Adding Real-Time Bayesian Ranks to Error-Related Potential Scores Improves Error Detection and Auto-Correction in a P300 Speller

Abstract

Brain-computer interface (BCI) spellers could improve access to communication for people with profound physical disabilities; however, improved speed and accuracy of these spellers is required to make them practical for everyday use. Here we introduce the combination of P300-speller confidence with the error-related potential (ErrP) to improve online single-trial error detection and correction accuracies in a BCI speller. First, we present a mechanism for obtaining P300-confidence using a real-time Bayesian dynamic stopping framework that makes novel use of additional stimuli that occur due to epoch and filter delays. Second, we propose an ensemble of decision trees to combine ErrP and P300-confidence features. Third, we describe the unique attentional differences between error and correct feedback in our spelling interface and discuss how these differences affect ErrP physiology. We tested online error detection on 11 typically developed adults using a BCI system trained on a previous day and found an average sensitivity of 86.67% and specificity of 96.59%. Automatic correction increased selection accuracy by 13.67% and utility grew by a factor of 4.48. We found, however, that the improved performance was primarily attributable to the inclusion of P300 confidence in error detection, calling into question the significance of single-trial ErrP detection.

A Review of Error-Related Potential-Based Brain–Computer Interfaces for Motor Impaired People

“…Margaux et al [8] and Spuler et al [107] led the work and employed the automatic error correction system (ECS) in their BCI speller. More recently, Zeyl et al [19] employed a two-step row-column speller and reported a 13.67% improvement in selection accuracy for 2.54 symbols/minute with the ECS.…”

“…A typical BCI provides an alternative path of communication to a disabled person with the aim of decoding their intention through their neurophysiological signals recorded using EEG [13], [17]. Misclassification of the user's intent results in an erroneous condition which elicits an errorrelated potential (ErrP) signal in the human brain following the perception of the error [18], [19]. The ErrP signal can be integrated with conventional BCIs to form a hybrid-BCI system that can take corrective action on the detection of ErrP to prevent the erroneous action from being executed and ultimately improving the efficacy of the BCI [20]- [22].…”

An EEG Feature Extraction Method Based on Sparse Dictionary Self-Organizing Map for Event-Related Potential Recognition

“…Zhang et al used ICA and Kalman in combination to reduce the interference of white noise on ICA [10] and improve the performance of ICA and ERP extraction; Fukami et al used Particle Filter to extract P300 waveforms [11], and obtained more accurate delay estimation and P300 component amplitude estimation; Ting et al used Kalman filter to extract ERP, by adding the EM algorithm to Kalman filter to achieve a more accurate amplitude estimation [12]; Delaney-Busch et al used Bayesian estimation to study semantic understanding in the process of learning-the trial by trial change of the N400 component in the ERP waveform cannot be achieved by the superimposed average method [13]. Zeyl et al used Bayesian ranks to analyze and calculate the Event related potential scores of each trial, and use event-related potential scores as the time domain features to improve the accuracy of the P300-based speller [14]. This kind of method can handle delay estimation and amplitude estimation on a single-signal frame and improve the classification accuracy.…”

An Optimal Transport Based Transferable System for Detection of Erroneous Somato-Sensory Feedback from Neural Signals

“…We kept the higher cut-off at 6 Hz to remove any overlapping of the motor imagery signals with our signal of interest. ErrP signals are dominant in the range of [0.1, 10] Hz [10] while motor imagery signals are dominant in the range of [8,12] Hz [40]. As the online experiment is based on motor imagery control, some overlap between motor imagery and ErrP waveforms were found in electrode Cz during preliminary analysis.…”