Transition from Visually Evoked to Purely Imagined Steady State Potentials in Brain-Computer Interfaces: Computational Schemes and Feasibility Assessments

Brain-computer interfaces (BCIs) are systems that allow the translation of the brain activity into commands to control a dedicated application without any muscular activity. For example, BCIs provide patients suffering from amyotrophic lateral sclerosis (ALS), locked-in syndrome, or paralysis, the capability to interact with their environment. The main building blocks of a BCI are (1) the signal acquisition, (2) the pre-processing and signal enhancement, (3) the feature extraction and classification, and (4) the translation of the features into commands. Between all available options for cerebral activity recordings, one of the most widespread choices is the electroencephalography (EEG), a non-invasive technique which records potential differences caused by the neural activity through electrodes placed on the scalp of the subject. Each BCI implementation depends on a characteristic paradigm that exploits distinct brain signals. Between the different EEG-based BCIs, some of the most robust rely on steady-state visually evoked potentials (SSVEPs), that are elicited by external visual stimuli flickering at specific frequencies. In these approaches, the subject focuses his sight on one of several visual cues flashing at their own unique frequency. The BCI system will then determine which stimulus the patient is observing by analyzing the EEG scalp recordings in the frequency domain and identifying peaks at the specific rate of the different stimuli and at their higher order harmonics. The BCI scheme will then translate this information into a predetermined command -- one for each stimulus -- that it will execute. This paradigm allows consistently high accuracy and has one of the highest bit-rates among all the other BCIs present in literature. However, since the subject needs to fix his attention on a screen, there are potential drawbacks for this technique. For example, it is challenging for patients with severely impaired vision or incapable of moving their eyes, and it may cause tiredness of the user. Hence, the aim of this work is to create a novel protocol for a BCI system driven by purely imagined SSVEP-eliciting stimuli, namely visual imagery (VI), examining whether mental reproductions of flashing images could be adopted to guide a robust and reliable BCI. This model may solve the aforementioned issues related to SSVEPs, allowing a wider range of patients to drive a BCI. Our implementation of this new paradigm showed noticeable performances in preliminary experiments, even when adopted in multiple classes sessions. Although the accuracy is not at the same level as classical SSVEPs, there is ample space for improvements. One direction that has been explored, which has already shown encouraging results, is the use of spatial filters based on the resolution of inverse problems, designed for this specific application. The filters allow the diminution of the noise components of the signals used for classification, leading to a higher accuracy of the overall system. This promising setting will be explored and enhanced in future works, with the hope to achieve state-of-the-art performances of EEG-based BCIs.