H. de Bruin

Classification of brain-stem auditory evoked potentials by syntactic methods ☆

A syntactic pattern recognition procedure for classification of brain-stem auditory evoked potential (BSAEP) is presented. A pre-processing stage of zero-phase bandpass filtering enhances the peaks and suppresses the noise. A finite-state grammar was designed to identify the peaks. Attributes of the peaks (latencies and amplitudes) that are identified are checked for their acceptability. A training run on 70 subjects of known diagnosis was performed to fine-tune the system and build up necessary acceptance criteria. Peak latency differences are used for the classification rather than absolute peak latencies. Acceptance criteria for peak latency differences were empirically optimized. A data base of normal BSAEPs, created during the training run, was updated and used during the test run. Test of the classifier using 60 subjects yielded a classification accuracy of 83%. The classifier has acceptable accuracy and can be modified for other evoked potentials such as visual and somatosensory by establishing relevant attribute tables.

A Simple and Fast Algorithm for Automatic Suppression of High-Amplitude Artifacts in EEG Data

In this paper we present a simple and fast technique for correcting high amplitude artifacts that contaminate EEG signals. Examples of such artifacts are ocular movement, eye blinks, head movement, etc. Since the measured EEG data can be modeled as a linear combination of brain sources and artifacts, the proposed technique is based on multiplying the observed data matrix by a blocking matrix that has the effect of blocking high amplitude artifacts, while linearly transforming the other sources without any distortion. The advantages of using this technique are: 1) it is relatively fast, so it can be applied in real time, 2) it is completely automatic, and 3) can be successfully applied to signals which fail with ICA-based algorithms.

Electroencephalogram signals classification for sleep- state decision - a Riemannian geometry approach

In this work, the authors study the classification of electroencephalogram (EEG) signals for the determination of the state of sleep of a patient. They employ the power spectral density (PSD) matrices as the feature for the distinction between different classes of EEG signals. This not only allows us to examine the power spectrum contents of each signal as well as the correlation between the multi-channel signals, but also complies with what clinical experts use in their visual judgement of EEG signals. To establish a metric facilitating the classification, the authors exploit the specific geometric properties, and develop, with the aid of fibre bundle theory, an appropriate metric in the Riemannian manifold described by the PSD matrices. To use this new metric effectively for the EEG signal classification, the authors further need to find a weighting for the PSD matrices so that the distances of similar features are minimised whereas those for dissimilar features are maximised. A closed form of this weighting matrix is obtained by solving an equivalent convex optimisation problem. The effectiveness of using these new metrics is examined by applying them to a collection of recorded EEG signals for sleep pattern classification based on the k-nearest neighbour decision algorithm with excellent outcome.

Crosstalk resistant adaptive noise cancellation

Presence of signal components in the reference input is detrimental to the performance of practical adaptive noise cancellation systems. Using a modeling approach, we analyse the performance of adaptive noise cancellation in the presence of signal cross talk. We demonstrate a crosstalk resistant adaptive noise cancellation method. After showing that the original signal cannot be recovered if the ability to prevent adaptation does not exist, we discuss the use of a weighted exact least squares lattice algorithm in the joint estimation form, where adaptation can be controlled. Using stimulated data, it is shown that signal can be estimated with good accuracy, even when there is significant signal crosstalk in the reference input.

Near-optimal control of high-order systems using low-order models †

This paper considers the use of low-order models for the determination of an approximate optimal control of a high-order system. For a number of such models of a given system, derived using different methods, the approximate optimal control input is calculated. The result of the application of these inputs to the system is compared with the actual optimum. It is shown that optimum low-order models provide a near-optimal control, which may serve as a good first approximation to the optimum. Together with a suitable scheme for on-line identification these models may be used effectively for adaptive control.

Designing and constructing a magnetic stimulator: theoretical and practical considerations

Magnetic nerve stimulation has proven to be an effective non-invasive technique that can be used to excite peripheral and central nervous systems. In this technique, the excitement of the neural tissue depends on exposing the body to a transient magnetic field. This field can be generated by passing a high pulse of current through a coil over a short period of time. This paper presents general guidelines for designing and constructing a magnetic stimulator. These guidelines cover theoretical concepts, hardware aspects and components required to build these systems. The critical points discussed in this paper are based on key findings and difficulties encountered during the process of building the system used for this study. Furthermore, some suggestions were addressed to improve future designs.

Magnetic nerve stimulation: field focality and depth of penetration

Magnetic nerve stimulation is a non-invasive method of exciting neural tissue. The major limitation of using magnetic stimulation is the lack of a focused field. At sufficiently high magnetic pulses the diffused field not only stimulates the target population of neurons, but also stimulates adjacent structures as well. Further, for deeply penetrating fields, as is the case in transcranial stimulation, excessively high amplitude current pulses are required in the coils because a significant fraction of the field energy is spread throughout the tissue under the coil. In this paper we propose two new coil designs that can be used for magnetic stimulation of the peripheral or central nervous system. The purpose of the design was to increase field focality and depth of penetration. The magnetic fields produced by these coils, when driven by biphasic pulses, were simulated using a finite element technique coupled with a transient solver. The resultant field densities and gradients were compared with those obtained from the commonly used Figure-8 coil. Both the air core and the ferromagnetic core designs have superior results when compared to the Figure-8 coil.

Measurement of electrical and contractile properties of motor units using EMG and accelerometry

Presents a novel method for recording the electrical and contractile properties of a number of motor units using surface electromyography (EMG) and accelerometry. Preliminary results have been obtained for the left and right thenar muscles of three healthy male subjects. The motor units are recruited by stimulating the median nerve at the wrist using surface electrodes and graded stimulus amplitudes. The resulting M-waves and acceleration signals are decomposed into individual motor unit action potentials and twitches using a pattern recognition technique. Typically 15 to 20 motor units can be studied for each recording sequence.

Criteria to select kernel functions for positive time-frequency distributions of theoretical time-varying signals

The development of positive time frequency distributions (PTFD) is an important issue in the framework of time frequency analysis of nonperiodic signals. A PTFD can be interpreted as the signal energy localized in both time and frequency domains. A PTFD representation depends on a kernel function which is related to the signal to be analyzed. For any particular signal, it is difficult to know its PTFD by only using time and frequency marginal distributions. We show that the uncertainty coefficient based on the entropy principle is a good approach to deciding which kernel function could be appropriate for analyzing a set of sinusoidal signals. As well, a lower bound of the uncertainty coefficient was defined based on the marginal product (correlationless case of PTFD). Three known kernel functions were used to calculate the PTFD of six sinusoidal signals. These signals were considered as a simple approximation to oscillation changes in the electrical activity of the human colon. One of the kernel functions was useful for analyzing almost all sinusoidal signals.<<ETX>>

Development of an Electroencephalography Data Acquisition System for Clinical Research Into Transcranial Magnetic Stimulation Evoked Potentials

We have developed a module to be inserted in-line between electroencephalography (EEG) electrodes and any standard EEG headbox that will prevent amplifier saturation during transcranial magnetic stimulation (TMS). By preventing amplifier saturation TMS evoked potentials (EPs) seen in the millisecond range immediately following TMS can be collected accurately. The new system uses a pre-proven EEG instrumentation concept, sample-and-hold circuitry (Ilmoniemi, 1997), to keep standard EEG front-end amplifiers stable during TMS pulses.