Artificial neural network detects changes in electro-encephalogram power spectrum of different sleep-wake states in an animal model of heat stress

An effective application is presented of a back-propagation artificial neural network (ANN) in differentiating electro-encephalogram (EEG) power spectra of stressed and normal rats in three sleep-wakefulness stages. The rats were divided into three groups, one subjected to acute heat stress, one subjected to chronic heat stress and one a handling control group. The polygraphic sleep recordings were performed by simultaneous recording of cortical EEG, electro-oculogram (EOG) and electromyogram (EMG) on paper and in digital form on a computer hard disk. The preprocessed EEG signals (after removal of DC components and reduction of base-line movement) were fragmented into 2s artifact-free epochs for the calculation of power spectra. The slow-wave sleep (SWS), rapid eye movement (REM) sleep and awake (AWA) states were analysed separately. The power spectrum data for all three sleep-wake states in the three groups of rats were tested by a back-propagation ANN. The network contained 60 nodes in the input layer, weighted from power spectrum data from 0 to 30 Hz, 18 nodes in the hidden layer and an output node. The ANN was found effective in differentiating the EEG power spectra from stressed to normal spectral patterns following acute (92% in SWS, 85.5% in REM sleep, 91% in AWA state) as well as chronic heat exposure (95.5% in SWS, 93.8% in REM sleep, 98.5% in AWA state).     

A neural network confirms that physical exercise reverses EEG changes in depressed rats

The use of an artificial neural network (ANN) system to differentiate the EEG power density spectra in depressed from normal rats was tried. The beneficial effects of chronic physical exercise in reducing the effects of stress and therefore depression was also to be tested in animals by the same method. In this study, rats were divided into 4 groups, subjected to (i) chronic stress (D group); (ii) chronic exercise by treadmill running (EO group); (iii) exercise with stress (ES group) and (iv) handling (C group). The prefrontal cortical EEG, EMG and EOG were recorded simultaneously on paper and the digitized EEG signals were also stored in the hard-disk of a PC-AT through an ADC. After filtering the digitize signals, the EEG power spectra were calculated by an FFT routine. Three successive 4 s artefact-free epochs were averaged. The REM and NREM sleep periods as well as the awake period signals were analyzed separately. The FFT values from each of the 3 states, in the 4 groups of animals were tested by an ANN with 30 first layer neurons and a 2nd layer of a majority-vote-taker. The ANN could distinguish the depressed from the normal rats' EEG very well in REM (99%) sleep, NREM (95%) sleep and awake (81%) states. In most of the cases it identified the exercised rats' EEG as normal.     

Neural Network-Based Evaluation of Chronic Non-Thermal Effects of Modulated 2450 MHz Microwave Radiation on Electroencephalogram

The effects of chronic exposure (2 h daily for 21 days) of 1 kHz square wave-modulated 2450 MHz microwave radiation (non-thermal) on sleep-EEG, open field behavior, and thyroid hormones (T₃, T₄, and TSH) have been analyzed in an animal model. Results revealed significant changes in these pathophysiological parameters (p < 0.05 or better), except body temperature, grooming behavior, and TSH levels. The sleep-EEG power spectrum data for slow wave sleep (SWS), rapid eye movement (REM) sleep, and awake (AWA) states in two experimental groups of rats (microwave exposed and the control) were tested by an artificial neural network (ANN), containing 60 nodes in input layer, weighted from power spectrum data from 0 to 30 Hz, 18 nodes in hidden layer and an output node. The target output values for this network were determined with another five-layered neural network (with the structure of 6-14-1-14-6). The input and output of this network was assigned with the six confirmed pathophysiological changes. The most important feature for chronic exposure of 2450 MHz microwave exposure and for control subjects was extracted from the third layer single neuron and used as the target value for the three-layered ANN. The network was found effective in recognizing the EEG power spectra with an average of 71.93% for microwave exposure and 93.13% for control subjects, respectively. However, the lower percentage of pattern identification agreement in the microwave-exposed group in comparison to the control group suggest only mild effects of microwave exposure with this experimental setup.     

