Real-time Strategies in Brain Computer Interfaces when Driven by Purely Visual Imagery Signals

Brain-computer interfaces (BCIs), when used for medical application, allow patients suffering from severely debilitating diseases to improve their life quality. A BCI translates brain signals in commands that can be used by the patient to drive several instruments, without the need of an electromyography (EMG) that is instead necessary in applications like myoelectric prosthesis. In fact, the BCI approach grants the possibility to interact with the external environment also to patients suffering from severe neuromuscular diseases, such as locked-in syndrome or advanced stages of multiple sclerosis. Currently, the most widespread technique used to monitor the brain signals in a BCI pipeline is electroencephalography (EEG), where a set of electrodes placed on the subject’s head measures the potential differences related to neural activity. The EEG driven BCIs can be categorized depending on the kind of signals used that they try to detect and classify. One of the most successful approaches present in literature is to use steady-state visually-evoked potential (SSVEP), signals that are generated when visual stimuli consisting of an image flickering at specific frequencies is shown to a subject. Indeed, when analyzing the frequency spectrum of SSVEP signals, it is possible to observe peaks located at frequencies that are related to the flickering frequency of the image. For this reason, SSVEP based BCIs can reach a high level of accuracy and a high bit-rate compared to alternative approaches. However, SSVEP signals present a series of issues, as for example the need to gaze at a screen for extended periods of time or the need to be able to concentrate on a specific image, both tasks that can prove difficult for patients with severe neuro-muscular impairment, the same patients that would benefit the most from using a BCI to drive instruments. In order to solve the just mentioned issues, in this work we focus on the usage of BCI based on visual imagery (VI) signals, a category of brain signals that has been rarely used to drive a BCI. Specifically, following our paradigm, we have the subject trying to visually imagine a flickering pattern, in order to elicit brain signals similar to the ones of SSVEP, but without the need for the user to gaze at a screen. The main novelty content of this work is the extension of an already tested VI based BCI to work online. In the previous implementation, the considered BCI was collecting EEG data of a user exercising on VI signals, and only in a second time classifying them. With the new extension it is possible to classify the VI signals in real time, opening the usage of the paradigm to numerous applications. In addition, the flexibility of VI-based BCIs has been further investigated by studying the usage of different frequencies and different patterns to imagine compared to the original study. The obtained results are encouraging, giving hope that VI-based BCIs could reach state-of-the-art performances also in terms of accuracy when compared to alternative EEG driven BCI approaches.