Phase Analysis in Steady-State Visual Evoked Potential (SSVEP) based BCIs

Brain-computer interfaces (BCI) based on Steady State Visual Evoked Potential (SSVEP) can provide higher information transfer rate than other BCI modalities. For the sake of safety and comfort, the frequency of the repetitive visual stimulus (RVS) necessary to elicit an SSVEP, should be higher than 30 Hz. However, in the frequency range above 30 Hz, only a limited number of frequencies can elicit sufficiently strong SSVEPs for BCI purposes. Consequently, the conventional approach, consisting in presenting various repetitive visual stimuli having different frequency each, is not practical for SSVEP based BCI functioning. Indeed this would bring low communication bitrates. In order to increase the number of possible repetitive visual stimuli, we consider modulating the phase of the stimulus instead of the frequency. Thus, several stimuli, sharing the same frequency, but with different phase can be presented to the user. The approach presented in this document, to detect the phase of the stimulus is termed phase synchrony. It consists in using as feature, to identify a subject's focus of attention, the phase difference between the SSVEP and the stimulus. The phase is extracted through the Hilbert transform applied on an univariate signal resulting from spatially filtering the electroencephalogram. We have conducted experiments with seven subjects to estimate the information transfer rate that can be achieved using the phase synchrony analysis method. Conclusions: The phase in high frequency band (32Hz-40Hz) can be use to increase the number of possible commands in BCI operation. The Hilbert transform (phase synchrony) effectively extracts the phase PR-TN 2010/00081 Unclassified iv Koninklijke Philips Electronics N.V. 2010 difference Spatial filter can improve the classification accuracy It is represent that the coefficient of spatial filter (w) is different for each subject, so the spatial filter is subject dependent. Information transfer rate calculated by Nykopp equation is more than Walpaw equation. Information transfer rate (bit rate) by calculating spatial filter is more than Oz-Cz derivation The result of information transfer rate achieving with spatial filter in both Synchronous and Asynchronous compare to Oz-Cz combination, prove the advantage of spatial filter enhancement. Unclassified PR-TN 2010/00081 Koninklijke Philips Electronics N.V. 2010 v Figures Index Figure 2.1.Brain main parts ........................................................................................................... 11 Figure 2.2.The four lobes of cerebral cortex ................................................................................. 11 Figure 2.3.Comparison methods of brain activity measurement [8].............................................. 12 Figure 2.4.Synaptic transmission-communication between neurons ............................................ 13 Figure 2.5.Left picture shows cortical field potentials[9]the right picture shows the structure of pyramidal cells as a dipole ............................................................................................................ 13 Figure 2.6.The surface electrode position [9] ................................................................................ 14 Figure 2.7.Standard 10-20 system [9] ........................................................................................... 14 Figure 2.8.Schematic diagram of typical BCI system [8] .............................................................. 15 Figure 3.1.Calculate Energy in stimuli frequency as feature ......................................................... 17 Figure 3.2.Extract feature from signal by sliding window-the statistical properties like mean or mode is used to extract feature from each window ....................................................................... 18 Figure 3.3.Extract phase as a feature. .......................................................................................... 18 Figure 3.4.Extract feature from signal by sliding window-the statistical properties like mean or mode is used to extract feature from each window ....................................................................... 19 Figure 3.5.Example of two extracted features of signal versus each other for one subject with 1second window without overlap..................................................................................................... 20 Figure 3.6.Example of two extracted features of signal versus each other for one subject with spatial filter (1-second window without overlap) ............................................................................ 23 Figure 3.7.Classification system.................................................................................................... 24 Figure 3.8.Single layer Neural Network with two features and three classes ............................... 24 Figure 3.9.The definition of each cell of confusion Matrix Probability-The diagonal elements indicate TP and the off-diagonal elements indicate FP (classes have the same probability) ................................................................... 26 Figure 3.10.Indicate the number of TP, FP, TN and FN ............................................................... 27 Figure 3.11.Confusion Matrix probability ....................................................................................... 27 Figure 3.12.Calculate the information transfer rate and accuracy with the specific shift and window size, the blue arrows show the shift. ................................................................................ 29 Figure 4.1.Distribution of LEDs ..................................................................................................... 31 Figure 4.2.The sounds hear from speaker two seconds before starting the stimuli to tell the subject to be ready for looking at the new direction ...................................................................... 32 Figure 4.3.EEG electrode placement according to the 10-20 system ........................................... 33 Figure 5.1.Feature energy versus feature phase with Oz-Cz ....................................................... 35 Figure 5.2.Feature energy versus feature phase with Spatial filter ............................................... 36 Figure 5.3.Compare the energy of Oz-Cz and the spatial filter SSVEP signal for S1 .................. 36 Figure 5.4.Information transfer rate result with different window size (0.1-2.7sec) with different overlap for subject 1-red indicates spatial filter and blue indicates Oz-Cz combination ............... 37 Figure 5.5.The best result for Information transfer rate and accuracy .......................................... 38 Figure 5.6.Display energy versus phase feature with Oz-Cz combination and after spatial filtered signal for best window size and overlap-Left figures show, the Train and test with Oz-Cz and Right figures show the train and Test trials after spatial filtering for S1 ........................................ 38 Figure 5.7.Calculated bit/trial for spatial filtered signal (red) and Oz-Cz signal (blue) for subject 1 with different window size (0.1 to 2 sec) ....................................................................................... 40 Figure 5.8.The best Information Transfer rate and test accuracy for subject one ........................ 41 Figure 5.9.Comparison of Instantaneous phase difference between SSVEP and stimuli signal for 1-second EEG record .................................................................................................................... 43 Figure 5.10.Compare spatial filtered signal for subjects ............................................................... 44 Figure 5.11.The best ROC curve for each subject ........................................................................ 44 PR-TN 2010/00081 Unclassified vi Koninklijke Philips Electronics N.V. 2010 Tables Index Table 1.Data classification ............................................................................................................ 22 Table 3.The optimal frequency selects for each subject ............................................................... 31 Table 4.The stimuli place that have the most bit/trial for subject 1-40Hz with different window size and best shift ................................................................................................................................. 41 Table 5.The AUC for the best ROC curve for each subject .......................................................... 45 Table 6.The best information transfer rate in Synchronous and Asynchronous methods for each subject ........................................................................................................................................... 45 Table 7.The location of best information transfer rate with Asynchronous and Synchronous method for each subject ................................................................................................................ 46 Unclassified PR-TN 2010/00081 Koninklijke Philips Electronics N.V. 2010 vii