Characteristics of slow waves on EEG associated with epileptic spasms

PURPOSE: The high-voltage slow waves (HVSs) on EEG associated with epileptic spasms were investigated to clarify their characteristics and their relation to the pathophysiology of spasms in West syndrome and related disorders. METHODS: In 14 patients, digitally recorded EEG segments showing the ictal HVSs were extracted and their traces were overlaid by using an average reference. The ictal HVSs were also averaged to build maps for investigation of the pattern of potential distribution over the scalp. RESULTS: In a total of 685 recorded spasms, 346 (50.5%) with minimal artifacts were selected to demonstrate that the ictal HVSs had a largely consistent waveform and distribution in each patient. The ictal HVSs were symmetrical in 10 patients and asymmetrical in the other four, and were relatively negative over the posterior region and positive over the frontal or temporal regions in 11 patients. Two symptomatic patients showed a marked deviation of the distribution of HVSs to the pathologically more involved hemisphere. An infant with Aicardi syndrome had two different types of spasms, each type showing a consistent pattern of HVSs with a lateralized distribution. CONCLUSIONS: The patterns of distribution of the ictal HVS may be related to the abnormal activation of the brain in the generation of spasms.     

An approach to seizure detection using an artificial neural network (ANN)

We have developed an EEG seizure detector based on an artificial neural network. The input layer of the ANN has 31 nodes quantifying the amplitude, slope, curvature, rhythmicity, and frequency components of EEG in a 2 sec epoch. The hidden layer has 30 nodes and the output layer has 8 nodes representing various patterns of EEG activity (e.g. seizure, muscle, noise, normal). The value of the output node representing seizure activity is averaged over 3 consecutive epochs and a seizure is declared when that average exceeds 0.65.Among 78 randomly selected files from 50 patients not in the original training set, the detector declared at least one seizure in 76% of 34 files containing seizures. It declared no seizures in 93% of 44 files not containing seizures. Four false detections during 4.1 h of recording yielded a false detection rate of 1.0/h. The detector can continuously process 40 channels of EEG with a 33 MHz 486 CPU.Although this method is still in its early stages of development, our results represent proof of the principle that ANN could be utilized to provide a practical approach for automatic, on-line, seizure detection.     

Radial Basis Function Network Configuration Using Mutual Information and the Orthogonal Least Squares Algorithm

Input nodes of neural networks are usually predetermined by using a priori knowledge or selected by trial and error. For example, in pattern recognition applications the input nodes are usually the given pattern features and in system identification applications the past input and output data are often used as inputs to the network. Some of the input variables may be irrelevant to the task in hand and therefore may cause a deterioration in network performance. Some may be redundant and may increase the complexity of the network and consume expensive computation time. In the present study, the mutual information between the input variables and the output of the network is used to select a suboptimal set of input variables for the network. The variables are selected according to the information content relevant to the output. Variables which have a higher mutual information with the output and lower dependence on other selected variables are used as network inputs. The algorithms are derived based on heuristics and performance is assessed by using radial basis function (RBF) networks trained with the orthogonal least squares algorithm (OLS), which selects the hidden layer nodes of the network according to the error reduction ratios on the network output. Both real and simulated data sets are used to demonstrate the effectiveness of the new algorithms. Copyright © 1996 Elsevier Science Ltd.     

Interaction between the alpha2-noradrenergic and muscarinic Systems in the regulation of neocortical high voltage spindles

The present experiments were carried out in order to study the interaction between alpha₂-noradrenergic and muscarinic systems in regulating high voltage spindle (HVS) activity in neocortex. Alpha₂-antagonist (atipamezole 1 and 10 mg/kg) blocked HVS activity. Atipamezole at 0.1 mg/kg dose had no effect on HVS activity. Alpha₂-agonist (guanfacine 0.004, 0.02 and 0.1 mg/kg) increased dose dependently HVSs. Guanfacine-induced HVSs were blocked by nucleus reticularis (NRT) lesions and by stimulation of either noradrenergic or cholinergic (pilocarpine) systems. Moreover, combined injections of atipamezole 1 mg/kg and pilocarpine 3 mg blocked HVSs more effectively than either of the drugs alone. Our results suggest that the NRT is jointly modulated by the noradrenergic and cholinergic afferents.     

