Journal of Clinical Monitoring and Computing (2008) 22:261–268 DOI: 10.1007/s10877-008-9128-x

EEG POWER SPECTRUM AND NEURAL NETWORK BASED SLEEP-HYPNOGRAM ANALYSIS FOR A MODEL OF HEAT STRESS Rakesh Kumar Sinha, PhD

 Springer 2008

Sinha RK. EEG power spectrum and neural network based sleephypnogram analysis for a model of heat stress. J Clin Monit Comput 2008; 22:261–268

ABSTRACT. Objective. An effective application of backpropagation artificial neural network (ANN) in preparation of sleep-hypnogram based on electroencephalogram (EEG) power spectra under acute as well as chronic heat stress has been presented. Methods. Rats were divided in three groups (i) acute heat stress—subjected to a single exposure for four hours at 38
C; (ii) chronic heat stress—exposed for 21 days daily for one hour at 38
C, and (iii) handling control groups. The preprocessed EEG signals were fragmented in two-second artifact free epochs for calculation of power spectra, training and testing of ANN. Results. The power spectrum analyses of EEG show that changes in higher frequency components (b2) were significant in all sleep-wake states following both acute and chronic heat stress conditions. The power of b2 activity after acute heat exposure was significantly decreased during SWS (slow wave sleep) (P < 0.05) and REM (rapid eye movement) sleep (P < 0.05), while reverse was observed in AWA (awake state) (P < 0.05). Following chronic heat exposure, b2 activity was found increased in all three sleep-wake stages (P < 0.05). The ANN used for sleep-hypnogram preparation contains 64 nodes in input layer, weighted from power spectrum data from 0 to 32 Hz, 14 nodes in hidden layer and 3 output nodes. The results obtained from the study, suggest increased sleep efficiency following acute exposure to heat stress while fragmented sleep with decreased sleep efficiency following chronic heat stress. Conclusion. The ANN can be used for the analysis of stressful events by calculating the sleep-EEG alterations. KEY WORDS. ANN, hypnogram.

EEG

Power

spectrum,

Rat,

Sleep-

INTRODUCTION

From the Department of Biomedical Instrumentation, Birla Institute of Technology, Mesra, Ranchi, Jharkhand 835215, India. Received 4 March 2008. Accepted for publication 15 May 2008. Address correspondence to R. K. Sinha, Department of Biomedical Instrumentation, Birla Institute of Technology, Mesra, Ranchi, Jharkhand 835215, India. E-mail: [email protected]

The review of literature on the effects of high environmental heat on sleep variables has suggested that the acute exposure to thermal environment decreases the waking episodes (AWA) with a shift toward increase in slow wave sleep (SWS) and rapid eye movement (REM) sleep. Recently it was well demonstrated that high environmental temperature (30
C) increases the deep sleep in rats and total SWS episodes after longer duration of exposure to 24
C and 30
C of ambient temperature [1, 2]. In different models of chronic heat stress, it has been observed that the chronic exposure to environmental heat stress reduced the total sleep time (TST), while the amount of wakefulness increased. The mean duration of REM episodes and the REM cycle length were also reported to become shorter [3, 4]. The most accurate way to assess the depth of sleep is through polygraphic recording of electrophysiological

262

Journal of Clinical Monitoring and Computing

signals. It has been done for the last several decades in different psychiatric and clinical electroencephalography laboratories to extract information, which are not available in awake and resting states. Sleep investigators have demonstrated that sleep stages in mammals can be classified by the recording of three channels of bioelectric signals, such as EEG (electroencephalogram), EOG (Electrooculogram) and EMG (Electromyogram) [5, 6]. Further, manual scoring of the effect of heat stress on different sleep-wake stages is very difficult and takes great deal of time and effort. It has also been shown that the performance of computer assisted pattern classification is very sensitive to the inter-observer variations in manual classification [7]. Many automated systems including different soft computing based algorithms are available for the classification of sleep stages. However, very few reports are existing that detects the sleep-EEG variations under psychophysiological states [8–10]. Among different automated sleep-wake classifiers, backpropagation artificial neural networks (ANN) in combination with fast Fourier Transform (FFT) or power spectrum of EEG signals have successfully used for many automated identification and classification problems associated with sleepEEG [11–16]. Motivated by the confirmed sleep-EEG spectral changes due to acute and chronic heat stress [15, 17], in the present study, a system has been designed with the help of EEG power spectrum and ANN to identify the sleep-wake variations due to acute and chronic heat exposure in rats.

