EXPLORING THE EFFECTS OF OFFLINE PARADIGMS AND FEATURE EXTRACTION TECHNIQUES ON PERFORMANCE OF MOTOR IMAGERY BASED BRAIN-COMPUTER INTERFACE

Motor Imagery (MI) Brain-Computer Interface (BCI) has become a popular way of allowing disabled and healthy individuals to use brain signals to communicate with their environment. However, despite its potential, MI-based BCI still possesses technical and human factor challenges that affect classification performance. This research study tested two offline training protocols involving modern technical and human factors to improve BCI users' classification performance. Area Under the Curve (AUC) Receiver Operating Characteristics (ROC) curve was the metric for classification performance. Due to the COVID-19 pandemic, the researcher was the only participant in the study. The experimental scenario involved comparing two improved static image paradigms and phase synchronization-based features to the standard (S) arrow paradigm and Common Spatial Pattern (CSP) features. The tasks for the new paradigms were lifting: (i) a bottle (Target-Direct/TD) and (ii) a kettle to pour some tea into a mug (Purpose-driven/PD). The primary dataset involved three experiments, each involving eight sessions. The effect of training on performance was investigated by comparing the primary datasets to the older supplemental datasets. The supplementary datasets were recorded months before the primary datasets. This study also explored the influence of paradigm choice and phase synchronization-based features on classification performance. Results showed that performance can improve with time and that TD-CSP-wPLI (16-30Hz) and S- CSP-wPLI (12-15Hz) both produced the most noticeable change in performance. Results showed that using both advanced paradigms and features significantly improves both classification and usability