Automatic recognition of alertness and drowsiness from EEG by an artificial neural network

We present a novel method for classifying alert vs drowsy states from 1 s long sequences of full spectrum EEG recordings in an arbitrary subject. This novel method uses time series of interhemispheric and intrahemispheric cross spectral densities of full spectrum EEG as the input to an artificial neural network (ANN) with two discrete outputs: drowsy and alert. The experimental data were collected from 17 subjects. Two experts in EEG interpretation visually inspected the data and provided the necessary expertise for the training of an ANN. We selected the following three ANNs as potential candidates: (1) the linear network with Widrow-Hoff (WH) algorithm; (2) the non-linear ANN with the Levenberg-Marquardt (LM) rule; and (3) the Learning Vector Quantization (LVQ) neural network. We showed that the LVQ neural network gives the best classification compared with the linear network that uses WH algorithm (the worst), and the non-linear network trained with the LM rule. Classification properties of LVQ were validated using the data recorded in 12 healthy volunteer subjects, yet whose EEG recordings have not been used for the training of the ANN. The statistics were used as a measure of potential applicability of the LVQ: the t-distribution showed that matching between the human assessment and the network output was 94.37+/-1.95%. This result suggests that the automatic recognition algorithm is applicable for distinguishing between alert and drowsy state in recordings that have not been used for the training.     

Electro-encephalogram disturbances in different sleep-wake states following exposure to high environmental heat

In this study, cerebral electrical activity or electro-encephalogram (EEG) was studied following exposure to high environmental heat, in three different age groups of freely moving rats. Each age group was subdivided into three groups: the acute heat stress group, subjected to a single exposure of 4h at 38°C in the biological oxygen demand incubator; the chronic heat stress group, exposed for 21 days, for 1 h each day, at 38°C in the incubator; and the handling control group. The polygraphic sleep-wake recordings involved simultaneous recordings of cortical EEG, electrooculogram (EOG), and electromyogram (EMG), on paper and in digital form on computer hard disk, just after the heat exposure for the acute stressed rats and on the 22nd day for the chronic stressed rats. The power spectrum was calculated for 2s epochs of the EEG signals. Quantitative analyses of EEG (qEEG) showed that, in all three age groups, changes in higher-frequency components (β₂) were significant in all sleep-wake states following both acute and chronic heat stress conditions. The power of β₂ activity in all three age groups after acute heat exposure was significantly decreased during slow wave sleep (SWS) (p<0.05) and rapid eye movement sleep (p<0.05), whereas the reverse was observed in the awake state (p<0.05). Following chronic heat exposure, β₂ activity was found to increase in all three sleep-wake stages in all groups of rats (p<0.01 for SWS in the weaning group and p<0.05 for other data). Thus the study demonstrated that the cortical EEG is sensitive to environmental heat, and alterations in EEG frequencies in different states of mental consciousness due to high heat can be differentiated efficiently by EEG power spectrum analysis.     

Artificial neural network analysis of gustatory responses in the thalamic taste relay of the rat

We used an artificial neural network (ANN) as a model for analyzing single-neuron responses from the thalamic taste relay of rats. The network consisted of: (1) a layer of 44 input units, representing the responses of the 44 thalamic taste cells; (2) a layer of hidden units of varying numbers; and (3) a layer of four output units. We used the back-propagation algorithm to train the output units to discriminate among tastants based on inputs from the thalamic neurons. As the network became fully trained, we found that: (1) only two hidden units were necessary to provide nearly the full discriminative capacity of the network; (2) the loss of even a few of the input units that had the highest impact on hidden units caused a drastic reduction of discriminative power, implying that not all neurons contribute equally to the neural code; and (3) adding a temporal component to the input, by representing each 100-ms time bin as a separate input unit, increased the accuracy with which output units were able to identify tastants. We used data from behavioral discrimination tasks as a measure of the capacity of the network to identify stimuli correctly. A network with two hidden units was about as effective as an across-pattern analysis in accounting for the rat's discriminative ability.     