Heat stress model The stress was produced in the rats, by subjecting them in the Biological Oxygen Demand (BOD) incubator at preset temperature of 38 ± 1
C and relative humidity 45– 50% [10, 15], simulated with the environmental conditions of Varanasi (India) in the months of May and June. Acute heat stress Each rat of this group, following implantation of polygraphic electrodes and recovery from surgical stress was subjected for a single exposure in the incubator at the temperature of 38 ± 1
C for four hours from 8.00 A.M. to 12.00 P.M. IST, just before the recording of polygraphic signals. Chronic heat stress Rats were subjected in the incubator for one hour daily for 21 days of chronic heat exposure from 8.00 A.M. to 9.00 A.M. at 38 ± 1
C. The electrodes for polygraphic sleep recording were implanted on 14th day of chronic exposure to hot environmental heat. The recordings of electrophysiological signals were performed on 22nd day. Control Separate groups of rats were handled and processed as the acute and chronic stressed rats, respectively, but at controlled incubator temperature of 24 ± 1
C (same as the room temperature). These groups of rats were considered as controls for acute and chronic heat stressed groups for the comparison of different electrophysiological data.

MATERIALS AND METHODS

Polygraphic recordings Subjects and electrode implantation Polygraphic sleep data were recorded from male Charles Foster rats of age of 9–11 weeks. Rats were individually housed in polypropylene cages (30 cm  20 cm  15 cm) with drinking water and food ad libitum. The animal room was artificially illuminated with a 12 h light cycle, light during 7.00–19.00 h Indian Standard Time (IST) at 24 ± 1
C. For the present study, previously recorded sleep-EEG data were used [10, 15]. The whole procedure of electrode implantation for the recording of polygraphic sleep was conducted under Pentobarbital (35 mg/kg i.p.) anesthesia. After implantation of electrodes, animals were allowed for 7 days of post-operative recovery period before recording of the electrophysiological signals. All experimental procedures on rats were performed in compliance with ‘‘Committee for the purpose of control and supervision of experiments on animals (CPCSEA)’’, India.

Through 8-channel polygraph, continuous four hours recordings of single channel EEG, EOG and EMG were performed from 12.00 to 16.00 P.M. IST on the recording day for all stressed and control rats at room temperature (24 ± 1
C). EOG and EMG signals were recorded to confirm the sleep stages and EEG data were used for processing and analyses. The parameters of amplifier setting for different electrophysiological signals were set as described by Sarbadhikari et al. [6]. The digitized data, collected, stored and processed with the help of analog to digital converter (ADLiNK, 8112HG, NuDAQ, Taiwan) and its supporting data acquisition and processing software (VISUAL LAB-M, Version 2.0c, Blue Pearl laboratory, USA). All recordings were done with a sampling frequency of 256 Hz and selected data were stored at regular intervals in computer hard disk. Data were stored in two minutes segments in separate files.

Sinha: EEG Power Spectrum and ANN-based Sleep-Hypnogram Analysis

263

EEG signal processing

Statistical analyses

The recorded EEG signals were band pass filtered using infinite impulse response (IIR) Butterworth filter [18, 19] with lower cutoff of 0.25 Hz and higher cutoff of 35 Hz, as the frequency band of EEG between 0.5 and 30 Hz is having the most clinical importance in conscious state. The EEG signals were visually analysed as SWS (about 50% synchronized EEG signal, amplitude 50–300 lV), REM sleep (desynchronised EEG waves, amplitude > 40 lV) and AWA (synchronization of EEG < 25%, amplitude about 30 lV) according to the criteria of sleep-wake states as presented by Sarbadhikari et al. [6] for the power spectrum calculation. EEG data from all three sleep-wake episodes were fragmented to two-second epochs. Using FFT that was developed to calculate Discrete Fourier Transform (DFT) efficiently and rapidly performed the final processing of the EEG signals for the analysis of dominant frequency. The graph of the square of amplitudes of Fourier transform against frequency is referred as ‘Power Spectrum’. The linear scale power spectrum is converted in the dB scale, because it present better energy distribution pattern. A smoothed power spectral analysis method—simple moving average method was used to investigate the composition of the EEG power spectra before applying them for the training and testing of the ANN. The ‘Rectangular Window’ was used for all calculations of EEG power spectra in the present study. Quantitative changes in EEG power for each of five frequency bands, such as ¶ (between 0.5 and 3.99 Hz), h (between 4 and 7.99 Hz), a (between 8 and 12.99 Hz), b1 (between 13 and 17.99 Hz) and b2 (between 18 and 30 Hz) were analysed in the power spectrum of each epoch from stressed subjects with respect to the control rats. The EEG analyses were performed according to the method described earlier [6].

The statistical analyses (student’s t-test) for determining the effects of acute and chronic heat stress on EEG and sleep variables were manually performed in the laboratory and tested with standard statistical software package.

