Subspace averaging of steady-state visual evoked potentials

A new algorithm for doing signal averaging of steady-state visual evoked potentials (VEP's) is described. The subspace average is obtained by finding the orthogonal projection of the VEP measurement vector onto the signal subspace, which is based on a sinusoidal VEP signal model. The subspace average is seen to out-perform the conventional average using a new signal-to-noise-ratio-based performance measure on simulated and actual VEP data.     

Optimal detection of visual evoked potentials

The authors consider the problem of detecting visual evoked potentials (VEP's). A matched subspace filter is applied to the detection of the VEP and is demonstrated to perform better than a number of other evoked potential detectors. Unlike single-harmonic detectors, the matched subspace filter (MSF) detector is suitable for detecting multiharmonic VEP's. Moreover, the MSF is optimal in the uniformly most powerful sense for multiharmonic signals with unknown noise variance     

Estimation of single sweep steady-state visual evoked potentials byadaptive line enhancement

The adaptive line enhancer (ALE) was used to obtain estimates of the single sweep steady-state visual evoked potential (SSVEP). Some relevant theoretical properties of the ALE are reviewed in the context of designing the ALE for this particular application. Using power spectral density estimates of the unprocessed and enhanced SSVEP data it was found that the method enhanced the estimated signal-to-noise ratio of the single sweep SSVEP by as much as 10 dB     

Preprocessing and meta-classification for brain-computer interfaces

A brain-computer interface (BCI) is a system which allows direct translation of brain states into actions, bypassing the usual muscular pathways. A BCI system works by extracting user brain signals, applying machine learning algorithms to classify the user's brain state, and performing a computer-controlled action. Our goal is to improve brain state classification. Perhaps the most obvious way to improve classification performance is the selection of an advanced learning algorithm. However, it is now well known in the BCI community that careful selection of preprocessing steps is crucial to the success of any classification scheme. Furthermore, recent work indicates that combining the output of multiple classifiers (meta-classification) leads to improved classification rates relative to single classifiers (Dornhege et al., 2004). In this paper, we develop an automated approach which systematically analyzes the relative contributions of different preprocessing and meta-classification approaches. We apply this procedure to three data sets drawn from BCI Competition 2003 (Blankertz et al., 2004) and BCI Competition III (Blankertz et al., 2006), each of which exhibit very different characteristics. Our final classification results compare favorably with those from past BCI competitions. Additionally, we analyze the relative contributions of individual preprocessing and meta-classification choices and discuss which types of BCI data benefit most from specific algorithms.     

Toward unsupervised adaptation of LDA for brain-computer interfaces

There is a step of significant difficulty experienced by brain-computer interface (BCI) users when going from the calibration recording to the feedback application. This effect has been previously studied and a supervised adaptation solution has been proposed. In this paper, we suggest a simple unsupervised adaptation method of the linear discriminant analysis (LDA) classifier that effectively solves this problem by counteracting the harmful effect of nonclass-related nonstationarities in electroencephalography (EEG) during BCI sessions performed with motor imagery tasks. For this, we first introduce three types of adaptation procedures and investigate them in an offline study with 19 datasets. Then, we select one of the proposed methods and analyze it further. The chosen classifier is offline tested in data from 80 healthy users and four high spinal cord injury patients. Finally, for the first time in BCI literature, we apply this unsupervised classifier in online experiments. Additionally, we show that its performance is significantly better than the state-of-the-art supervised approach.     

Alternate half-field stimulation technique for SSVEP-based brain-computer interfaces

A technique of half-field alternate visual stimulation, combined with differential EEG signal measurement, is applied to acquire steady-state brain-evoked signals. Taking the difference of two signals, measured with properly placed electrodes, enhances the visual-evoked potential (VEP) and suppresses the noise components. An array of different-flash-frequency light-emitting-diode (LED) pairs forms a multiple-choice table. By fixating at different LED pairs, the user communicates (by the VEP of corresponding frequency) his/her decision about the selection of the table entry.     