Automated interictal EEG spike detection using artificial neural networks.

Feed-forward, error-back-propagation artificial neural networks were applied to recognition of epileptiform patterns in the EEG. The inherent network properties of generalization and variability tolerance were effective in identifying wave forms that differed from the training patterns but still maintained 'epileptiform' spatio-temporal characteristics. The certainty of recognition was measured as a continuous function with a range of 0-1. Two levels of certainty (0.825 and 0.900) were used to indicate recognition of spikes and sharp waves (SSW). An average 94.2% (+/- 7.3) of the SSW were recognized; 20.9% (+/- 22.9) of all recognized SSW were false-positive recognitions. The time required for pattern recognition was well within the time required for digitizing the analogue data. This study provides evidence that neural network technology is, in principle, an effective pattern recognition strategy for identification of epileptiform transients in the EEG. The analysis is sufficiently rapid to be of potential value as a strategy for data reduction of long recordings stored on bulk media.     

A multistage knowledge-based system for EEG seizure detection in newborn infants

OBJECTIVE: Automatic seizure detection has attracted attention as a method to obtain valuable information concerning the duration, timing, and frequency of seizures. Methods currently used to detect EEG seizures in adults show high false detection rates in neonates because they lack information about specific age-dependent features of normal and pathological EEG and artifacts. This paper describes a novel multistage knowledge-based seizure detection system for newborn infants to identify and classify normal and pathological newborn EEGs as well as seizures with a reduced false detection rate. METHODS: We developed the system in a way to make comprehensive use of spatial and temporal contextual information obtained from multichannel EEGs. The system development consists of six major stages: (i) EEG data collection and bandpass filtering; (ii) automatic artifact detection; (iii) feature extraction from segments of non-seizure and seizure activities; (iv) feature selection via the relevance and redundancy analysis; (v) EEG classification and pattern recognition using a trained multilayer back-propagation neural network; and (v) knowledge-based decision-making to examine each of possible EEG patterns from a multi-channel perspective. The system was developed and tested with the EEG recordings of 10 newborns aged between 39 and 42 weeks. RESULTS: The overall sensitivity, selectivity, and average detection rate of the system were 74%, 70.1%, and 79.7%, respectively. The average false detection of 1.55/h was also achieved by the system with a feature reduction up to 80%. CONCLUSIONS: The expert rule-based decision-making subsystem accompanying the classifier helped to reduce the false detection rate, reject a wide variety of artifacts, and discriminate various patterns of EEG. SIGNIFICANCE: This paper may serve as a guide for the selection of discriminative features to improve the accuracy of conventional seizure detection systems for routine clinical EEG interpretation and brain activity monitoring in newborns especially those hospitalized in the neonatal intensive care units.     

Interaction between the alpha 2-noradrenergic and muscarinic systems in the regulation of neocortical high voltage spindles

The present experiments were carried out in order to study the interaction between alpha 2-noradrenergic and muscarinic systems in regulating high voltage spindle (HVS) activity in neocortex. Alpha 2-antagonist (atipamezole 1 and 10 mg/kg) blocked HVS activity. Atipamezole at 0.1 mg/kg dose had no effect on HVS activity. Alpha 2-agonist (guanfacine 0.004, 0.02 and 0.1 mg/kg) increased dose dependently HVSs. Guanfacine-induced HVSs were blocked by nucleus reticularis (NRT) lesions and by stimulation of either noradrenergic or cholinergic (pilocarpine) systems. Moreover, combined injections of atipamezole 1 mg/kg and pilocarpine 3 mg blocked HVSs more effectively than either of the drugs alone. Our results suggest that the NRT is jointly modulated by the noradrenergic and cholinergic afferents.     