Artificial neural network model A backpropagation ANN [21] simulation program, written in C++ programming language was used for differentiating the EEG power spectra of different sleep-wake states. It has already been described earlier that single hidden layered ANN are universal approximator and universal classifier [22]. Thus, single hidden layered ANN architecture has been used in the present study. However, number of nodes in hidden layer has been fixed by trial and error method. The sigmoidal function was used in this study. The ANN used in the present study had 64 nodes in input, 14 nodes in the hidden and 3 nodes in the output layer. The Number of neurons in input layer was fixed by the digital values of EEG power spectrum data of three different sleep-wake states from 0 to 32 Hz with increment of 0.5 Hz. The first output is representing the EEG activity (synchronized/desynchronized EEG), second output represents the EOG activity (presence/absence of eye ball movement) and the third one shows the EMG activation (high/low muscle tone) (Figure 1). Thus, the SWS (100) are scored when there is synchronized EEG activity, REM sleep (010) is scored when there is desynchronized EEG activity with uneven EOG and AWA (001) is scored when input patterns with high EMG activity. For this study, other sets of outputs were treated as unclassified sleep patterns.

Sleep analysis Changes in sleep parameters after acute as well as chronic heat exposure with respect to their control groups, was analyzed according to the method described by Andersen and Tufik [20]. (a) Total sleep time (TST): Sum of all sleep periods during the recording, (b) Total time of SWS (TSWS): Sum of all periods of SWS throughout the recording, (c) Total time of REM sleep (TREM): Sum of all periods of REM sleep throughout the recording, (d) Total wake time (TWT): Sum of all periods of waking during the recording, (e) Latency of SWS (LSWS): Time lag between onset of recording and the first episode of SWS, (f) Latency of REM sleep (LREM): Time lag between onset of recording and the first episode of REM and (g) Stage shift (StS): Number of times that the animal changes from one sleep phase to the other.

Fig. 1. The sleep classification criteria have been analyzed with the help of the representative paper recording of EEG, EOG and EMG for SWS, REM sleep and AWA states, respectively.

264

Journal of Clinical Monitoring and Computing

Training and testing of ANN The training set of ANN contains 60 patterns of EEG power spectrum (20 from each of thee sleep-wake states), calculated from different rats and randomly arranged in a file ‘TRAINING.DAT’. For training, the error tolerance and learning rate parameters were assigned as 0.01 and 0.1, respectively, to activate the network. Once the stimulator reaches the error tolerance specified or achieved the maximum numbers of iterations, assigned for training, the stimulator save the state of the network by saving all its weights in ‘WEIGHT.DAT’. This file was subsequently used for the testing purpose. In testing mode, the ANN was provided a set of test data, similar to the training data but without assigning output value and stored in different test data files, which were named as ‘TEST.DAT’ before actual testing with the ANN. When each test file was applied to the trained network, the network goes through a cycle of operation, covering all test data sets and generated ‘OUTPUT.DAT’ file containing outputs from the network for all the input data sets. For the optimization of the ANN, different learning rates were investigated in the range of 0.01–0.5. The best performance obtained when the learning rate was 0.1, with which an overall accuracy of 100% for the 60 training sets (20 sets each from SWS, REM and AWA states) and 98.33% for the 120 manually differentiated, selected test sets (40 sets each from SWS, REM and AWA states). With a fixed number of iterations (1 millions), 14 hidden nodes resulted in the best performance compared with the other combination of hidden nodes. At initial

stage of training, the performance was found poor (nearly 20%) which improved quickly to nearly 60% after 0.2 million cycles, but after 0.4 million cycles, the ANN performance has nearly became moderately high and stagnant at average of 97.5% at the end of 1 million iterations. RESULTS

Sleep and EEG variations due to heat stress (a) Acute heat stress The quantitative analyses of changes in EEG power spectra in SWS, REM sleep and AWA conditions following acute heat exposure are shown in Table 1a. The percent of power of b2 was observed significantly decreased in SWS and REM sleep (P < 0.05), and the reverse occurred in AWA state (P < 0.05). Simultaneously, significant increase in the percent of ¶ activity in SWS was recorded (P < 0.05). Decrease in the percentage of a activity in SWS (P < 0.05) and increase in percentage b1 activity in REM sleep (P < 0.01) were observed. The statistical analyses for the results of sleep-wake parameters revealed significant alterations due to acute heat stress (Figure 2a). Due to acute exposure to high environmental heat, significant increase in TST and TSWS with corresponding decrease in TWT (P < 0.05) was analyzed. Alongside, The LSWS and LREM were also found significantly (P < 0.05) increased following the acute heat stress.