Triple wavelength LED driver for optical brain-computer interfaces

A dedicated triple wavelength LED driver is presented for optical brain?computer interfacing (BCI). The solution caters for the constraints of a common-anode grounded case and modulation up to several kilohertz that allows source separation of light that has backscattered from the brain. With total harmonic distortion of 0.95% and a frequency range of ∼40 kHz, the driver has application in a continuous wave optical BCI. Other modulation strategies such as time division multiplexing (TDM) are catered for, owing to input DC coupling. Linearity in the optical output is maintained by the 'load sensing' differential op-amp on the LED?s current limiting resistor, which is the basis for the V-I conversion.     

The adaptive chirplet transform and visual evoked potentials

We propose a new approach based upon the adaptive chirplet transform (ACT) to characterize the time-dependent behavior of the visual evoked potential (VEP) from its initial transient portion (tVEP) to the steady-state portion (ssVEP). This approach employs a matching pursuit (MP) algorithm to estimate the chirplets and then a maximum-likelihood estimation (MLE) algorithm to refine the results. The ACT decomposes signals into Gaussian chirplet basis functions with four adjustable parameters, i.e., time-spread, chirp rate, time-center and frequency-center. In this paper, we show how these four parameters can be used to distinguish between the transient and the steady-state phase of the response. We also show that as few as three chirplets are required to represent a VEP response. Compared to decomposition with Gabor logons, a more compact representation can be achieved by using Gaussian chirplets. Finally, we argue that the adaptive chirplet spectrogram gives a superior visualization of VEP signals' time-frequency structures when compared to the conventional spectrogram.     

A periodogram-based method for the detection of steady-statevisually evoked potentials

The task of objective perimetry is to scan the visual field and find an answer about the function of the visual system. Flicker-burst stimulation-a physiological sensible combination of transient and steady-state stimulation-is used to generate deterministic sinusoidal responses or visually evoked potentials (VEPs) at the visual cortex, which are derived from the electroencephalogram by a suitable electrode array. Here, the authors develop a new method for the detection of VEPs. Based on the periodogram of a time-series, they test the data for the presence of hidden periodic components, which correspond to steady-state VEPs. The method is applied successfully to real data     

An adaptive approach to spectral analysis of pattern-reversal visual evoked potentials

A method for spectral analysis of pattern-reversal visual evoked potentials (PRVEP's) is presented that results in spectral peaks of uniform width in the frequency domain for signals with a wide range of time-domain duration. Uniformity of spectral peak width is necessary for accurate comparison of spectra. The desired frequency domain characteristics can be achieved through the application of "tunable" data windows prior to transformation. The Io-sinh (Kaiser), Gaussian, and cosine-taper (Tukey) windows were evaluated as to their ability to produce power spectra with uniform spectral peak width. Objective comparison of power spectra is based on the "spectral parameter," which is a numerical index of power distribution. Application of the method to PRVEP waveforms of normal subjects (N = 20) and to a population of Alzheimer's Disease patients (N = 15) showed the Io-sinh window to be the most effective method, yielding correct classification of all normal and abnormal subjects. The Gaussian window also performed well, with only two misclassifications. Use of the rectangular window resulted in seven misclassifications. The tapered-cosine window was very limited in its applicability, and was about equal in performance to the rectangular window.     

Dynamic extraction of visual evoked potentials through spatialanalysis and dipole localization

The dynamic extraction of evoked potential is a problem of great interest in EEG signal processing. Here, a comprehensive method is presented which integrates spatial analysis and dipole localization to make full use of the spatial-temporal information contained in the multichannel stimulation records. A realistic double boundary head model is constructed through CT scans and a two-step method devised to overcome the ill-posed nature of the forward problem of EEG caused by the low conductivity of the skull. As a result, visual evoked potentials can be effectively extracted from only two consecutive records and the dynamic information of visual evoked potential thus procured. The efficiency of the presented method has been verified by means of computer simulation and a clinical experiment     

