Resumen de Eeg-based brain-computer interfaces for detection of gait intention in healthy and stroke subjects

Andreea Ioana Sburlea

• Movement is the main way we can interact with the world. The motor system is involved in all types of movements, including speech, gestures, walking and many others. In particular, the ability to walk plays an essential role in daily life activities. Cortical injuries (e.g., stroke) can cause motor impairments and hinder the execution of these movements. Therefore, great effort is devoted to restore walking in people with motor impairments. Passive movements performed with the help of the therapist have shown effectiveness in recovery of walking patterns, however by actively participating in the therapy the patient can also induce plastic changes in the affected brain areas.

  For a deeper understanding of cortical involvement preceding walking and for the control of robotic devices using neural information, it is necessary to develop decoding models, which are capable of detecting the intention to walk from brain activity. Such models could facilitate the development of novel rehabilitation strategies in the future. One technology that enhances the detection of EEG neural correlates of movement intention and further employs them for the control of external devices is brain-computer interfaces (BCIs). EEG signals show an inherent variability among subjects and recording sessions. Therefore, EEG-based BCIs require a calibration period at the beginning of each session. However, since calibration is a time-consuming and tiring process the usability of EEG-based BCIs would benefit from reducing or even removing the recalibration time before new sessions or new subjects.

  This thesis addresses the usage of EEG-based BCIs for the detection of walking intention. To this end, we designed two self-paced walking experimental protocols, for healthy and stroke subjects. Next, we developed detection models based on temporal and spectral EEG features of walking intention and expanded our models to alternative features (instantaneous phase representation) which has a higher signal-to-noise ratio and anticipates movement sooner. Furthermore, we applied these models for detection within sessions, between sessions or between subjects, with small amounts of recalibration or without recalibration. By combining features we attained robust results in transfer cases (between sessions and subjects) in both healthy and stroke subjects, relative to the results obtained within sessions.