Facilitating gamified lower-extremity rehabilitation through novel VR MI-BCI system

• Main Author: Pramotedham, Chavalit
• Other Authors: Guan Cuntai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computer applications::Life and medical sciences
• Online Access: https://hdl.handle.net/10356/156693
• Institution: Nanyang Technological University

Within the rapidly progressing field of Brain Computer Interfaces (BCI), Motor-Imagery (MI) has been one of the most well-researched paradigms due to its potentially high impact in the medical domain, by facilitating rehabilitation in motor impaired post-stroke patients. However, there has been a general lack of focus on Lower-Extremity MI analysis, and the development of Serious Games catered specifically to the stroke rehabilitation use case. More systematic studies are also required to quantify the effects of Virtual Reality (VR) stimuli on BCI performance, to guide the development of future applications. In this project, a Serious Game in VR was developed to serve 2 main purposes. Firstly, the game is designed to facilitate lower-limb rehabilitation in post-stroke patients using MI BCI. Secondly, the game is designed to serve as an interface for data collection, to create a highly versatile dataset consisting of both Lower-Extremity (LE) and Upper-Extremity (UE) MI Electroencephalography (EEG) data, in VR and 2D environments. In the post-development phase, a cross-over experiment protocol was designed to study the effectiveness of LE- vs. UE-MI classification, which is a novel approach in MI BCI, as well as the effects of VR vs. 2D environments on BCI performance. Specifically, the protocol comprises 2 environments (VR and 2D), where participants are instructed to perform LE- and UE-MI tasks. The EEG data collected through the experiments were analyzed offline, to classify LE- vs. UE-MI states in VR vs. 2D environments, using existing Deep CNN frameworks. A 10-fold cross validation strategy was applied, and classification accuracies for each subject was computed for VR and 2D environments. In the preliminary results of this study, the average classification accuracies over subjects were 71.03% and 67.15% for VR and 2D respectively. This performance surpassed the BCI literacy threshold established by past studies, indicating the effectiveness of this novel LE- vs. UE-MI approach. Additionally, while the difference in BCI performance in VR vs. 2D was not deemed to be statistically significant (p = 0.38), this is likely due to the extremely small sample size (n = 12) within this preliminary study. General observations across subjects, however, point towards higher BCI performance in VR, supporting the use of VR stimuli in MI BCI systems.