Multi-resolution dyadic wavelet denoising approach for extraction of visual evoked potentials in the brain

More powerful techniques need to be developed to extract small and weak visual evoked potentials (VEPs) from the spontaneous cerebral electric activity EEG. The authors present a wavelet decomposition algorithm suitable for identification and detection of very weak VEPs. The cross-correlation analysis between Daubechies wavelet (i.e. dbN) functions /spl psi//sub N/(t), for N=4, 5 ,...,10 and a representative noiseless VEP signal is performed to choose the proper wavelet function, say that with maximum correlation coefficient (highest resemblance) with respect to the representative VEP signal sequence. In this way, the specific choice of the best wavelet prototype function is no longer arbitrary for the application of obtaining pattern reversal VEPs. Extensive clinical experiments have demonstrated that the multiresolution wavelet analysis method can identify and estimate the peak latency of VEP signal well, with only a much reduced trial of ensemble averaging (EA) required. The major advantages of the wavelet transform are that it can 'zoom-in' to time discontinuities, and that orthonormal bases, localised in time and frequency, can be constructed. With this zoom-in property of the wavelet analysis, the irregularities or abnormalities of signals can easily be detected. Also the characteristics of EP signals can be captured by means of wavelet analysis, which can be further used for the detection and recognition of the abnormalities in the brain.     

Analysis of pattern reversal visual evoked potentials (PRVEPs) byspline wavelets

The pattern-reversal visual evoked potentials (PRVEPs) collected from normal and demented subjects are investigated by applying the quadratic spline wavelet analysis. The data are decomposed into six octave frequency bands. For quantitative purposes, the wavelet coefficients in the residual waveform representing the delta-theta band activity (0-8 Hz) are explored to characterize the (N70-P100-N130) complex. Specifically, the coefficients corresponding to the location of N70, P100, and N130 peaks are investigated for their sign in order to test whether they represent a consistent (N70-P100-N130) complex in the averaged waveform. Waveforms with normal latency (N70-P100-N130) complex are observed to have positive second, negative third, and positive fourth coefficients in amplitude in their residual scale standing for the delta-theta (0-8 Hz) band activity. The method allows for the analysis of oscillatory-phase behaviour of the normal and pathological PRVEPs in their delta-theta band based on a few quantitative measures consistent with the time-frequency occurrence of the major components of the evoked potential     

Detection of multiple sclerosis with visual evoked potentials - anunsupervised computational intelligence system

This paper describes the application of a novel unsupervised pattern recognition system to the classification of the visual evoked potentials (VEPs) of normal and multiple sclerosis (MS) patients. The method combines a traditional statistical feature extractor with a fuzzy clustering method, all implemented in a parallel neural network architecture. The optimization routine, ALOPEX, is used to train the network while decreasing the livelihood of local solutions. The unsupervised system includes a feature extraction and clustering module, trained by the optimization routine ALOPEX. Through maximization of the output variance of each node, and an architecture which excludes redundancy, the feature extraction network retains the most significant Karhunen-Loeve expansion vectors. The clustering module uses a modification to the fuzzy c-means (FCM) clustering algorithms, where ALOPEX adjusts a set of cluster centers to minimize an objective error function. The result combines the power of the FCM algorithms with the advantage of a more global solution from ALOPEX. The new pattern recognition system is used to cluster the VEPs of 13 normal and 12 MS subjects. The classification with this technique can, without supervision, separate the patient population into two groups which largely correspond to the MS and control subject groups. A suitable threshold can be chosen so that the recognizer chooses no false negatives. The use of multiple stimulation patterns appears to improve the reliability of the decision. The reasoning of most neural networks in their decision making cannot easily be extracted upon the completion of training. However, due to the linearity of the network nodes, the cluster prototypes of this unsupervised system can be reconstructed to illustrate the reasoning of the system. In this application, this analysis hints at the usefulness of previously unused portions of the VEP in detecting MS. It also indicates a possible use of the system as a training aide     