Synchronous high-voltage spindles in the cortex-basal ganglia network of awake and unrestrained rats

Synchronous oscillations in various frequency ranges have been recorded in several nuclei of the basal ganglia (BG) and are thought to be an information processing mechanism. High-voltage spindles (HVSs) are 5-13 Hz spike-and-wave oscillations, which are commonly recorded in rats and which have been reported in some recent studies where their occurrence in the BG has been investigated. We recorded single neurons and local field potentials (LFPs) simultaneously in the motor cortex, striatum and substantia nigra pars reticulata (SNr) of the freely moving rat. We took advantage of the high level of synchronization observed during HVSs to study signal transmission in the cortex-BG network in the awake animals. The results show that LFPs are synchronized in the motor cortex, striatum and SNr during HVS episodes and that the latter propagate from the cortex to the SNr via the striatum. Moreover, > 50% of single neurons in each of these structures are triggered by the HVS. Following the discharge of cortical cells, SNr neurons are first inhibited after approximately 19 ms and then activated after approximately 45 ms. This response is probably driven by the direct and indirect pathways, respectively, without any involvement of the hyperdirect pathway. Here, it is shown that cortex-BG connectivity can be studied using physiological signals in the freely moving animal as opposed to artificial stimulation under anaesthetized conditions. This opens the door to further studies under various experimental conditions, such as animal models of basal ganglia disorders.     

Automatic recognition of alertness level by using wavelet transform and artificial neural network

We propose a novel method for automatic recognition of alertness level from full spectrum electroencephalogram (EEG) recordings. This procedure uses power spectral density (PSD) of discrete wavelet transform (DWT) of full spectrum EEG as an input to an artificial neural network (ANN) with three discrete outputs: alert, drowsy and sleep. The error back propagation neural network is selected as a classifier to discriminate the alertness level of a subject. EEG signals were obtained from 30 healthy subjects. The group consisted of 14 females and 16 males with ages ranging from 18 to 65 years and a mean age of 33.5 years, and a body mass index (BMI) of 32.4 +/- 7.3 kg/m2. Alertness level and classification properties of ANN were tested using the data recorded in 12 healthy subjects, whereby the EEG recordings were not used been used to train the ANN. The statistics were used as a measure of potential applicability of the ANN. The accuracy of the ANN was 96 +/- 3% alert, 95 +/- 4% drowsy and 94 +/- 5% sleep. The results suggest that the automatic recognition algorithm is applicable for distinguishing between alert, drowsy and sleep state in recordings that have not been used for the training.     

A self-supervised learning system for pattern recognition by sensory integration

Artificial neural networks are useful tools for pattern recognition because they realize nonlinear mapping between input and output spaces. This ability is tuned by supervised learning methods such as back-propagation. In the supervised learning methods, desired outputs of the neural network are needed. However, the desired outputs are usually unknown in unpredictable environments. To solve this problem, this paper presents a self-supervised learning system for category detection. This system learns categories of objects automatically by integrating information from several sensors. We assume that these sensory inputs are always ambiguous patterns that include some noises according to deformations of the objects. After the learning, the system recognizes objects, also controlling the priority of each sensor, according to the deformation of the sensory input pattern.

In the simulation, the system is applied to several learning and recognition tasks using artificial or actual sensory inputs. In all tasks, the system found the categories. Particularly, we applied the new system to the learning of five Japanese vowels with the corresponding shapes of the mouth. As result, the system became to yield specific outputs corresponding to each vowel.     

Indications for network regularization during absence seizures: Weighted and unweighted graph theoretical analyses

Previous studies with intracranial recordings suggested that a more random spatial structure of functional brain networks could be related to seizure generation. Here, we studied whether similar network changes in weighted and unweighted networks can be found in generalized absence seizures recorded with surface EEG. We retrospectively selected EEG recordings of eleven children with absence seizures. The functional neural networks were characterized by calculating both coherence and synchronization likelihood (SL) between 21 EEG signals that were either broad band filtered (1-48 Hz) or filtered in different frequency bands. From both weighted and unweighted networks the clustering coefficient (C) and path length (L) were computed and compared to 500 random networks. We compared the ictal with the pre-ictal network structure. During absence seizures there was an increase of synchronization in all frequency bands, seen most clearly in the SL-based networks, and the functional network topology changed towards a more ordered pattern, with an increase of C/C-s and L/L-s. This study supports the hypothesis of functional neural network changes during absence seizures. The network became more regularized in weighted and unweighted analyses, when compared to the more randomized pre-ictal network configuration.     

