Abstract
Brain-computer interfaces (BCI) have the potential to improve and enhance our lives, enabling us to control devices with our thoughts and intentions. Currently, BCIs primarily aim to benefit a small patient population by, for example, providing an alternative communication channel for severely paralyzed people that are not able to speak,
... read more
or enabling control over a prosthetic limb. While the current BCIs already deliver impressive results, the degree of control and the practical (daily) use is very limited.
This thesis contributes to advancing brain-computer interfacing by exploring the optimal utilization of brain activity in a diversity of brain functions and regions. Here, I consider ECoG as the primary candidate for a daily home-use BCI and leverage fMRI measurements to perform fundamental research that is transferable to an ECoG-based BCI.
Within the paradigm of sensorimotor BCIs, we showed that electrophysiological representations of hand movements are preserved after long-term amputation of the contralateral arm. These unique ECoG measurements confirm distinguishable finger representations, fast temporal onset and sustained decodability over time from just a small piece of the cortex. Such properties are encouraging for utilization in a BCI, in particular when considering a clinical need to drive neural prosthetic devices or communication devices. In utilizing sensorimotor activity with ECoG, it is essential that the most relevant information is captured to provide optimal control and reliability to a BCI. Using fMRI data-driven simulations in healthy participants, we show how the design (i.e. physical characteristics) and precise surgical placement of the electrode grid determine, to a large extent, how well neural information on hand movements is captured. Optimal results can already be achieved with a few well-placed electrodes, being beneficial in terms of battery life, cost of implantable amplifiers, real-time processing and clinical invasiveness. Our results strongly argue in favor of fMRI data-driven simulations to be used pre-operatively to ensure the most optimal control over an implanted BCI later. Venturing to other brain areas and functions, we investigated the use of visual imagery and the decoding of imaged characters. We demonstrate that imagined characters can be decoded from the early visual cortex with minimum training, thus providing an intuitive way of per-letter spelling in a BCI. Prolonged imagery did not result in better classification, implying that prolonged imagery does not add to the quality of the information present in early visual areas. However, prolonged imagery did extend decodability over time, suggesting sustained neuronal activity that can be utilized for BCI purposes. Finally, we explore the left dorsolateral prefrontal cortex (DLPFC) as a working memory area that can be voluntarily controlled and provide input to a BCI when motor representations are impaired. Control over the activity in the DLPFC was shown to be achieved within a short period of time when provided with closed-loop neurofeedback. In addition, we show that a potential practice effect (i.e. automatization of the task) and the associated decrease of activity in the WM network can be negated by providing closed-loop feedback, so the reliability of the DLPFC input signal is preserved.
show less
