Brain keyboard interface design with deep learning method using EEG signals

Brain-computer interface (BCI) studies continue to captivate researchers with new dimensions. Predicting letters through brain signals, termed as a brain keyboard, stands out as a significant research area. The primary motivation behind these studies is to address communication challenges for millions of people worldwide who, for various reasons, cannot speak. EEG signals, such as P300 and Steady-State Visual Evoked Potentials (SSVEPs), have been commonly used in BCI research. However, individuals exposed to paradigms containing continuous stimuli like light while receiving these signals may experience fatigue over time, leading to a decline in attention to stimuli and performance. Additionally, acquiring and processing these signals is time-consuming. This study explores the effects of four proposed paradigms on different EEG channels for designing a brain keyboard using EEG signals to address these challenges. The paradigms are based on conditions such as imagining letters, moving the tongue and lips to pronounce letters while keeping them silent, and pronouncing letters silently or aloud. Power spectral density (PSD) values in the 8-30 Hz frequency range of EEG data serve as feature vectors for paradigm-induced brain signals. These feature vectors are classified using Long Short-Term Memory (LSTM), Support Vector Machine (SVM), and k- Nearest Neighbors (k-NN) algorithms. The highest accuracy is achieved with the LSTM deep learning algorithm. The highest accuracy rate, 98.66%, is obtained in recordings involving the pronunciation of letters through tongue and lip movements in the O1 channel corresponding to the 13th channel. Furthermore, a 97.88% accuracy is achieved in recordings involving silent pronunciation of letters with tongue and lip movements in the AF4 electrode corresponding to the 2nd channel. In conclusion, the classification output obtained with the innovative EEG recording paradigm holds promising potential for future applications in human-machine interaction.