Network with shunting synapses as a non-linear frequency modulator

The role of ‘noisy’ excitation in synchronizing interneuron networks with shunting synapses was studied. The excitatory input was simulated as a Poisson pattern of presynaptic conductance with varying frequencies and amplitudes. We find that higher excitation frequencies induce stronger synchronisation of the network. Within the range of 1-10000 Hz, only frequencies between 20 Hz and 200 Hz affected network synchronisation. No detectable network synchronisation was found at excitation frequencies below 20 Hz, and the network’s synchronisation was either almost independent of the external input or falling down to zero when the input frequency was greater than 200 Hz. Thus the network transformed the input signals with frequencies above 20 Hz into output signals with the network’s synchronisation frequency. The network’s synchronisation frequency in our model ranged from 20 to 68 Hz depending on the frequency of the excitatory input. We conclude that a network of interconnected interneurons is capable of converting an asynchronous excitatory input into a synchronous inhibitory output as a frequency amplifier with the amplification coefficient dependent on the number of converging excitatory inputs. Another important result of our work revealed that the external frequency may affect, in opposite ways, the frequency of the network with shunting synapses depending on the excitatory synaptic conductance and the magnitude of leak conductance.     

Neural networks for improved target differentiation and localization with sonar.

This study investigates the processing of sonar signals using neural networks for robust differentiation of commonly encountered features in indoor robot environments. Differentiation of such features is of interest for intelligent systems in a variety of applications. Different representations of amplitude and time-of-flight measurement patterns acquired from a real sonar system are processed. In most cases, best results are obtained with the low-frequency component of the discrete wavelet transform of these patterns. Modular and non-modular neural network structures trained with the back-propagation and generating-shrinking algorithms are used to incorporate learning in the identification of parameter relations for target primitives. Networks trained with the generating-shrinking algorithm demonstrate better generalization and interpolation capability and faster convergence rate. Neural networks can differentiate more targets employing only a single sensor node, with a higher correct differentiation percentage (99%) than achieved with previously reported methods (61-90%) employing multiple sensor nodes. A sensor node is a pair of transducers with fixed separation, that can rotate and scan the target to collect data. Had the number of sensing nodes been reduced in the other methods, their performance would have been even worse. The success of the neural network approach shows that the sonar signals do contain sufficient information to differentiate all target types, but the previously reported methods are unable to resolve this identifying information. This work can find application in areas where recognition of patterns hidden in sonar signals is required. Some examples are system control based on acoustic signal detection and identification, map building, navigation, obstacle avoidance, and target-tracking applications for mobile robots and other intelligent systems.     

Automated staging of sleep in cats using neural networks.

Manual staging of sleep based on visual EEG criteria is a laborious and time-consuming task. In an effort to automate sleep staging, we have developed a neural network that 'learns' to stage sleep on the basis of wave band count data alone, in the cat. Wave band count data are collected on a microcomputer, using period-amplitude analysis. Delta waves, spindle bursts, ponto-geniculo-occipital (PGO) waves, electro-oculogram (EOG), basal electromyogram (EMG) amplitude, and movement artifact amplitude are collected, and used to 'train' the network to score sleep. These wave count data serve as the input patterns to the net, and the corresponding manually scored sleep stages serve as a 'teacher.' We demonstrate that, when used to score the states of wake, slow wave sleep (SWS), desynchronized sleep (D), and the transition period from SWS to D (SP), these neural networks agree with manual scoring an average of 93.3% for all epochs scored. Neural network programs can learn both rules and exceptions, and since the nets teach themselves these rules automatically, a minimum of human effort is required. Because programming requirements are small for neural nets, this approach is readily adaptable to microcomputer-based systems and is widely applicable to both animal and human EEG analyses. The utility of this approach for the detection and classification of a variety of clinical neurophysiological disorders is discussed.     

Prediction of the side of hand movements from single-trial multi-channel EEG data using neural networks.

Thirty channels of EEG data were recorded prior to voluntary right or left hand movements. Event-related desynchronization (ERD) was quantified in the 8-10 Hz and 10-12 Hz bands in single-trial data and used as training input for a neural network comprised of a learning vector quantizer (LVQ). After a training period, the network was able to predict the side of hand movement from single-trial EEG data recorded prior to movement onset.     

