Characterization of EEG Signal Patterns During Visual Imageries of Basic Structures for the Development of Brain-Computer Typing Interface for Locked-In Syndrome Patients

The paper aims to characterize Electroencephalogram (EEG) signals during visual imagery of basic shapes that includes square, triangle and circle with and without visual stimulus and neutral state using a 14- channel EEG Emotiv EPOC+. Principal Component Analysis (PCA) was utilized to reduce the dimensionality of the features and the transformed features or biomarkers were used to train the classifiers. Classifiers used in this study are Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and k Nearest Neighbors (KNN). The results obtained from 5 study volunteers indicated that the highest contributions to the 56 new biomarkers are the higher order even central moments and more predominantly from channel T7. Also, the extracted features from EEG signals during visual imagery of different shapes with and without visual stimuli consistently showed a low degree of correlation. Furthermore, the dataset used to train the classifiers were subdivided into two: one containing neutral state with visual stimulus, and the other comprising neutral state without visual stimulus. Performance of different classifiers trained with and without visual stimulus yielded similar accuracies; however, the dataset with the absence of visual stimulus exhibit higher classification accuracies for all classifiers. In addition, all classifiers obtained high classification accuracies (>96%) for both datasets and the SVM performed best among the classifiers having accuracies of 97.5% and 99.5% for datasets with and without visual stimulus respectively. The study supports the feasibility of a brain-computer typing interface that utilizes visual imagery as an input modality. Furthermore, the findings of this study will serve as a basis for the development of a brain-computer typing interface using visual imagery of characters and letters.