A set of EEG parameters to predict clinically anaesthetized state in humans for halothane anaesthesia

This article evaluates all the EEG parameters suggested in the literature that undergo changes due to anaesthetic dose, and suggests a set of EEG parameters that act as best signatures of anaesthetic state of a patient. This set of EEG parameters is validated by an artificial neural network. PRIMARY OBJECTIVE: The purpose of this study is to arrive at a set of EEG parameters that 'best' distinguish between awake and anaesthetized states of human patients for halothane anaesthesia. METHODS AND PROCEDURES: A total of 21 EEG parameters were evaluated for 40 patients. Stepwise discriminant analysis (SDA) pruned them to a set of five parameters. They were fed to a 5-3-1 artificial neural network (ANN) for classification into awake and anaesthetized state. To confirm the results, variance analysis was applied to the set of 21 parameters. Five parameters were finalized after validation by the ANN. MAIN OUTCOMES AND RESULTS: The classification accuracy of the ANN with SDA parameters was found to be 96%. With variance analysis parameters, it returned an accuracy of 100%. CONCLUSION: The set of five EEG parameters - approximate entropy, average frequency, Lempel Ziv (LZ) complexity, delta power and beta power forms the best set to distinguish between awake and anaesthetized state of human patients. Variance analysis is a better tool to converge at the optimal set than SDA.     

基于BP网络的睡眠分阶方法及睡眠质量评估研究

我们利用不同睡眠期脑电复杂性特征与睡眠深度的关系及多道睡眠图功率谱特征,基于3层BP网络进行了睡眠自动分阶的研究,并提出了能部分反映睡眠质量的睡眠时间、浅睡时间、深睡时间、REM时间、觉睡比、醒转次数等参数。通过6例全睡眠监护实验说明,该方法可为睡眠质量的评价提供途径。     

MDA Publications

This article evaluates all the EEG parameters suggested in the literature that undergo changes due to anaesthetic dose, and suggests a set of EEG parameters that act as best signatures of anaesthetic state of a patient. This set of EEG parameters is validated by an artificial neural network.

Primary objective: The purpose of this study is to arrive at a set of EEG parameters that ‘best’ distinguish between awake and anaesthetized states of human patients for halothane anaesthesia.

Methods and procedures: A total of 21 EEG parameters were evaluated for 40 patients. Stepwise discriminant analysis (SDA) pruned them to a set of five parameters. They were fed to a 5–3–1 artificial neural network (ANN) for classification into awake and anaesthetized state. To confirm the results, variance analysis was applied to the set of 21 parameters. Five parameters were finalized after validation by the ANN.

Main outcomes and results: The classification accuracy of the ANN with SDA parameters was found to be 96%. With variance analysis parameters, it returned an accuracy of 100%.

Conclusion: The set of five EEG parameters - approximate entropy, average frequency, Lempel Ziv (LZ) complexity, delta power and beta power forms the best set to distinguish between awake and anaesthetized state of human patients. Variance analysis is a better tool to converge at the optimal set than SDA.     

用脑电互信息及复杂度预测异氟醚麻醉时患者对切皮刺激的反应

自从麻醉应用于临床以来，麻醉深度的可靠监测是十分必要的。但到目前，尚没有一个公认可靠准确的方法。本文提出一种麻醉深度监测的新方法，即用脑电的互信息序列及其复杂度分析来反应异氟醚麻醉条件下患者的麻醉情况。首先计算出四导脑电的互信息时间序列，然后计算该序列的复杂性测度，借助于神经网络可实现用脑电来监测麻醉深度。神经网络的输入是复杂度值和对应的MAC水平，输出即是麻醉深度状况的结果。从98个自愿患者进行的实验中得到98个不同程度异氟醚麻醉时切皮前脑电片断，同时监测血液动力学参数和患者的呼吸模式。切皮后，仔细观察每个患者两分钟，以检查患者对切皮的反应，把有反应时的脑电标上0．0，无反应时的脑电标上1．0。训练和测试神经网络用“去掉一个”方法。从患者对切皮的反应和神经网络的输出结果可检测系统的预测情况。实验表明，系统对切皮后患者反应的平均正确判断率为91．84％，该方法比传统脑电分析方法如边缘康率法、中心频率法、双谱分析法有更高的准确性。另外，该方法计算时间短，适合临床实时使用。     

