Enhancing the Hybrid BCI Performance With the Common Frequency Pattern in Dual-Channel EEG

The brain-computer interface establishes a direct communication pathway between the human brain and an external device by recognizing specific patterns in cortical activities. The principle of hybridization stands for combining at least two different BCI modalities into a single interface with the aim of improving the information transfer rate by increasing the recognition accuracy and number of choices available for the user. This study proposes a simultaneous hybrid BCI system that recognizes the motor imagery (MI) and the steady-state visually evoked potentials (SSVEP) using the EEG signals from a dual-channel EEG setting with sensors placed over the central area (C3 and C4 channels). The data processing implements a supervised optimization algorithm for the feature extraction, named the common frequency pattern, which finds the optimal spectral filter that maximizes the separability of the data by classes. The experiment compares the classification accuracy in a two-class task using the MI, SSVEP and hybrid approaches on seventeen healthy 18-29 years old subjects with various dual-channel setups and complete set of thirty EEG electrodes. The designed system reaches a high accuracy of 97.4 ± 1.1% in the hybrid task using the C3-C4 channel configuration, which is marginally lower than the 98.8 ± 0.5% accuracy achieved with the complete set of channels while applying the support vector classifier; in the plain SSVEP task the accuracy drops from 91.3 ± 3.9% to 86.0 ± 2.5% while moving from the occipital to central area under the dual-channel condition. The results demonstrate that by combining the principles of hybridization and data-driven spectral filtering for the feature selection it is feasible to compensate a lack of spatial information and implement the proposed BCI using a portable few channel EEG device even under sub-optimal conditions for the sensors placement.