Brain-Computer Interface for Lower-Limb Neurorehabilitation: From Signal Analysis to Practical Applications

Abstract

In recent years, the development of Brain-Computer Interfaces (BCIs) or Brain-Machine Intarfaces (BMIs) with Electroencephalography (EEG) has gained recognition in the scientific community for implementing robotic systems for rehabilitation. For instance, Motorized Mini Exercise Bikes (MMEBs) have been used for passive assistance with control driven by Motor Imagery (MI). However, these BCIs face challenges, such as long calibrations and low customization in applications. In addition, intentionality detection with EEG signals during pedaling tasks has not been fully explored. This dissertation aims to use different strategies on EEG signals for the detection of pedaling tasks using several algorithms. In addition, these methodologies approaches to implement real-time neurorehabilitation BCIs. For this, protocols with active pedaling, passive pedaling, and MI tasks were executed and different signal processing methodologies were addressed. Some processing methodologies were based on Common Spatial Patterns (CSP), Power Spectrum Density (PSD) or Riemannian Geometry, whereas Machine and Deep learning techniques, such as Linear Discriminant Analysis (LDA) or Extreme Learning Machine (ELM), were used here to classify EEG signals with accuracies close to 0.95 for MI, and 0.80 for active pedaling. Riemannian geometry-based methods were also used to identify MI tasks after passive pedaling at three different speeds (30, 45, and 60 rpm) with accuracies close to 0.78. As main contribution, a BCI was designed with visual neurofeedback, passive pedaling assistance, and MI, which was evaluated in the online phase, achieving an accuracy of approximately 0.80, and providing a feedback to the subject, aiming to encourage modulations. Subsequently, it was possible to observe the cortical response in the parieto-central cortex of the brain during the session. The results allow concluding that the implemented methodologies are feasible and accurate for the design of robotic lower limb BCIs that allow more personalized physical and neural neurorehabilitation and better human-machine interaction, which could help in the restoration of skills of people with neuromotor disabilities. The results presented here open the door to continue exploring brain information during the development of lower-limb tasks that may allow technological innovation in BCI systems for rehabilitation. Additionally, the proposed system can be used in therapeutic interventions for people with neuromotor impairments, such as post-stroke or spinal cord injury populations.