Psychological stress is associated with heightened physiological arousal during NREM sleep in primary insomnia

The objective of this study was to evaluate cross-sectional relationships among symptoms of psychological stress, sleep, and physiological arousal during non-rapid eye movement (NREM) sleep in a sample of 30 patients with chronic, primary insomnia (mean age, 30.2 years, 60% female). Study measures included indexes of subjective stress, visually scored sleep, and physiological arousal during NREM sleep: quantitative electroencephalogram (QEEG) and quantitative electrocardiogram (QEKG) measures. Psychological stress was more strongly related to indexes of physiological arousal during NREM sleep than to visually scored measures of sleep. Higher levels of perceived stress were associated with decreased EEG delta power (rho = -0.50, p < .01) and increased EEG beta power (rho = 0.38, p < .05). Increased frequency of stress-related avoidance behaviors was associated with decreased EKG high-frequency power (rho = -0.46, p < .05). Although QEEG measures were significantly correlated with sleep maintenance (QEEG delta power rho = 0.45, p < .01; QEEG beta power rho = -0.54, p < .01) and time spent in delta sleep (QEEG delta power rho = 0.65, p < .001; QEEG beta power rho = -0.65, p < .001), QEKG measures were unrelated to visually scored measures of sleep. Perceived stress and stress-related avoidance behaviors were associated with multiple indexes of physiological arousal during NREM sleep in patients with chronic, primary insomnia.     

Changes in sleep on chronic exposure to warm and cold ambient temperatures

Alterations in sleep induced by chronic exposure to mild changes in ambient temperature (Ta) were studied in male Wistar rats with chronically implanted electrodes for recording electrooculogram (EOG), electroencephalogram (EEG) and electromyogram (EMG), and a thermocouple to record the brain temperature (Tbr). Changes in sleep-wakefulness (S-W) and Tbr on exposure to warm (30+/-1 degrees C) and cold (18+/-1 degrees C) Ta for 4 weeks were studied in two groups of five rats each. Chronic heat exposure produced a persistent increase in sleep, primarily due to an increase in the durations of sleep episodes. A disproportionate increase in sleep during the dark period resulted in reduced circadian variation. The paradoxical sleep (PS)/total sleep time (TST) ratio also remained increased, during heat exposure. On chronic cold exposure, the sleep was decreased initially, but it recovered after 3 weeks, due to an increase in the frequency of slow wave sleep (SWS) episodes. The Tbr was not altered on exposure to warm Ta, but it remained high throughout the 4 weeks of cold exposure. The increase in the amount of sleep, especially the PS with enhanced ambient temperature, may be considered as an adaptation to thermal load aimed at energy conservation. Though the increased wakefulness is suggested to enable the organism to optimize thermoregulation during acute cold stress, thermoregulation itself may be readjusted to ensure homeostatic restoration of sleep during chronic cold exposure.     

Energy based feature extraction for classification of sleep apnea syndrome

In this paper it is aimed to classify sleep apnea syndrome (SAS) by using discrete wavelet transforms (DWT) and an artificial neural network (ANN). The abdominal and thoracic respiration signals are separated into spectral components by using multi-resolution DWT. Then the energy of these spectral components are applied to the inputs of the ANN. The neural network was configured to give three outputs to classify the SAS situation of the subject.The apnea can be mainly classified into three types: obstructive sleep apnea (OSA), central sleep apnea (CSA) and mixed sleep apnea (MSA). During OSA, the airway is blocked while respiratory efforts continue. During CSA the airway is open, however, there are no respiratory efforts. In this paper we aim to classify sleep apnea in one of three basic types: obstructive, central and mixed. A significant result was obtained.     

