Towards SSVEP-based BCI applications for real-world environments

Abstract

Brain-computer interfaces (BCIs) provide direct communication and control of applications through brain signals. This work focuses on non-invasive BCIs that use electroencephalography (EEG) to record brain signals and are based on the neural response known as steady-state visually evoked potentials (SSVEPs). SSVEPs are electrical potentials evoked in response to a repetitive visual stimulus flickering at a specific frequency. This neural response consists of oscillatory activity at the fundamental frequency and harmonics of the visual stimulus. An SSVEP-based BCI application exploits this response by uniquely associating visual stimuli flickering at specific frequencies, to specific commands, which are presented to the user who may select a command by attending to the corresponding stimulus. The BCI identifies the SSVEP response in the EEG signal and translates it into a particular signal to control the application. In the past few decades SSVEP-based BCIs have improved significantly with vast research aimed at enhancing their performance. However, several limitations to their practicality and user experience are still evident that these restrict the use of this technology in real-world environments. This work investigates and proposes innovative solutions to address these challenges in a collection of studies towards practical SSVEP-based BCI applications. These include the presentation of a benchmark to evaluate the signal quality of EEG acquisition devices for SSVEP-based BCIs; the development of a double blink mechanism as a solution to address the annoyance of the flickering stimuli and to reduce the execution of unwanted BCI actions; an investigation on the performance of SSVEP-based BCIs in an uncontrolled environment; an investigation of the demanding training requirements to use such systems; and the invention of a switch-and-train (SAT) framework as a novel SSVEP detection method to reduce this training time. This work also investigates the use of a novel probabilistic autoregressive modelling framework and its extension to an autoregressive switching multiple model (AR-SMM) framework for the detection of SSVEPs in BCI applications. The benefit of the developed approaches for this type of application, as opposed to the standard feature extraction and classification techniques used in SSVEP-based BCIs, is that minimal training data is required, a minimum of two electrodes are necessary, a probabilistic decision for classification is provided which can be used as a measure of certainty in the decision making process, and the ability to discriminate between rest and control states. Furthermore, the AR-SMM framework provides classification on a sample-by-sample basis that is shown to lead to a faster detection of SSVEPs, shorter flickering time and improved BCI information transfer rate.