Miniaturization effects and node placement for neural decoding in EEG sensor networks

Electroencephalography (EEG) is a non-invasive neurorecording technique, which has the potential to be used for 24/7 neuromonitoring in daily life, e.g., in the context of neural prostheses, brain-computer interfaces, or for improved diagnosis of brain disorders. Although existing mobile wireless EEG headsets are a useful tool for short-term experiments, they are still too heavy, bulky and obtrusive, for long-term EEG-monitoring in daily life. However, we are now witnessing a wave of new miniature EEG sensor devices containing small electrodes embedded in them, which we refer to as Mini-EEGs. Mini-EEGs ideally consist of a wireless node with a small scalp area footprint, in which the electrodes, amplifier and wireless radio are embedded. However, due to their miniaturization, these mini-EEGs have the drawback that only a few EEG channels can be recorded within a small area. The latter also implies that the distance between the recording electrodes is small, leading to weaker and more focal EEG signals. Therefore, to capture more spatial information, a multitude of these mini-EEGs can be attached at relevant positions on the scalp forming a `wireless EEG sensor network’ (WESN). This thesis investigates whether optimizing such a WESN topology can compensate for miniaturization effects in a neural decoding task, with a main application focus on, auditory attention decoding (AAD). To this end, we devised the following strategy. First, starting from standard full-cap EEG data, several candidate mini-EEG nodes are emulated which locally collect EEG data with embedded electrodes separated by short distances. The node emulation is followed by a selection of a subset of these candidate nodes to form a WESN. The node selection to form a WESN makes use of channel selection methods for neural decoding which in itself is a commonly encountered problem in EEG. Since evaluating all possible channel combinations is usually infeasible, one usually has to settle for heuristic methods or convex approximations without optimality guarantees. We propose a utility-based greedy channel selection method and compare it against two popular approximate channel selection methods. First, a method based on the least absolute shrinkage and selection operator (LASSO) and second, a decoder magnitude-based greedy method. We show that our proposed utility-based greedy channel selection method performs significantly better than the other two approximate methods. Subsequently, this method is used for the node selection to form WESNs. AAD performance obtained using WESNs-consisting of nodes with electrodes separated by short distances-is compared against those using long-distance EEG recordings. We demonstrate with this comparison that the AAD performance using short-distance EEG measurements is comparable to those using an equal number of long-distance EEG measurements; if in both cases the optimal electrode positions are selected. Next, the performance gap in the context of AAD, between the selection made by the three approximate methods mentioned before and a truly optimal EEG channel selection is investigated and quantified. To this end, we reformulated the channel selection problem as a mixed-integer quadratic program (MIQP which allowed the use of efficient MIQP solvers to find the optimal channel combination in a feasible computation time for up to 100 candidate channels. In a context of AAD, we demonstrated that our proposed utility-based greedy channel selection does not show a significant optimality gap compared to optimal channel selection, whereas the other two methods do show a significant loss in performance. Furthermore, we demonstrate that our proposed MIQP formulation also provides a natural way to incorporate topology constraints in the selection, e.g., for electrode placement in WESNs with galvanic separation constraints, i.e. nodes that do not share an electrode. A combination of the utility-based greedy selection with an MIQP solver allowed us to perform a topology constrained electrode placement, even in large scale problems with more than 100 candidate positions. Having established that short-distance EEG measurements lead to similar AAD performance as long-distance benchmarks, the next goal was to identify and quantify the lower bound on miniaturization for EEG-based stimulus decoding. Previous studies were carried out using a lower (64) channel-density dataset, limiting the inter-electrode distances in emulated nodes to 3.5 cm. Therefore, we collected a new 255-channel EEG dataset in an AAD task. As opposed to previous studies with a lower channel density, this new high-density dataset allowed us to emulate mini-EEGs with inter-electrode distances down to 1 cm. We demonstrate that the performance remains reasonably stable for inter-electrode distances down to 3 cm, but decreases quickly for shorter distances, if the mini-EEG nodes can be placed at optimal scalp locations and orientations selected by a data-driven algorithm. With the help of this thesis, we established that WESN-like platforms allow achieving similar neural decoding performance as with long-distance EEG recordings, while adhering to the stringent miniaturization constraints for ambulatory EEG. We also demonstrate their applicability in an AAD task, which is the crucial ingredient for the design of neuro-steered auditory prostheses.