Abstract: Nearly a century has elapsed since the discovery of the electroencephalogram (EEG) by Hans Berger. In this time, electroencephalography (EEG) has installed itself as clinical standard for the diagnosis of multiple neurological disorders as well as a relatively cheap and versatile research tool. Over the last 90 years, EEG has continuously been improved by researches, clinicians and engineers. One notable improvement, on which this thesis builds, is the use of an ever-increasing number of spatial measurement points (electrodes). EEG started being measured with only a handful of electrodes targeting specific brain locations. Clinicians expanded the EEG montage to ~20 electrodes covering the whole scalp and defined nomenclatures and positions in the 10-20 system. While the 10-20 system is still clinical standard, in research this system has been extended to ~74 electrodes (10-10 system) and >300 electrodes (10-5 system). The push for more electrodes has been driven by the advent of modern day computing making it possible to digitize and store large amounts of data, combined with new mathematical methods capable of disentangling the sources of the ongoing brain activity and localizing them in the brain, namely source localization. Increasing computational power made it progressively possible to refine the level of detail included in the head models and the forward solutions underlying the source localization (also called inverse solution). Including these details made the forward solutions, and thus the inverse solutions too, more accurate and create the possibility of investigating the effect of individual anatomy on electromagnetic volume conduction. Despite the progress of EEG and source localization, EEG still not close to the gold standard for measuring and localizing neuronal activity in the brain. The gold standard is still to measure the activity directly on the brain surface or even directly within the brain. These methods are commonly called invasive or intracranial EEG (both iEEG). Because of the superior spatial and signal resolution of iEEG, these methods have seen relatively few attempts at improvements. Thus, the aim of this thesis was to further improve the spatial and signal resolution EEG to drive it as close as possible to iEEG, and to improve iEEG itself. The improvements achieved during this thesis are reflected both in the write-up of the thesis as well as in publications and manuscripts available online. Some of these were not included in the write-up in an attempt to keep the scope focused on one topic, namely the spatial simulation of EEG and iEEG. For methodological and experimental improvements the interested reader is referred to my scholar profile. In the following the results included in the write-up are briefly presented. In chapter V, a shortcoming of electrocorticography (ECoG), an iEEG method, is addressed. In EEG, the description of the spatial, temporal and frequency characteristics of the electric activity of muscles (electromyogram, electromyography, EMG) contaminating EEG has been a necessity to separate it from brain activity. That ECoG, and iEEG in general, is also contaminated by EMG, albeit to a lesser extent than EEG, has been largely disregarded. Thus, a first step to improve iEEG is to create guidelines to identify EMG and differentiate it from brain activity. One major source of EMG artifacts in iEEG recordings is the chewing musculature. In chapter V we therefore describe the spatial, temporal and frequency characteristics of chewing-related (ChR) EMG signals recorded with ECoG. We show that ChR EMG can be differentiated from brain activity based on spatial, temporal and frequency characteristics. Its spatially wide distribution, which bridges anatomical borders, is in stark contrast to the mostly focal spatial extend of brain activity, which is also restricted to cortical structures with relatively sharp borders. The repetitive temporal characteristic of ChR EMG can also be used to differentiate it from brain activity, which is usually either transient or sustained. Lastly, the frequency profile of ChR EMG has a much more broadband energy distribution than brain activity, which is distributed across different narrow frequency bands, some of which behave in an anti-correlated fashion. The insights won in chapter V can now be applied to either discard iEEG data contaminated by EMG or to design methods to attempt to clean the contaminated data, improving the quality and yield of iEEG. In chapter VI, we present results based on the data described in chapter V extended using volume conductor simulations. The results suggest that the cortical electric fields occurring during strong mastication could be steep enough to entrain neuronal activity. This claim is based on multiple recent findings showing that neurons passively influence the membrane potential of their neighbors because of the changing ionic concentration of the extracellular medium during ongoing activity. This influence creates synchronous fluctuations of the membrane potential of local neuronal populations, making the whole population synchronously more, or less, receptive to incoming information. We argue that this bidirectional mechanism is potentially active in a unidirectional fashion, from ChR muscles to cortical neurons, during strong mastication. Based on the current literature we discuss the possibility that this interaction might influence our cognition on both short-term and long-term timescales. In chapter VII, we improve our understanding of the influence of mm-scale anatomical detail on EEG volume conduction using cutting edge imaging data and simulations. Although anatomy is increasingly being taken into account when simulating volume conduction, many details are being disregarded under the assumption that their influence on volume conduction is negligible. We show that blood vessels, which are quite difficult to model and have thus up to now always been disregarded, do have a non-negligible influence of volume conduction. Source localization errors > 20 mm could be observed in the vicinity of the major cerebral arteries. While minor arteries did not skew source localization considerably ( 5 mm), in densely vascularized brain areas, like the insula and the medial temporal lobe, spatial accumulation of minor arteries also led to source localization errors 20 mm. These brain areas are targets of clinical source localization and play important roles for the normal function of the brain. Clinical EEG source localization could thus be further improved if more attention would be payed to small details. Disregarding seemingly negligible details could potentially bear large consequences and should thus be investigated as soon as methodologically possible. While the accuracy of most models is sufficient for current research questions, emerging topics like layer specificity of neuronal activity do need a sub-mm localization accuracy. In chapter VIII, we show that increasing the spatial sampling of ECoG leads to significant improvements in the spatio-frequential content of recordings. In last couple of decades, spatial sampling of EEG has been drastically increased. The spatial sampling of iEEG however, because of clinical certification constraints, has stayed constant. The resent push in research to improve ECoG has triggered the development of μECoG where contact size and intervals are in the μm range. In chapter VIII we not only show that μECoG has a richer spatio-frequential content than standard ECoG, but also that the spatio-frequential content further increases the smaller the μECoG grids get. Combined with recent findings that neuronal spikes can be recorded from the cortical surface, our findings put in question the traditional assumptions regarding the limitations of the optimal spatial sampling of EEG and iEEG recordings. In chapter IX, we present results based on the data described in chapter VIII, extended using volume conductor simulations, showing that μECoG recordings have the potential to be used to perform cortical layer specific source localization of neuronal activity. The laminar or laminated recordings currently used to investigate the laminar specificity of neuronal activity are a tedious and technically challenging procedure. Moreover, such recordings are, in the clinical context of medically intractable epilepsy, which is the main source of iEEG data in humans, ethically questionable and bear strong regulatory constraints. The medical and scientific prospect of having access to this information using superficial recordings performed in the clinical context is thus huge. Our results show that it is possible to generate a frequency-resolved depth mapping of μECoG activity matching the current penetrating electrode literature using simple methods based on spatial properties. This represents the first step towards layer specific source localization of neuronal activity and leads the way for further analysis using more sophisticated methods. Concluding, this thesis shows that volume conductor simulations of electromagnetic potentials are a potent tool for improving EEG and exploring new possibilities. Simulations can be used to localize the sources of the measured activity, to infer data at point where no measurement took place, to investigate how the measured data is influenced by the underlying anatomy and to test new ideas and hypotheses. Although the methodological and experimental aspects are crucial for improving EEG, simulations also opened the doors for further improvements. Some of which, as described in this thesis, could probably not have been envisioned without the use of simulations

