Simulating and classifying EEG scalp data for code-modulated visual evoked potential brain computer interfaces

Thumbnail Image
Issue Date
Journal Title
Journal ISSN
Volume Title
Objective. Within the field of brain-computer interface (BCI) research, a common challenge is the lack of ground truth data. Additionally, a BCI typically requires a calibration session in order to reach strong performances with new users, which increases the amount of time required to performexperiments. In this study, a simulation framework is implemented as a forward model which could generate EEG data artificially, in order to investigate whether simulated data could be used instead of empirical data when trainingmachine learning methods. Approach. Four versions of a forward model were implemented. Furthermore, two inverse models were implemented, namely a reconvolutionmodel using canonical correlation analysis and an EEGNet model. These inverse models were trained to classify the classes of c-VEP data, using either empirical data or simulated data, followed by a performance evaluation by testing on empirical data. Main results. By training inverse models on data generated with a forward model, the inverse models can learn to classify empirical c-VEP data to a certain extent which surpasses chance level. However, this performance is not yet at a level where it can be used in real life applications. Significance. This study shows that it is possible to achieve performances higher than chance level on empirical data by using machine learning techniques that were trained using simulated data. Further improvement of the forward models that were used in this study could potentially reduce the amount of empirical data that is needed in future studies within the field of VEP-based BCI.
Faculteit der Sociale Wetenschappen