Design and development of a clinical tool based on brain-computer interface (BCI) technology for motor rehabilitation
Motor impairments resulting from neurological disorders such as stroke often lead to significant disability, requiring intensive rehabilitation. Depending on the severity of the brain lesion, some patients may not qualify for conventional physical therapy due to limited residual motor function. Brai...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/430889 |
| Acceso en línea: | https://hdl.handle.net/2117/430889 |
| Access Level: | acceso abierto |
| Palabra clave: | Brain-computer interfaces Electroencephalography Machine learning Interfícies cervell-ordinador Electroencefalografia Aprenentatge automàtic Àrees temàtiques de la UPC::Enginyeria biomèdica::Electrònica biomèdica |
| Sumario: | Motor impairments resulting from neurological disorders such as stroke often lead to significant disability, requiring intensive rehabilitation. Depending on the severity of the brain lesion, some patients may not qualify for conventional physical therapy due to limited residual motor function. Brain-Computer Interfaces (BCIs), particularly those based on Motor Imagery (MI), offer a promising alternative for supporting neurorehabilitation in severely impaired individuals. MI has been proven to elicit brain activation patterns similar to actual movement, making it especially useful in cases where overt motor execution is not possible. This project aims at designing and developing a functional framework for an MI-based BCI capable of classifying left versus right hand motor imagery and delivering real-time neurofeedback. EEG signals were acquired from eleven healthy subjects during a standardized MI task. The data were used to implement an offline processing pipeline that included signal preprocessing, feature extraction using Common Spatial Patterns (CSP) algorithm, dimensionality reduction through feature selection, and classification using various supervised learning algorithms. Classification performance varied considerably across individuals, with 7 out 11 subjects achieving accuracies above 80%, and the best subject reaching 91.6%, while others scores close to chance level. These results confirm both the feasibility of the proposed system and the strong inter-subject variability often observed in MI-BCI applications. An online prototype was also developed, integrating real-time EEG acquisition, CSPbased feature extraction, classification, and visual feedback. Although online testing was limited to a small number of sessions due to time constraints, the system demonstrated its capacity to detect and respond to MI tasks in real time. Two additional BCI applications based on blink detection were implemented as demonstrators, showing high responsiveness and positive user feedback in public science outreach events. Despite promising results, the system’s generalization remains limited, and the lack of data from motor-impaired individuals represents a noteworthy constraint. Future work should focus on improving the robustness and adaptability of the real-time system through extended calibration, adaptive learning, and testing with clinical populations. With continued development and validation, the proposed MI-BCI framework holds potential as a cost-effective and accessible tool for neurorehabilitation. |
|---|