Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of GraphSearch.
Publication
Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of GraphSearch.
By directly analyzing brain activity, Brain-Computer Interfaces (BCIs) allow for communication that does not rely on any muscular control and therefore constitute a possible communication channel for the completely paralyzed. Typically, the user performs different mental tasks, that correspond to different output commands as recognized by the system. From the recorded brain signals (Electroencephalogram, EEG), features that characterize the mental tasks and allow their discrimination by a classifier have to be extracted. This dissertation addresses the extraction of features in the framework of BCIs. On the one hand, new features are proposed. On the other hand, feature selection algorithms are investigated in order to select relevant features. Currently existing BCIs mostly use power estimates in some pre-defined frequency bands, which are single-channel features. Some authors report on the use of multichannel features, but interactions between specific brain regions have not yet been studied. We propose to use the synchronization feature Phase Locking Value (PLV) for the classification of spontaneous EEG recorded during different mental tasks. It is fast to compute and can be applied to relatively short time windows, two important assets for BCI applications. In a first instance, average synchronization values are considered. Tests on offline data show that significant classification accuracies can be obtained by the sole use of PLV. This demonstrates the relevance of synchronization features for the classification of EEG in this context. We found that PLV and power features do not clearly outperform each other, but their combination often leads to significantly improved results and never significantly deteriorates the classification accuracies obtained by the separate subsets. In the next step, feature selection algorithms are investigated in order to select the most interesting features. We show that Genetic Algorithms (GAs) as well as SVM-based recursive feature elimination (SVM-rfe) select physiologically meaningful features. As they are slow (computation times on the order of days and hours respectively) and thus cannot practically be used for BCIs, a modified version of the Fast Correlation-Based Filter (FCBF) is proposed. In this study, FCBF generalizes well and achieves good classification accuracies with very few features. The correspondence of the selected EEG signals with neurophysiological evidence is even stronger than for GAs and SVM-rfe. In addition, this algorithm is fast (computation time on the order of minutes) and so it can be applied between two recording sessions. Comparing the classification results obtained with broadband and narrowband power and PLV features selected by FCBF learns that power features are preferably computed in the narrower 8–12Hz frequency band and that for PLV features, the 8–30Hz frequency band is the better one. Furthermore, FCBF is used to evaluate a set of features comprising, on the one hand, the PLV for all possible electrode pairs, PLV averages, the cluster participation index and power features (band power and statistical mean frequency (SMF)), computed from the EEG signals. On the other hand, power (total power and SMF) and synchronization features derived from the empirical mode decomposition of these signals were included. They all proved useful. It is recommended to initially consider power features and only the PLV averages, because the resulting set of features is substantially smaller than when using PLV for all possible electrode pairs. Automatic detection of artifacts and rejection of the corresponding data window for selecting the features and training the classifier, results in, if any, an increase of classification accuracy that is generally not greater than 1%, for our data. Application of global and local PCA-based denoising techniques never yielded improved results for two of the five subjects we analyzed. If it yields better results for the other 3 subjects, the increase never exceeds 1.84%. Application of the proposed methods to blind data of the third international BCI competition held in 2005 gave significant classification accuracies. This further illustrates the relevance and the potential of the investigated techniques. We conclude that PLV may complement currently used features and improve future BCI systems. PLV is well suited for BCI applications, because of its fast computation, needed for online feedback systems. The Fast Correlation-Based Filter is a valuable tool for evaluating features and selecting a subset of relevant features.
Loading
Loading
Loading
Loading
Loading
black-box' generative and discriminative approaches. In order to take potential advantage of the temporal nature of the EEG, we use two temporal models: the standard generative hidden Markov model, and the discriminative input-output hidden Markov model. For this latter model, we introduce a novel apposite' training algorithm which is of particular benefit for the type of training sequences that we use. We also asses the advantage of using these temporal probabilistic models compared with their static alternatives. We then investigate the incorporation of more specific prior information about the physical nature of EEG signals into the model structure. In particular, a common successful assumption in EEG research is that signals are generated by a linear mixing of independent sources in the brain and other external components. Such domain knowledge is conveniently introduced by using a generative model, and leads to a generative form of Independent Components Analysis (gICA). We analyze whether or not this approach is advantageous in terms of performance compared to a more standard discriminative approach, which uses domain knowledge by extracting relevant features which are subsequently fed into classifiers. The user of a BCI system may have more than one way to perform a particular mental task. Furthermore, the physiological and psychological conditions may change from one recording session and/or day to another. As a consequence, the corresponding EEG signals may change significantly. As a first attempt to deal with this effect, we use a mixture of gICA in which the EEG signal is split into different regimes, each regime corresponding to a potentially different realization of the same mental task. An arguable limitation of the gICA model is the fact that the temporal nature of the EEG signal is not taken into account. Therefore, we analyze an extension in which each hidden component is modeled with an autoregressive process. The second part of the thesis focuses on analyzing the EEG signal and, in particular, on extracting independent dynamical processes from multiple channels. In BCI research, such a decomposition technique can be applied, for example, to denoise EEG signals from artifacts and to analyze the source generators in the brain, thereby aiding the visualization and interpretation of the mental state. In order to do this, we introduce a specially constrained form of the linear Gaussian state-space model which satisfies several properties, such as flexibility in the specification of the number of recovered independent processes and the possibility to obtain processes in particular frequency ranges. We then discuss an extension of this model to the case in which we don't know a priori the correct number of hidden processes which have generated the observed time-series and the prior knowledge about their frequency content is not precise. This is achieved using an approximate variational Bayesian analysis. The resulting model can automatically determine the number and appropriate complexity of the underlying dynamics, with a preference for the simplest solution, and estimates processes with preferential spectral properties. An important contribution from our work is a novel `sequential' algorithm for performing smoothed inference, which is numerically stable and simpler than others previously published.