Investigating EEG-based motor imagery decoding in the continuous control BCI paradigm

Moeini, Mahdi (2025) Investigating EEG-based motor imagery decoding in the continuous control BCI paradigm. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (2MB)

Abstract

Brain-computer interfaces (BCIs) based on motor imagery (MI) hold significant potential for restoring autonomy to individuals with severe motor disabilities. While decoding of short, discrete MI tasks has been extensively studied, the reliability of sustained MI decoding for continuous control paradigms remains under-explored. This study evaluates the feasibility of classifying four MI tasks (Hand, Feet, Tongue, Singing) versus rest over extended intervals (8–20 seconds) using EEG. Fifteen participants performed cued MI tasks in a simulated continuous control paradigm. A machine learning pipeline—optimized for real-time compatibility—was systematically assessed, focusing on training data volume, epoching strategies, feature selection, and classifiers. Under the optimal configuration (four training blocks, 4-second epochs with 75% overlap, 10 features, SVM), comparable decoding accuracies were obtained on average across participants for the four tasks (Hand: 78.4% ± 9.5, Feet: 74.0% ± 10.0, Tongue: 74.5% ± 6.4, Singing: 74% ± 5.6), with no significant inter-task differences (Friedman test, p = 0.41). Within participants, however, there was considerable variability in decoding accuracy among the tasks. Comparative analysis against a shorter-interval MI dataset revealed a significant reduction in accuracy under the continuous paradigm only for Singing MI (Mann-Whitney U, U = 44, p = 0.01). Subjective feedback highlighted Singing MI as the participants’ most preferred task (53.3% of participants selected it as their favorite), while Tongue MI was the least preferred (33.3% of participants selected it as their least favorite). These results underscore the viability of continuous MI-based control while emphasizing the need to balance technical performance with user-centric design. The novel task of singing MI emerged as a promising candidate for intuitive BCI applications, warranting further exploration of hybrid paradigms and ecological validation to enhance real-world usability.

Actions (login required)

View Item