Investigation of short-term changes in visual evoked potentials with windowed adaptive chirplet transform

We propose a new application of the adaptive chirplet transform that involves partitioning signals into non-overlapping sequential segments. From these segments, the local time-frequency structures of the signal are estimated by using a four-parameter chirplet decomposition. Entitled the windowed adaptive chirplet transform (windowed ACT), this approach is applied to the analysis of visual evoked potentials (VEPs). It can provide a unified and compact representation of VEPs from the transient buildup to the steady-state portion with less computational cost than its non-windowed counterpart. This paper also details a method to select the optimal window length for signal segmentation. This approach will be useful for long-term signal monitoring as well as for signal feature extraction and data compression.     

Multiple media interfaces for music therapy

This article describes interfaces (and the supporting technological infrastructure) to create audiovisual instruments for use in music therapy. In considering how the multidimensional nature of sound requires multidimensional input control, we propose a model to help designers manage the complex mapping between input devices and multiple media software. We also itemize a research agenda.     

Weighted averaging of evoked potentials

Weighted averages of brain evoked potentials (EP's) are obtained by weighting each single EP sweep prior to averaging. These weights are shown to maximize the signal-to-noise ratio (SNR) of the resulting average if they satisfy a generalized eigenvalue problem involving the correlation matrices of the underlying signal and noise components. The signal and noise correlation matrices are difficult to estimate and the solution of the generalized eigenvalue problem is often computationally impractical for real-time processing. Correspondingly, a number of simplifying assumptions about the signal and noise correlation matrices are made which allow an efficient method of approximating the maximum SNR weights. Experimental results are given using actual auditory EP data which demonstrate that the resulting weighted average has estimated SNR's that are up to 21% greater than the conventional ensemble average SNR.     

A beamformer for the acquisition of evoked potentials

Evoked potentials (EP) contain information about various physiological parameters and the estimation and detection of these signals can aid in the diagnosis of many pathological conditions. However, the signal-to-noise ratio (SNR) for EP measurement is often very low, and thus signal processing techniques must be employed to enhance the SNR. A delay and sum beamformer acquisition system has the potential for significant SNR improvement in EP measurements. In this communication it is shown that an electrode array acquisition system implements a uniform coherent delay and sum beamformer. The performance of the beamformer is characterized in terms of the number of electrodes, and cross-channel correlation. When compared to conventional ensemble averaging, the beamformer reduces the number of response repetitions required to achieve a given SNR by a factor which approaches the number of channels in the acquisition system.     

Real-time filtering for the estimation of steady-state visualevoked brain potentials

It is shown that EEG visual evoked potentials elicited by repetitive stimuli in the range of 2 to 20 per second can be readily estimated in real time using a simple filtering approach. This measurement takes advantage of the fact that a comb filter will pass the important Fourier harmonics of the signal to provide an estimate of the evoked activity, plus track time-variations in the signal. Results on human subjects demonstrate the effectiveness of the approach.     

Frequency domain dipole localization: extensions of the method andapplications to auditory and visual evoked potentials

A statistical frequency domain approach to localizing equivalent dipole generators of human brain evoked potentials is described. The frequency domain representation allows considerable data reduction, constrains the magnitude function of the dipoles to be smooth, and accounts for the statistical properties of the background EEG. A general model in which dipole orientation can vary over time, and which includes multiple dipole generators is considered. The varying orientation model has the practical advantage of being more nearly linear and more flexible than a fixed orientation model, which facilitates convergence of the iterative fitting algorithm. A measure of goodness-of-fit that compares the likelihood of the dipole model with the likelihoods of saturated and null models is suggested. The results of fitting the model report recorded auditory and visual evoked potentials are reported. A single dipole with fixed orientation seems to be an adequate model of the auditory midlatency response, while two dipoles with varying orientation are needed to fit the later P200 component. Analysis of the visual P100 response to unilateral stimulation localized a generator in the contralateral occipital cortex, as expected from anatomical considerations