Abstract: The aim of this bachelor thesis was to study EEG correlates of inner speech. In particular, I want to know if it is possible to distinguish different semantic categories when the subjects speak the words with an inner voice. It would be a great opportunity for brain-computer- interfaces (BCI) if the computer could decode from the EEG-recordings what the patient wishes, be it food or the realisation of an action directly from the specific thought and not through surrogate thoughts. Many recent studies showed with neuroimaging technics that different semantic classes induce activations in different cortical areas. A word does not only have a semantic meaning but also evokes related associations, which are processed in dif- ferent cortical regions. As different categories have different associations, various brain re- gions are activated. Besides, these differences in EEG-correlates between semantic classes have not yet been described for inner speech. Ten subjects repeated 112 words from the categories 'animals', 'food', 'tools' and 'pseudowords' for three seconds in their minds. 'Tools' is expected to induce stronger activation over the motor cortex, 'animals' is expected to in- duce more activation over the visual cortex and 'food' in cortical regions related to gustatory or olfactory associations. 'Pseudowords' are used as second baseline to detect activation in brain areas in addition to the language system. The other baseline was a neutral condition before the stimulus occurred. Unfortunately, there are no differences between the semantic categories. All visible effects appear in all semantic classes and were evoked by the visual stimulus or the working memory. Possible reasons for this result are discussed and im- provements for the experimental design are proposed

Abstract: As autonomous service robots become more affordable and thus available for the general public, there is a growing need for user-friendly interfaces to control these systems. Control interfaces typically get more complicated with increasing complexity of the robotic tasks and the environment. Traditional control modalities as touch, speech or gesture commands are not necessarily suited for all users. While non-expert users can make the effort to familiarize themselves with a robotic system, paralyzed users may not be capable of controlling such systems even though they need robotic assistance most. In this paper, we present a novel framework, that allows these users to interact with a robotic service assistant in a closed-loop fashion, using only thoughts. The system is composed of several interacting components: non-invasive neuronal signal recording and co-adaptive deep learning which form the brain-computer interface (BCI), high-level task planning based on referring expressions, navigation and manipulation planning as well as environmental perception. We extensively evaluate the BCI in various tasks, determine the performance of the goal formulation user interface and investigate its intuitiveness in a user study. Furthermore, we demonstrate the applicability and robustness of the system in real world scenarios, considering fetch-and-carry tasks and tasks involving human-robot interaction. As our results show, the system is capable of adapting to frequent changes in the environment and reliably accomplishes given tasks within a reasonable amount of time. Combined with high-level planning using referring expressions and autonomous robotic systems, interesting new perspectives open up for non-invasive BCI-based human-robot interactions

Abstract: Error detection in motor behavior is a fundamental cognitive function heavily relying on local cortical information processing. Neural activity in the high-gamma frequency band (HGB) closely reflects such local cortical processing, but little is known about its role in error processing, particularly in the healthy human brain. Here we characterize the error-related response of the human brain based on data obtained with noninvasive EEG optimized for HGB mapping in 31 healthy subjects (15 females, 16 males), and additional intracranial EEG data from 9 epilepsy patients (4 females, 5 males). Our findings reveal a multiscale picture of the global and local dynamics of error-related HGB activity in the human brain. On the global level as reflected in the noninvasive EEG, the error-related response started with an early component dominated by anterior brain regions, followed by a shift to parietal regions, and a subsequent phase characterized by sustained parietal HGB activity. This phase lasted for more than 1 s after the error onset. On the local level reflected in the intracranial EEG, a cascade of both transient and sustained error-related responses involved an even more extended network, spanning beyond frontal and parietal regions to the insula and the hippocampus. HGB mapping appeared especially well suited to investigate late, sustained components of the error response, possibly linked to downstream functional stages such as error-related learning and behavioral adaptation. Our findings establish the basic spatio-temporal properties of HGB activity as a neural correlate of error processing, complementing traditional error-related potential studies