Connectionist networks learn to transmit chaos

Evidence presented in the preceding paper indicates that the activity of some neurons during the generation of coordinated motor patterns may be attributable to chaos. Because even "simple" biological systems are difficult to control, we have used connectionist networks in order to inquire into the question of whether a chaotic signal originating in one part of the nervous system can be learned and transmitted by another. We have examined a number of different architectures, and report here the findings for a simple network consisting of one input unit, four hidden units, and one output unit. During training sessions, the input of the circuit was given analog values of either the 3.60 or 3.95 logistic equation, or of one variable of the three-variable R?ssler attractor. The backpropagated error in the learning algorithm was a function of the difference between the input value and the output at each iteration. Iterations involving small changes in analog value resulted in good similarity between the input and output signals, but little learning occurred because of the small error propagated back to the synapses. With larger differences in the analog values (and larger feedback error) at each iteration, we found that networks learned to transmit different chaotic attractors. Once the network learned one input, it could transmit another without changing the synapses. Increasing the number of hidden units increased the rate of learning.(ABSTRACT TRUNCATED AT 250 WORDS)     

The vanishing point of the mismatch negativity at sleep onset.

OBJECTIVE: To determine when the mismatch negativity (MMN) disappears at sleep onset, event-related potentials (ERPs) were recorded continuously from wakefulness to sleep. METHODS: Ten healthy young students were told to fall asleep ignoring the tones presented through a loudspeaker above their head. Standard (1000 Hz, P=0.90), high deviant (1200 Hz, P=0.05), and low deviant (1050 Hz, P=0.05) tones were presented in a quasirandom order with a fixed stimulus onset asynchrony of 500 ms. ERP waveforms were obtained separately for 5 successive stages characterized by typical electroencephalographic (EEG) patterns of the sleep onset period. The EEG staging was made manually with very short (5 s) scoring epochs. RESULTS: The MMN appeared in wakefulness and in the early phase of stage 1 sleep (EEG stage II) but disappeared when low-voltage theta waves emerged after alpha flattening (EEG stage III). Instead, P240 and N360 developed particularly for high deviant tones. CONCLUSIONS: Concurrently with the disappearance of alpha waves, the automatic change detection system in wakefulness seems to stop operating and a different sleep-specific system becomes dominant.     

Neural network tomography: Network replication from output surface geometry

Multilayer perceptron networks whose outputs consist of affine combinations of hidden units using the tanh activation function are universal function approximators and are used for regression, typically by reducing the MSE with backpropagation. We present a neural network weight learning algorithm that directly positions the hidden units within input space by numerically analyzing the curvature of the output surface. Our results show that under some sampling requirements, this method can reliably recover the parameters of a neural network used to generate a data set.     

The functional connectivity of different EEG bands moves towards small-world network organization during sleep.

OBJECTIVE: To analyze the functional connectivity patterns of the different EEG bands during wakefulness and sleep (different sleep stages and cyclic alternating pattern (CAP) conditions), using concepts derived from Graph Theory. METHODS: We evaluated spatial patterns of EEG band synchronization between all possible pairs of electrodes (19) placed over the scalp of 10 sleeping healthy young normal subjects using two graph theoretical measures: the clustering coefficient (Cp) and the characteristic path length (Lp). The measures were obtained during wakefulness and the different sleep stages/CAP conditions from the real EEG connectivity networks and randomized control (surrogate) networks (Cp-s and Lp-s). RESULTS: We found values of Cp and Lp compatible with a small-world network organization in all sleep stages and for all EEG bands. All bands below 15Hz showed an increase of these features during sleep (and during CAP-A phases in particular), compared to wakefulness. CONCLUSIONS: The results of this study seem to confirm our initial hypothesis that during sleep there exists a clear trend for the functional connectivity of the EEG to move forward to an organization more similar to that of a small-world network, at least for the frequency bands lower than 15Hz. SIGNIFICANCE: Sleep network "reconfiguration" might be one of the key mechanisms for the understanding of the "global" and "local" neural plasticity taking place during sleep.     

Connectionism and an inescapable defect

Linear connectionist models of neurocomputing show how input patterns may be recognized, stored, compared, and recalled for output in serial-parallel, quasi-Hebbian networks. This aids the design of hardware and software for better robotics, while offering useful insights to neuroscientists studying sensorimotor systems, but connectivity via quasi-Hebbian nodes and back-propagation layers alone cannot show us how vertebrate cerebellum, allocortex, and neocortex work.