Tuning Methodology for a Hybrid BCI

This paper describes the methodology used for the selection of the best parameters to generate commands in a brain computer interface (BCI),based on a hybrid combination of two paradigms: Steady-State Visual Evoked Potentials (SSVEP) and Imaginary Movement (IM). The state of the art mentions that associated patterns are identifiable from the electroencephalographic signal (EEGS), however the preprocessing of the data is compulsory in order to generate the patterns. The protocol and methodology for the extraction of characteristics from the EEGS to each recording is described for both techniques. The Principal Component Analysis was used to select the characteristics that most contribute to the separation of the data from SSVEP. Besides, wavelet decomposition with Independent Components Analysis was used to obtain the characteristic patterns for the considered movement intention. A comparison was made of the efficiencies obtained with different classification methods: neural networks, nearest neighbors, minimum distance and Mahalanobis distance for SSVEP and a classifier based on correlation was proposed for IM. The number of electrodes to be used was also optimized for the use of these two techniques. Using this automatic extraction, the selection of parameters and the identification of the best classifier following this tuning methodology proves to reduce the number of required electrodes to 50%, resulting in 7 electrodes for this hybrid BCI. Besides it provides a systematic analysis of the minimal time required for a threshold of 90% of efficiency for each user. In this study case, 4 seconds for IM and 2 seconds for SSVEP were obtained and it was found that all participants are good candidates to be BCI users and the best characteristics and classifiers were found.