Implementation of a framework to characterize electroencephalographic correlates of walking

This thesis investigates the application of machine learning algorithms for the prediction of human gait kinematics utilizing electroencephalogram (EEG) signals, with an initial focus on healthy subjects to evaluate the potential of EEG data in estimating movement during ambulation. Specifically, the research aims to predict the angular dynamics of the joints by employing both EEG signals in isolation and integrating a limited number of kinematic samples. The accuracy of these predictions is assessed in terms of joint angle precision within a defined tolerance, as well as the model's capacity to differentiate between the stance and swing phases, which are crucial for a comprehensive understanding of the gait cycle. The analysis of EEG data obtained from healthy subjects revealed several significant limitations. The EEG signal is highly sensitive to motion artifacts, mechanical disturbances, and muscular interference, complicating the precise extraction of neural activity specifically associated with motor control during gait. Furthermore, the repetitive and cyclic nature of gait performed at a constant speed suggests that the predictability of movement may stem not solely from EEG data, but rather from the inherent periodicity of the motor action itself. This raises pertinent questions regarding the actual informational content of EEG in the context of constant-speed walking, indicating that EEG may offer greater value in scenarios characterized by speed variations or alterations in gait dynamics. In light of these observations, this thesis posits that a promising direction for future research may involve the expansion of experimental protocols to encompass more complex scenarios, such as walking with variations in speed, initiating and ceasing ambulation, as well as navigating inclines and declines. Such situations may activate EEG signals that are more relevant to the regulation of motor control. Additionally, further experimentation could include individuals utilizing robotic-assisted gait (RAG) technologies and patients with spinal cord injuries, thereby assessing the ability of EEG to capture predictive signals in rehabilitation contexts or while utilizing assistive devices, in which EEG may play a pivotal role in facilitating mobility. The findings of this research serve as a foundational step toward the development of novel applications of machine learning in the prediction of gait based on EEG signals, laying the groundwork for future investigations that incorporate more complex gait scenarios and clinical conditions. Ultimately, this work aims to expand the potential applications of EEG signals within rehabilitative and assistive contexts.