Jorge I. Aunon

Improved Waveform Estimation Procedures for Event-Related Potentials

Several methods of estimating the waveform of event-related potentials are presented. The techniques of conventional averaging, Woody cross-correlation averaging, latency corrected averaging, continuous latency corrected averaging, and enhanced averaging are described and their results compared. It was found that the continuous latency corrected average appears to offer the most useful representation of the waveform of the event-related potential.

Signals and Noise in Evoked Brain Potentials

Event-related brain potentials measured with scalp electrodes are always corrupted by unrelated electrical discharges occurring in the brain. These unrelated electrical discharges, generally referred to as noise, have temporal and spectral characteristics similar to evoked potential waveforms, and they greatly increase the difficulty of detecting and estimating the parameters of the evoked potential waveforms themselves. This problem has been analyzed by computing the probability distributions for measured amplitudes and latencies of ERP components measured in the presence of the ongoing EEG. The analytical results have been verified over a wide range of signal-to-noise ratios by computer simulation. Comparisons of theoretical results to measured data indicate that the latency variations found experimentally greatly exceed what would be expected if they were due only to additive noise. It may be concluded, therefore, that the single ERP is not a signal whose components are deterministically related to the stimulus, but is made up of components that shift significantly in both amplitude and latency from one stimulus application to the next. Using the expressions developed in the paper, it is possible to separate the contributions to the variance due to interference from the ongoing EEG and that inherent in the ERP.

Classification and Detection of Single Evoked Brain Potentials Using Time-Frequency Amplitude Features

The classification and detection of event-related brain potentials was investigated using signal processing and statistical pattern recognition techniques. Amplitudes at sampled time points and frequency quantities have previously been used as features. Improvements to these procedures were obtained by using features from the time-frequency plane to utilize the geometric relationship between time and frequency, capitalizing on the nonstationarity of the evoked potential signals. These features were transformed from the original data sets based upon a two-step classification/feature selection procedure which uses selected frequencies from step 1 as parameters for data filtering in step 2. Features were selected from the filtered data, classifiers were designed, and the estimated classification accuracies were computed.

Dipole Localization of Average and Single Visual Evoked Potentials

A single dipole source was chosen as a model of the neurological generator of evoked potentials elicited by illuminated checker-board stimulation of halves and quadrants of the visual field, and a homogeneous sphere was chosen as a model for the head. Dipole locations and moments were calculated for the average evoked potentials and average dipole locations and moments were calculated for the preprocessed single evoked potentials. The average dipole locations for the single potentials appear to be more anatomically appropriate than the location due to the average evoked potentials. A relationship was demonstrated between the dipole parameters and polarity inversions in the average evoked potentials.

Computer techniques for the processing of evoked potentials

A computer technique is described for the systematic characterization of brain evoked potentials. Preprocessing of the data by a causal digital filter is followed by a peak search and identification procedure. A latency histogram of the peaks found is constructed to determine the ranges or clusters of peaks within the single evoked potential. Peaks found within the clusters are then corrected to the mean latency of the cluster and a latency corrected average version of the evoked potential constructed.

Discrimination among visual stimuli by classification of their single evoked potentials

A correlation classifier and a maximum likelihood quadratic classifier are used to classify single evoked potentials from four different, visual stimuli. Forward sequential feature selection was employed with the quadratic classifier to reduce data dimensionality. The features selected by this method were the signal amplitudes at latencies at which polarity reversals were present in the averages. Examination of detected peaks from the single v.e.p.s suggested that the polarity reversals were also present in the single v.e.p.s. The quadratic classifier gave better than 90% recognition for the case of four classes and three classes.

Nonlinear system identification and overparameterization effects in multisensory evoked potential studies

Traditional signal processing techniques have not been suitable in establishing contributions from different sensory paths in multisensory evoked potentials. In this paper, a nonlinear modeling technique is proposed to demonstrate the possible mechanisms of interaction between sensory paths. The nonlinear autoregressive with exogenous inputs (NARX) model is explored to establish a relationship between electrical activities of the brain obtained by unimodal and by bimodal stimulation. The intersensory phenomenon concept is extended using nonlinear system theory and applied to show the possible interactions between the visual and auditory sensory paths. In addition, the paper addresses the compensation phenomenon caused by overparameterization in the NARX algorithm when it is applied to event-related potentials. It is hoped that the nonlinear modeling approach will generate hypotheses about the intersensory interaction phenomenon, improving and advancing its theoretical formulation.

Optimal and suboptimal feature selection for classification of evoked brain potentials

Exhaustive feature selection algorithms are optimal because all possible combinations of features are tested against a predetermined criterion. Suboptimal algorithms that trade performance for speed by considering only a subset of all feature combinations are generally preferred. An implementation of the exhaustive search feature selection (ESFS) method is described for the Bayes Gaussian statistics. The algorithm significantly reduces the computational and time requirements normally associated with optimal algorithms. The performance of this algorithm is compared to that of two suboptimal algorithms-forward sequential features selection and stepwise linear discriminant analysis. Results show that this implementation provides a moderate improvement in classification accuracy and is well suited for evaluating the performance of suboptimal algorithms.

EXPERIMENTAL COMPARISON OF THREE METHODS OF PROCESSING EVOKED POTENTIALS

Publisher Summary This chapter presents experimental comparison of three methods of processing evoked potentials. The most common method of processing stimulus evoked brain potentials is by averaging a large number of repeated measurements. The basic assumptions in this type of processing are that the electrical response to the stimulus is time locked to the stimulus application and that the ongoing electroencephalographic (EEG) activity or noise is additive and uncorrelated with the signal. Under these assumptions, the improvement in signal-to-noise (power) ratio is directly proportional to the number of responses that are averaged together. This type of averaging is not an optimum procedure when the data does not satisfy the above assumptions. The chapter describes the Woody crosscorrelation averaging technique. In this method, an initial approximation to the signal is crosscorrelated with each single evoked response and the latency difference for which the crosscorrelation function is a maximum is determined. Each response is then shifted, the appropriate direction in time by the measured latency difference recorded above. The average of all the shifted responses is then computed and used as the template for the next iteration of the technique. In addition to providing a better representation of the signal, the Woody crosscorrelation averaging technique can provide an approximation to the mean and standard deviation of the signal latency for each iteration.