Boyu Wang

Comparison of different classification methods for EEG-based brain computer interfaces: A case study

The performances of different off-line methods for two different Electroencephalograph (EEG) signal classification tasks - motor imagery and finger movement, are investigated in this paper. The classifiers based on linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), kernel fisher discriminant (KFD), support vector machine (SVM), multilayer perceptron (MLP), learning vector quantization (LVQ) neural network, k-nearest neighbor (k-NN), and decision tree (DT), are compared in terms of classification accuracy. The main purpose of this paper is to provide a fair and extensive comparison of some commonly employed classification methods under the same conditions so that the assessment of different classifiers will be more convictive. As a result, a guideline for choosing appropriate algorithms for EEG classification tasks is provided.

A high rate online SSVEP based brain-computer interface speller

This paper presents an online steady-state visual evoked potential (SSVEP) based brain-computer interface (BCI). Stimuli are displayed on a liquid crystal display (LCD) screen with a frame based encoding method to elicit SSVEPs with a wide range of frequencies. This system focuses strongly on practicability and convenience, including an adequate alphabet (42 characters) that can allow a wide range of options. Four healthy subjects attain a mean information transfer rate (ITR) of 61.64±3.61 bits/min.

EEG signals classification for brain computer interfaces based on Gaussian process classifier

Classification of electroencephalogram (EEG) is a crucial issue for EEG-based brain computer interface (BCI) system. In this paper, the performances of the Gaussian process classifier (GPC) for three different categories of EEG signals, i.e. steady state visually evoked potential (SSVEP), motor imagery and finger movement EEG data, are investigated. The main purpose of this paper is to explore the practicability of GPC for EEG signals classification of different tasks. Compared with some commonly employed algorithms, the GPC achieves similar or better performances. Furthermore, the probabilistic output provided by the GPC can also be of great benefit to the decision making for both online and offline EEG analysis.

Trial pruning based on genetic algorithm for single-trial EEG classification

We consider the problem of artifacts in electroencephalography (EEG) data. In a practical motor imagery based brain-computer interface (BCI) system, EEG signals are usually contaminated by misleading trials caused by artifacts, measurement inaccuracies, or improper imagination of a movement. As a result, the performance of a BCI system can be degraded. In this paper, we introduce a novel algorithm combining Gaussian mixture model (GMM) and genetic algorithm (GA) to detect the abnormal EEG samples. In addition, this algorithm can be also integrated with other data-driven feature exaction method (e.g., common spatial pattern (CSP)) so that a more reliable analysis can be obtained by pruning the potential outliers and noisy samples, and consequently the performance of a BCI system can be improved. Experimental results demonstrate significant improvement in comparison with the conventional mixture model.

Gaussian mixture model based on genetic algorithm for brain-computer interface

Gaussian mixture model (GMM) has been considered to model the EEG data for the classification task in brain-computer interface (BCI) system. In the practical BCI application, however, the performance of the classical GMM optimized by standard expectation-maximization (EM) algorithm may be degraded due to the noise and outliers, which often exist in realistic BCI systems. The motivation of this paper is to introduce the GMM based on the combination between the genetic algorithm (GA) and EM method to give a probabilistic output for further analysis and, more important, to achieve the reliable estimation by pruning the potential outliers and noisy samples in the EEG data, so the performance of BCI system can be improved. Experiments on two BCI datasets demonstrate the improvement in comparison with the classical mixture model.

A Solution to harmonic frequency problem: Frequency and phase coding-based brain-computer interface

In this paper, we propose a modified visual stimulus generation method and feature detection algorithm to design a frequency and phase coding steady-state visual evoked potential (SSVEP) based brain-computer interface (BCI). By utilizing both frequency and phase information, we solve the harmonic frequency problem in our proposed SSVEP-BCI system. The offline experimental results show that the proposed feature detection algorithm can enhance the classification rate over 10% (from 69%±12% to 82%±8%) even though only one signal electrode is used and the harmonic frequencies (6.67Hz, 13.33Hz, 8.57Hz and 17.14Hz) are employed.

An online SSVEP-based chatting system

This paper presents a steady-state visual evoked potential (SSVEP) based online brain-computer interface (BCI) chatting system. In this application, it helps disable subjects to chat by means of their brain signal activities. Stimuli illuminated at different frequencies are displayed on liquid crystal display (LCD) screen. The BCI system implements a chatting interface programmed in Java, and it provides a convenient and high efficient communication platform for paralyzed patients. A novel idle state detection is introduced in this proposed system, and a high transfer rate of 63.54 bits/min and 92% idle state detection accuracy are achieved.

Outlier detection for single-trial EEG signal analysis

The performance of a brain computer interface (BCI) system is usually degraded due to the outliers in electroencephalography (EEG) samples. This paper presents a novel outlier detection method based on robust learning of Gaussian mixture models (GMMs). We apply the proposed method to the single-trial EEG classification task. After trial-pruning, feature extraction and classification are performed on the subset of training data, and experimental results demonstrate that the proposed method can successfully detect the outliers and therefore achieve more reliable result.

Robust learning of mixture models and its application on trial pruning for EEG signal analysis

This paper presents a novel method based on deterministic annealing to circumvent the problem of the sensitivity to atypical observations associated with the maximum likelihood (ML) estimator via conventional EM algorithm for mixture models. In order to learn the mixture models in a robust way, the parameters of mixture model are estimated by trimmed likelihood estimator (TLE), and the learning process is controlled by temperature based on the principle of maximum entropy. Moreover, we apply the proposed method to the single-trial electroencephalography (EEG) classification task. The motivation of this work is to eliminate the negative effects of artifacts in EEG data, which usually exist in real-life environments, and the experimental results demonstrate that the proposed method can successfully detect the outliers and therefore achieve more reliable result.

Trial pruning for classification of single-trial EEG data during motor imagery

Due to the artifacts in electroencephalography (EEG) data, the performance of brain-computer interface (BCI) is degraded. On the other hand, in the motor imagery based BCI system, EEG signals are usually contaminated by the misleading trials caused by improper imagination of a movement. In this paper, we present a novel algorithm to detect the abnormal EEG data using genetic algorithm (GA). After trial pruning, a subset of the EEG data are selected, on which common spatial pattern (CSP) and Gaussian classifier are trained. The performance of the proposed method is tested on Data set IIa of BCI Competition IV, and the simulation result demonstrates a significant improvement for six out of nine subjects.