Improving P300 Speller performance by means of optimization and machine learning

Abstract

Brain-Computer Interfaces (BCIs) are systems allowing people to interact with the environment bypassing the natural neuromuscular and hormonal outputs of the peripheral nervous system (PNS). These interfaces record a user’s brain activity and translate it into control commands for external devices, thus providing the PNS with additional artificial outputs. In this framework, the BCIs based on the P300 Event-Related Potentials (ERP), which represent the electrical responses recorded from the brain after specific events or stimuli, have proven to be particularly successful and robust. The presence or the absence of a P300 evoked potential within the EEG features is determined through a classification algorithm. Linear classifiers such as stepwise linear discriminant analysis and support vector machine (SVM) are the most used discriminant algorithms for ERPs’ classification. Due to the low signal-to-noise ratio of the EEG signals, multiple stimulation sequences (a.k.a. iterations) are carried out and then averaged before the signals being classified. However, while augmenting the number of iterations improves the Signal-to-Noise Ratio, it also slows down the process. In the early studies, the number of iterations was fixed (no stopping environment), but recently several early stopping strategies have been proposed in the literature to dynamically interrupt the stimulation sequence when a certain criterion is met in order to enhance the communication rate. In this work, we explore how to improve the classification performances in P300 based BCIs by combining optimization and machine learning. First, we propose a new decision function that aims at improving classification performances in terms of accuracy and Information Transfer Rate both in a no stopping and early stopping environment. Then, we propose a new SVM training problem that aims to facilitate the target-detection process. Our approach proves to be effective on several publicly available datasets.

Cortico-muscular coupling to control a hybrid brain-computer interface for upper limb motor rehabilitation: A pseudo-online study on stroke patients

“…For this reason, identifying the parameters that allow the best trade-off between classification performances and speed is crucial. Over the past decades, many studies have explored feature extraction and classification approaches to improve the accuracy ( Bianchi et al, 2022 ), raise the number of commands ( Xu et al, 2020 ), increase the information transfer rate and reduce the calibration time ( Wong et al, 2020 ; Yao et al, 2022 ). P300-based speller and steady-state visual evoked potential-based BCIs have mainly taken advantages from those methodological improvements ( Nakanishi et al, 2018 ; Xu et al, 2020 ) in order to avoid patient frustration caused by false and delayed detections ( Eliseyev et al, 2021 ).…”

Artificial Intelligence Algorithms in Visual Evoked Potential-Based Brain-Computer Interfaces for Motor Rehabilitation Applications: Systematic Review and Future Directions

“…Hence, the importance of developing and studying visual BCI systems for these applications or combine them with motor paradigms, like the ones found on these SLR ( Horki et al, 2011 ; Choi et al, 2016 ). For P300 BCIs, multiple repetitions (5 or more) of the whole stimuli sequence are typically needed to predict accurately the user’s choice ( Bianchi et al, 2021 ). Depending on the number of possible targets and interstimulus interval, the selection time for a single command can be relatively slow (tens of s) ( Mainsah et al, 2015 ).…”

Optimized Collaborative Brain-Computer Interfaces for Enhancing Face Recognition

“…Alternatively, early stopping techniques can be used to detect when the decoder is confident enough about its predictions and stop the stimulation. A reliability score can be computed after a preliminary calibration phase, so that each neural response can be assigned a different weight, thus varying its contribution to the final classification [32]. In this way, responses contaminated by noise, for example, should be assigned a low score and then only marginally contribute to the selection of a letter.…”

“…Decision confidence w s,t for participant s during trial t is estimated by solving the binary classification problem where, for each trial, the output is whether the participant made the correct decision (label +1) or not (label -1). We build a separation hyperplane using SVM, and we measure the confidence looking at the distribution of the trials with respect to the separating hyperplane, drawing inspiration from the Optimized Score-Based decision Function (OSBF) proposed for P300 spellers [32]. We, therefore, call this method collaborative Optimized Score-Based decision Function (cOSBF).…”