Visual P300 Mind-Speller Brain-Computer Interfaces: A Walk Through the Recent Developments With Special Focus on Classification Algorithms

Abstract

Brain-computer interfaces are sophisticated signal processing systems, which directly operate on neuronal signals to identify specific human intents. These systems can be applied to overcome certain disabilities or to enhance the natural capabilities of human beings. The visual P300 mind-speller is a prominent one among them, which has opened up tremendous possibilities in movement and communication applications. Today, there exist many state-of-the-art visual P300 mind-speller implementations in the literature as a result of numerous researches in this domain over the past 2 decades. Each of these systems can be evaluated in terms of performance metrics like classification accuracy, information transfer rate, and processing time. Various classification techniques associated with these systems, which include but are not limited to discriminant analysis, support vector machine, neural network, distance-based and ensemble of classifiers, have major roles in determining the overall system performances. The significance of a proper review on the recent developments in visual P300 mind-spellers with proper emphasis on their classification algorithms is the key insight for this work. This article is organized with a brief introduction to P300, concepts of visual P300 mind-spellers, the survey of literature with special focus on classification algorithms, followed by the discussion of various challenges and future directions.

Robust Subject-Independent P300 Waveform Classification via Signal Pre-Processing and Deep Learning

“…Pre-processing P300 signals for classification requires extensive feature engineering, which can often be computationally intensive leading to slower response times [33]. In this work, we significantly reduce the pre-processing steps required for effective classification, and further, we completely eliminate domain transformations, which are typically used to learn salient P300 signal features.…”

“…Linear discriminant analysis (LDA) is the most common model used for P300 classification [33] The objective is to determine the classification probability, P (Y = y|X = x), of each input sample, x ∈ R d , where the output is given by y ∈ {0, 1} (corresponding to non-target and target classes, respectively) using Bayes' Theorem. Specifically, this is formulated by…”

A P300 Brain-Computer Interface With a Reduced Visual Field

“…Following the introduction of P300 speller in 1988 by Farwell and Donchin (1988), many studies have strived to improve this method (Fazel-Rezai et al, 2012;Allison et al, 2020;Philip and George, 2020). The speller performance is hindered by the necessity to run many trials to distinguish target and non-target stimuli based on the comparison of the event-related potentials (ERPs) they evoke.…”

“…Particularly, non-target items that are adjacent to the target attract attention and interfere with the decoding performance (Fazel-Rezai, 2007;Townsend et al, 2010). Several solutions to this problem have been explored, including using flashes of single items instead of flashing rows and columns (Guger et al, 2009), rearranging the spatial configuration of the simultaneously flashing stimuli (Townsend et al, 2010), suppressing the stimuli adjacent to targets (Frye et al, 2011), or all non-targets (Shishkin et al, 2011) during the calibration procedure, and optimizing the characteristics of visual stimuli (Salvaris and Sepulveda, 2009;Jin et al, 2017;Mainsah et al, 2017;Philip and George, 2020). Yet, all these approaches require a considerable amount of distracting stimuli for accurate spelling, which slows the decoding, and causes user fatigue (Boksem et al, 2005;Käthner et al, 2014;Oken et al, 2018).…”

A Spelling Paradigm With an Added Red Dot Improved the P300 Speller System Performance

“…Several studies have attempted to improve the spelling accuracy and speed of the P300 speller. However, its performance is still unable to meet the requirements of a real-world application (Kaufmann et al, 2011;Aya et al, 2018;Philip and George, 2020;Xu et al, 2020).…”