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.

A new mode of communication between man and his surroundings.

The material presented in this paper is the result of a research project that was designed to study the feasibility of establishing an alternative mode of communication between man and his surroundings. The new form of communication proposed uses only the subject's brain waves with no overt physical action required. Subjects' electroencephalograms (EEG) were recorded while they performed various mental tasks designed to elicit hemispheric responses. Features formed from the EEG recording were then used as inputs into a Bayes quadratic classifier to test classification accuracy between the various tasks. The results obtained indicate that it is possible to accurately distinguish between any pair of the five tasks investigated. A comparison between three different methods for creating the feature sets is also presented.

On intersensory evoked potentials.

It has been found in a variety of studies that, for human observers, reaction time is faster to a sound than to a light. The study being reported here was designed to test the hypothesis of audio-visual facilitation when a flash precedes a sound by about 30 to 50 milliseconds. It is hypothesized that the electrophysiological result of this facilitation will be an increased energy of the event related potential (ERP) resulting from the interaction between the visual and auditory processes. The stimuli consisted of a click (A), a flash (V), a simultaneous flash-click (AV), a flash preceding the click by 40 milliseconds (AV40) and by 80 milliseconds (AV 80). All of the stimuli were presented to the subject at random. Comparison of the ERP averages for the five subjects tested, and for the above mentioned conditions, supported the hypothesis of sensory integration. The AV40 averages indicated a larger recruitment of cells, as evidenced by a larger response than the AV, A, or V responses separately.

Man-machine communications through brain-wave processing.

The possibility of monitoring voluntarily produced changes in the electroencephalogram (EEG) of a subject and translating these changes into a set of commands to be issued to an external device was investigated. Subjects performed five distinct tasks under both eyes-open and eyes-closed conditions. A feature set consisting of the asymmetry ratios and the power values for each lead at four frequency bands-delta (0-3 Hz) theta (4-7 Hz), alpha (8-13 Hz), and beta (14-20 Hz)-was used to characterize the EEG. The feature sets created from an estimate of the spectral density of the EEG for each task were used to test classification accuracy among the various tasks using a Bayes quadratic classifier. The results show that it is possible to distinguish, to a high degree of accuracy, among the various mental tasks studied, using only the EEG.

On the classification of single evoked potentials using a quadratic classifier.

It has been shown that using a computationally simple technique it is possible to classify single event-related potentials associated with upper and lower visual field stimulation with a high degree of accuracy. A flow chart of the algorithm and the FORTRAN 77 computer program required to perform the classification just described are included in the Appendix. The program as given in the Appendix is for a 2 class problem only. Memory limitations (33000 bytes 16-bit words) per user in the time-shared system utilized limits the amount of data that may be in core at any one time. A 2 class, 5 step problem takes approximately 5 min in a PDP ll/45 System (Digital Equipment Corp.).

Signal processing in evoked potential research: averaging and modeling.

Three separate topics are covered in this review. The first deals with the technique of signal averaging. The concept of ensemble averaging is explored, and alternatives to this traditional tool are considered such as crosscorrelation averaging, latency corrected averaging, median averaging, etc. Different measures of variability of single evoked potentials are finally discussed. The second topic deals with modeling of the evoked potentials. The direct and inverse problems of source localization are discussed. Recent results of the application of these techniques to single evoked potentials are given. The third topic deals with the use of principal components for signal representation and comparison. Geometric consideration and varimax rotation of coefficients are discussed and examples given.

Signal processing in evoked potential research: applications of filtering and pattern recognition.

The separate but closely related topics of waveform estimation by filtering and information extraction by pattern recognition are covered in this review. Because of the low signal-to-noise ratio generally encountered in evoked potential research, a variety of filtering methods have been employed for improving waveform estimation. Initially the filtering was done with analog devices but with the availability of high performance minicomputers virtually all filtering is now done digitally. Filters of various types are considered. Among them are single and multiple channel Wiener filtering, Kalman filtering, minimum mean square error filtering, maximum signal-to-noise filtering, and several types of nonlinear filters. The application of adaptive filtering techniques is also considered. In recent years there has been a continual increase in the application of pattern recognition techniques to the processing of evoked potentials. The techniques are based on statistical decision theory and the underlying basis of these procedures is reviewed. The technique of linear stepwise discriminant analysis is considered as well as the use of general discriminant functions of linear and quadratic types. Applications of these procedures to psychophysiological testing are discussed with particular emphasis on auditory and visual event-related 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.

High frequency components in the spectrum of the visual evoked potential.

The occurrence of short term high frequency components in the spectra of the VEP is usually overshadowed by larger amplitude slow waves which occur simultaneously with the high frequency waves. The high frequency components become evident when the time varying power spectra of short duration (250 ms) segments of the VEP are examined. Two types of short duration time varying power spectra are considered: one is obtained directly from overlapping time segments of the average VEP; and, the other is obtained by computing the spectra from the average autocorrelation functions of overlapping time segments of the data following each stimulus. The first yields the time varying properties for the spectra of the stimulus-synchronized data while, the latter shows the time varying properties of both the synchronized and unsynchronized data present.