Significance of distributed representation in the output layer of a neural network in a pattern recognition task

In the cerebral cortex, it is assumed that information is represented by the activity pattern of an assembly of neurons and the synaptic efficacies among them. A distributed representation of pattern is incorporated in the output layer of a neural network with an error back-propagation algorithm, in order to study its technological merits. The network has three layers, which consist of a 32×32 array of units (1024) for the input layer, 6–25 units for the hidden layer and 12 units for the output layer. 12 triangular patterns with a variety of parameters are presented to the input layer. Three output-layer units are assigned to each input figure. After initial learning, the network responds to the learned pattern with high accuracy. In addition, it responds with high accuracy to similar but unpresented patterns, showing a generalisation for patterns. The network shows resistance to unit de-activation procedures. When the input layer is exposed to the learned pattern, the hidden-layer units show associative activation pattern. These results indicate that the organisation of information representation in the output layer in a neural network strongly influences both the performance of the whole network and information representation in the hidden layer.     

Circadian variation of EEG power spectra in NREM and REM sleep in humans: Dissociation from body temperature

In humans, EEG power spectra in REM and NREM sleep, as well as characteristics of sleep spindles such as their duration, amplitude, frequency and incidence, vary with circadian phase. Recently it has been hypothesized that circadian variations in EEG spectra in humans are caused by variations in brain or body temperature and may not represent phenomena relevant to sleep regulatory processes. To test this directly, a further analysis of EEG power spectra - collected in a forced desynchrony protocol in which sleep episodes were scheduled to a 28-h period while the rhythms of body temperature and plasma melatonin were oscillating at their near 24-h period - was carried out. EEG power spectra were computed for NREM and REM sleep occurring between 90-120 and 270-300 degrees of the circadian melatonin rhythm, i.e. just after the clearance of melatonin from plasma in the 'morning' and just after the 'evening' increase in melatonin secretion. Average body temperatures during scheduled sleep at these two circadian phases were identical (36.72 degrees C). Despite identical body temperatures, the power spectra in NREM sleep were very different at these two circadian phases. EEG activity in the low frequency spindle range was significantly and markedly enhanced after the evening increase in plasma melatonin as compared to the morning phase. For REM sleep, significant differences in power spectra during these two circadian phases, in particular in the alpha range, were also observed. The results confirm that EEG power spectra in NREM and REM sleep vary with circadian phase, suggesting that the direct contribution of temperature to the circadian variation in EEG power spectra is absent or only minor, and are at variance with the hypothesis that circadian variations in EEG power spectra are caused by variations in temperature.     

Evolution of sleep and sleep EEG after hemispheric stroke

The evolution of subjective sleep and sleep electroencephalogram (EEG) after hemispheric stroke have been rarely studied and the relationship of sleep variables to stroke outcome is essentially unknown. We studied 27 patients with first hemispheric ischaemic stroke and no sleep apnoea in the acute (1-8 days), subacute (9-35 days), and chronic phase (5-24 months) after stroke. Clinical assessment included estimated sleep time per 24 h (EST) and Epworth sleepiness score (ESS) before stroke, as well as EST, ESS and clinical outcome after stroke. Sleep EEG data from stroke patients were compared with data from 11 hospitalized controls and published norms. Changes in EST (>2 h, 38% of patients) and ESS (>3 points, 26%) were frequent but correlated poorly with sleep EEG changes. In the chronic phase no significant differences in sleep EEG between controls and patients were found. High sleep efficiency and low wakefulness after sleep onset in the acute phase were associated with a good long-term outcome. These two sleep EEG variables improved significantly from the acute to the subacute and chronic phase. In conclusion, hemispheric strokes can cause insomnia, hypersomnia or changes in sleep needs but only rarely persisting sleep EEG abnormalities. High sleep EEG continuity in the acute phase of stroke heralds a good clinical outcome.     

