Sort by: [year] [type] [author]

Advantages of a Gabor Wavelet Dictionary in ECoG Signal Analysis

Zachary Freudenburg, Nicholas Anderson, William D. Smart, Robert Pless, and Eric C. Leuthardt.
In "Society for Neuroscience Annual Meeting: Neuroscience 2008", Washington, DC, November 2008.

Our group has previously demonstrated the utility of electrocorticography (ECoG) as a platform for Brain-Computer Interfaces (BCIs). ECoG signals enable temporally accurate recording of brain activity in the higher gamma(> 40 Hz) frequency range as compared to EEG methodologies. Key to leveraging the temporal resolution advantages of ECoG is the ability to quickly and accurately capture fast changing high frequency features. In this study we explored to what degree using a Gabor wavelet dictionary (GWD) for ECoG signal frequency analysis can improve gamma frequency feature abstraction compared to previously used Fourier transform (FT) and autoregressive power spectral density analysis (ARPSDA) techniques.

A GWD that fits each frequency in the frequency feature space with an appropriately sized wavelet was created by combining a complex sine/cosine waveform with a Gaussian envelope. ECoG signal acquired during a hand and tongue movement screening task was analyzed in a 1 to 300 Hz frequency feature space using the GWD and standard windowed FT and ARPSDA techniques. An optimal window size was found for the task for both the FT and ARPSDA methods. The resulting spectrograms and correlation values of each frequency feature for each channel as related to the hand and tongue movements were compared.

In this study, a distinct increase in high frequency feature response to the hand movement task was seen with the Gabor which was not visualized when compared to both the widowed FT and ARPSDA spectrograms. The GWD spectrogram demonstrated a strong increase in the feature responses from 80 to 140 Hz with an approximate 750 ms time delay after a hand movement. This was also reflected in the correlations values of the hand movement verses rest. The GWD analysis results in statistically significant correlation values of 80 to 140 Hz features between hand movement and the lag corrected ECoG signal across several channels. Both the FT and ARPSDA methods failed to show any statically significant correlated gamma frequency features for either hand or tongue movement

The GWD frequency feature analysis method showed improved frequency feature abstraction because the wavelet window size used to abstract the feature values for each frequency is automatically fitted to the wavelength of the respective feature. The Gabor method has particular advantage for fast changing signal features which are common with gamma rhythms. Additionally, both the increased sensitivity of the Gabor method compared to fixed window time methods and the reduced computational time of the Gabor method to fitted sliding window FT and ARPSDA, make it potentially well suited for real time control applications.

  author = {Freudenburg, Zachary and Anderson, Nicholas and Smart, William D. and Pless, Robert and Leuthardt, Eric C.},
  title = {Advantages of a {Gabor} Wavelet Dictionary in {ECoG} Signal Analysis},
  booktitle = {Society for Neuroscience Annual Meeting: Neuroscience 2008},
  address = {Washington, DC},
  month = {November},
  year = {2008}