Neurophysiology, (103), 386-384.

Ramoser, H., Müller-Gerking, J., & Pfurtscheller, G. (2000). Optimal spatial filtering of single trial eeg during imagined hand movement. IEEE Transactions on Rehabilitation Engineering, 6(4), 441-446.