Discrimination of cerebral ischemic states using bispectrum analysis of EEG and artificial neural network

No doubt a noninvasive technique for detection of cerebral ischemic extent, before the formation of the focus, is extremely valuable. This paper presents a new approach to early evaluate the degree of ischemic injury by combining bispectrum estimation of electroencephalograms (EEGs) with artificial neural network (ANN). The graded ischemic injuries in 24 Sprague-Dawley (SD) rats were induced for different periods of 8, 18, 30 min by infusing physiological saline along the left blood stream, based on the model for rat ischemic cerebral injury described in this paper. Four channels of EEG were collected in each rat at scheduled time of ischemia. The maximum bicoherence index and the weighted center of bispectrum (WCOB) were extracted from the EEGs and were used as input feature vector of a four-layer (12-7-2-1) ANN for prediction. Training and testing the ANN used the 'leave one out' strategy. The levels of ischemic injury were verified and classified by observing the ischemic area by conventional hematoxylin and eosin (HE) staining and the heat shock protein (HSP70) test. The proposed method was able to correctly detect ischemic extent in average accuracy of 91.67% of the cases. The results show that this scheme can be expected to diagnose ischemic cerebral injury in its earlier phases.     

Automatic recognition of rapid eye movement (REM) sleep by artificial neural networks

SUMMARY  Artificial neural networks are well known for their good performance in pattern recognition. Their suitability for detecting REM sleep periods on the basis of preprocessed EEG data in humans under clinical conditions was tested and their performance compared with the manual evaluation. A single channel of the EEG signal was analysed in time periods of 20s and preprocessed into a vector of six real numbers, which served as input to the network. EOG and EMG information was ignored. Backpropagation was used as a learning rule for the network, which consisted of 12 neurons and 39 synapses. Training datasets were put together from the input vectors and the corresponding sleep stages were scored manually. In working mode different networks were compared in terms of the rate of misclassified time periods for data not belonging to the training sets. The indicator function of REM sleep was well approximated by the network output in the course of the night, which was especially true for REM onsets. The average rate of correctly classified time periods was 89%. The errors were analysed and suggestions for improvements developed.     

Normal sleep homeostasis and lack of epilepsy phenotype in GABA A receptor alpha3 subunit-knockout mice

Thalamo-cortical networks generate specific patterns of oscillations during distinct vigilance states and epilepsy, well characterized by electroencephalography (EEG). Oscillations depend on recurrent synaptic loops, which are controlled by GABAergic transmission. In particular, GABA A receptors containing the alpha3 subunit are expressed predominantly in cortical layer VI and thalamic reticular nucleus (nRT) and regulate the activity and firing pattern of neurons in relay nuclei. Therefore, ablation of these receptors by gene targeting might profoundly affect thalamo-cortical oscillations. Here, we investigated the role of alpha3-GABA A receptors in regulating vigilance states and seizure activity by analyzing chronic EEG recordings in alpha3 subunit-knockout (alpha3-KO) mice. The presence of postsynaptic alpha3-GABA A receptors/gephyrin clusters in the nRT and GABA A-mediated synaptic currents in acute thalamic slices was also examined. EEG spectral analysis showed no difference between genotypes during non rapid-eye movement (NREM) sleep or at waking-NREM sleep transitions. EEG power in the spindle frequency range (10-15 Hz) was significantly lower at NREM-REM sleep transitions in mutant compared with wild-type mice. Enhancement of sleep pressure by 6 h sleep deprivation did not reveal any differences in the regulation of EEG activities between genotypes. Finally, the waking EEG showed a slightly larger power in the 11-13-Hz band in alpha3-KO mice. However, neither behavior nor the waking EEG showed alterations suggestive of absence seizures. Furthermore, alpha3-KO mice did not differ in seizure susceptibility in a model of temporal lobe epilepsy. Strikingly, despite the disruption of postsynaptic gephyrin clusters, whole-cell patch clamp recordings revealed intact inhibitory synaptic transmission in the nRT of alpha3-KO mice. These findings show that the lack of alpha3-GABA(A) receptors is extensively compensated for to preserve the integrity of thalamo-cortical function in physiological and pathophysiological situations.     

一种基于EEG特征提取的癫痫棘波综合检测判决方法

癫痫特征的自动检测在临床应用上具有重要的意义。本研究综合小波变换、非线性能量算子、特征提取和神经网络等技术，提出了一种癫痫棘波检测系统，充分发挥各技术的优点，在对真实脑电数据的处理中，表现出良好的性能。