Combining multiple features for error detection and its application in brain–computer interface

Abstract

BackgroundBrain–computer interface (BCI) is an assistive technology that conveys users’ intentions by decoding various brain activities and translating them into control commands, without the need of verbal instructions and/or physical interactions. However, errors existing in BCI systems affect their performance greatly, which in turn confines the development and application of BCI technology. It has been demonstrated viable to extract error potential from electroencephalography recordings.MethodsThis study proposed a new approach of fusing multiple-channel features from temporal, spectral, and spatial domains through two times of dimensionality reduction based on neural network. 26 participants (13 males, mean age = 28.8 ± 5.4, range 20–37) took part in the study, who engaged in a P300 speller task spelling cued words from a 36-character matrix. In order to evaluate the generalization ability across subjects, the data from 16 participants were used for training and the rest for testing.ResultsThe total classification accuracy with combination of features is 76.7 %. The receiver operating characteristic (ROC) curve and area under ROC curve (AUC) further indicate the superior performance of the combination of features over any single features in error detection. The average AUC reaches 0.7818 with combined features, while 0.7270, 0.6376, 0.7330 with single temporal, spectral, and spatial features respectively.ConclusionsThe proposed method combining multiple-channel features from temporal, spectral, and spatial domain has better classification performance than any individual feature alone. It has good generalization ability across subject and provides a way of improving error detection, which could serve as promising feedbacks to promote the performance of BCI systems.

Invariance and variability in interaction error-related potentials and their consequences for classification

“…The timing of the different deflections, however, seems to be more consistent across subjects as indicated by the values of the coefficient of variation (i.e., the standard deviation divided by the mean) reported in the last row of table 2. As in [51], it can be argued that the high variance of classification accuracy across subjects, reported in table 3, can be attributed to the fact that these subjects have different concentration levels on the performed tasks.…”

“…However, despite this inter-subject variability, no significant difference in peak latency or peakto-peak amplitude of interaction ErrPs has been observed across the groups of healthy and motorimpaired subjects [30]. Further, it has been shown in [7,51] that a classifier learned from examples of interaction ErrPs obtained from the EEG of several subjects time-locked to erroneous/correct interactions can transfer to new subjects performing the same task with relatively good accuracy (75% on average). The across-subjects classifier transfer, for most subjects, performed worse than a classifier transfer among different types (observation to interaction and vice versa) of ErrPs within the same subjects [7].…”

“…The different experimental tasks produce imbalanced data sets (i.e. numbers of ErrP and noErrP trials are different), typically resulting in biased classification towards the majority class [51]. As a remedy, the normalized mutual information (NMI) [2] was additionally adopted as a single metric that incorporates both sensitivity and specificity.…”

Cortical Classification with Rhythm Entropy for Error Processing in Cocktail Party Environment Based on Scalp EEG Recording

“…The ERN arose after an erroneous response and the maximum peak was localized at the medial frontal regions 14 , 16 . Recently, researchers found that the ERN potential was used in BCI for adjusting command outputs of BCI systems when subjects observed incorrect outputs from BCI systems, thus facilitating the development of BCI systems with improved accuracy 17 .…”

A CAD system for pulmonary nodule prediction based on deep three-dimensional convolutional neural networks and ensemble learning

“…CAD systems are divided into two subcategories: computer-aided detection (CADe) and computer-aided diagnosis (CADx) systems. A CAD system for detecting pulmonary nodules usually involves four stages: lung segmentation, nodule detection, feature analysis, and false positive elimination [2]. To meet these challenges, many researchers have explored extracting information from pulmonary nodules to improve the sensitivity of nodule detection.…”