Abstract
Brain-computer interfaces (BCIs) capable of decoding continuous movement of a limb in three dimensions represent a significant advancement over traditional BCI systems that involve classification of specific limb movements. Three-dimensional motion trajectory decoding (3DMTD) BCIs potentially enable intuitive neural control of virtual or prosthetic limbs. However, challenges in decoding accuracy (DA), user adaptation, and understanding the impact of feedback modalities have limited their practical implementation beyond laboratory settings.This thesis investigates how decoding algorithms, embodied feedback (feedback designed to promote a sense of ownership and agency over a virtual limb), and user factors influence EEGbased MTD-BCIs for 3D virtual limb control. An integrated experimental framework combining multimodal sensing with immersive environments was designed to enable embodied feedback using a convolutional neural network and long short-term memory (CNN-LSTM) architecture, which significantly outperformed traditional multiple linear regression (mLR) across all spatial dimensions.
A comprehensive comparison between screen-based and virtual reality feedback across ten participants revealed that immersive spatial feedback substantially improves DA (14.7% increase), particularly for forward movements. Functional connectivity (FC) analysis identified enhanced integration between sensorimotor and frontal executive regions in VR, providing neural evidence for the benefits of embodiment. An investigation of human factors and training effects revealed that user adaptation was supported by assistance reduction during BCI control, while unexpected psychological predictors emerged: lower vividness of visual imagery and higher fear of incompetence were associated with better performance, challenging conventional assumptions about optimal cognitive states for BCI control.
This research advances brain-computer interfacing by establishing a robust framework for real-time 3D motion control, providing neurophysiological evidence for embodied feedback benefits, and identifying key human and training factors that influence performance. These findings offer new directions for developing personalised BCI training protocols and adaptive neural interfaces for rehabilitation and assistive technology applications.
| Date of Award | May 2026 |
|---|---|
| Original language | English |
| Supervisor | Darryl Charles (Supervisor), Karl Mc Creadie (Supervisor) & Damien Coyle (Supervisor) |
Keywords
- brain-computer interface
- EEG
- motor imagery
- virtual reality
- embodiment
- immersive feedback
- upper limb
- kinematic decoding
- motion trajectory decoding
- convolutional neural network
- long short-term memory
- deep learning
- functional connectivity
- sensorimotor cortex
- decoding accuracy
- neural engineering
- human factors
- user adaptation
- spatial cognition
Cite this
- Standard