Coupling Measures for BCIs

The submitted project with the running title "Coupling measures for BCIs" aims to analyze the suitability of novel features for brain-computer interfaces (BCIs). These features describe the relationships or coupling between single EEG signals. In contrast, conventional univariate features such as the band power totally ignore this information. This project is separated into two parts. The first part focusses on synchronized BCIs, where the time frames for controlling an application with thoughts are established by the computer (sometimes, these systems are also called "cue-based"). Different coupling measures that can all be derived from a multivariate autoregressive model will be analyzed and their suitability for a BCI will be assessed. To that end, an offline study will be conducted on existing data with the goal of comparing the performance of those features with classical univariate ones. Moreover, both feature types will be combined in order to find out whether the classification accuracy can be improved. Finally, the findings from this study will be applied to online experiments with feedback with several subjects. The second part is dedicated to self-paced BCIs (sometimes also called "asynchronous"), where the users can freely decide when to control the system. The additional challenge presented with this paradigm lies in the ability to detect the state when the users do not want to control the system (the so-called no control state). The various coupling measures will be analyzed whether they permit to distinguish this state from a control state. The first step will be once again an offline analysis, this time with the data already recorded in the first part of this project. Once more, coupling measures will be compared with univariate features and the combination of both types. The findings will be validated with another online study with feedback involving several subjects.

The submitted project with the running title "Coupling measures for BCIs" aims to analyze the suitability of novel features for brain-computer interfaces (BCIs). These features describe the relationships or coupling between single EEG signals. In contrast, conventional univariate features such as the band power totally ignore this information. This project is separated into two parts. The first part focusses on synchronized BCIs, where the time frames for controlling an application with thoughts are established by the computer (sometimes, these systems are also called "cue-based"). Different coupling measures that can all be derived from a multivariate autoregressive model will be analyzed and their suitability for a BCI will be assessed. To that end, an offline study will be conducted on existing data with the goal of comparing the performance of those features with classical univariate ones. Moreover, both feature types will be combined in order to find out whether the classification accuracy can be improved. Finally, the findings from this study will be applied to online experiments with feedback with several subjects. The second part is dedicated to self-paced BCIs (sometimes also called "asynchronous"), where the users can freely decide when to control the system. The additional challenge presented with this paradigm lies in the ability to detect the state when the users do not want to control the system (the so-called no control state). The various coupling measures will be analyzed whether they permit to distinguish this state from a control state. The first step will be once again an offline analysis, this time with the data already recorded in the first part of this project. Once more, coupling measures will be compared with univariate features and the combination of both types. The findings will be validated with another online study with feedback involving several subjects. Im beantragten Projekt mit dem Kurztitel "Kopplungsmaße für BCIs" geht es darum, neuartige Merkmale für Brain-Computer Interfaces (BCIs) auf ihre Eignung zu testen. Diese Merkmale beinhalten die Beziehungen oder Kopplungen zwischen einzelnen EEG-Signalen, ganz im Gegensatz zu den bisher weit verbreiteten univariaten Merkmalen (wie beispielsweise die Bandleistung).