Table 1. EEG power analysis for (a) acute stress and (b) chronic stress groups (n = 5 has been taken for each group of subjects) SWS

(a) ¶ h a b1 b2 (b) ¶ h a b1 b2

REM

AWA

Control

Stress

Control

Stress

Control

Stress

21.86 22.38 22.06 18.76 15.33

(0.08) (0.08) (0.10) (0.19) (0.05)

25.02* (0.10) 22.30 (0.18) 20.75* (0.12) 18.91 (0.08) 12.97* (0.15)

32.00 25.91 19.63 12.23 11.17

(0.18) (0.15) (0.16) (0.07) (0.11)

31.04 (0.12) 25.15 (0.15) 19.64 (0.13) 15.53** (0.10) 9.54* (0.06)

33.51 24.11 18.67 12.08 11.69

(0.04) (0.13) (0.21) (0.09) (0.12)

32.54 (0.17) 23.61 (0.11) 18.41 (0.14) 11.65 (0.15) 13.70* (0.14)

21.15 22.48 21.76 18.92 15.62

(0.14) (0.17) (0.12) (0.11) (0.08)

21.34 (0.08) 22.88 (0.16) 20.37* (0.09) 17.42* (0.09) 18.95* (0.10)

34.00 23.91 19.36 11.93 10.77

(0.12) (0.05) (0.10) (0.12) (0.08)

31.75* (0.08) 23.98 (0.17) 18.93 (0.21) 13.72* (0.13) 11.50* (0.07)

36.51 23.00 16.70 11.08 12.18

(0.14) (0.12) (0.15) (0.07) (0.09)

34.76* (0.07) 23.92 (0.10) 16.66 (0.07) 11.35 (0.08) 13.26* (0.08)

Data are expressed in mean ± S.E. The percentage of EEG power of five frequency bands for SWS, REM sleep and AWA states following acute and chronic heat stress are compared to their respective control groups, *P < 0.05, **P < 0.01.

Sinha: EEG Power Spectrum and ANN-based Sleep-Hypnogram Analysis

265

heat stress. Beside these results, LSWS and LREM were also analyzed to significantly decrease (P < 0.05 and P < 0.01, respectively) due to chronic exposure to hot environment. The findings of the alterations in different sleep-wake variable are shown in Figure 2b.

ANN in sleep classification in heat stressed subjects

Fig. 2. Figures show the results of sleep-wake changes due to (a) acute as well as (b) chronic heat stress (n = 5 has been taken for each group of subjects). Data are expressed in mean ± S.E. Changes in each sleep-wake parameters following acute and chronic heat stress are compared *P < 0.05, **P < 0.01 to their respective control groups.

(b) Chronic heat stress The quantitative analyses of percentage variations in EEG power spectra in SWS, REM sleep and AWA conditions following chronic exposure to the hot environment are presented in Table 1b. Analyses of the results represent significantly increased percentage of power of b2 activity in all three sleep-wakefulness states (P < 0.05) such as SWS, REM and AWA, respectively. Other than these changes, percentage of a and b1 were observed significantly decreased in SWS (P < 0.05), and the percentage of b1 activity was decreased in REM sleep (P < 0.05). Consequently, the percentage of ¶ activity in REM and AWA conditions was observed significantly decreased (P < 0.05). In sleep-wake parameters, TST and TSWS were observed significantly decreased (P < 0.05) with unchanged TREM. Correspondingly, the statistical analyses of the data of waking episodes also showed a significant increase in TWT (P < 0.05) in this group of rats. The results showed fragmented sleep as significantly large number of StS was counted (P < 0.05) following chronic

Following optimization and training of the ANN, twosecond data from all three sleep-wake states, separately from each subject were arranged randomly and stored in test data files. For each subject, the recorded four hours of EEG data (7200 epochs) were analyzed for sleep-wake classification. For the testing purposes, continuous epochs obtained from processed EEG data were stored in test files. Each testing file consists of 60 test data sets. Thus, for each subject 120 test files were prepared. Following the established rule base as explained in Figure 3, ANN used in the present study smartly prepared the sleep-hypnogram for all individual subjects. Typical sleep-hypnogram for acute and chronic heat stressed subjects and their comparison with respective control rats have been presented in Figures 4a, b and 5a, b, respectively. The analyses of the results obtained from ANN based sleep-hypnogram was found strikingly similar with the manually analyzed results (Figure 2a, b). Due to acute exposure to high environmental heat, significant increase in SWS with longer time durations were obtained through ANN based analyses, which finally represents increase in TST and TSWS with respective decrease in TWT. Consequently, due to chronic heat exposure, fragmented sleep was observed. Irrespective to the results analyzed for acute heat stress group and similar to the manual analyses, the ANN based hypnogram for this group of rats showed significantly increased TWT with decrease in TSWS and TST.

Fig. 3. An idealized hypnogram of sleep-waking states. AWA, wakefulness; SWS, slow wave sleep; REM, rapid eye movement sleep; a, duration of AWA; b, duration of SWS; c, duration of REM. Transition between states: 1, AWA to SWS; 2, SWS to AWA; 3, SWS to REM; 4, REM to SWS, and 5, REM to AWA.

266

Journal of Clinical Monitoring and Computing

Fig. 4. Figures show hypnograms created with the help of backpropagation ANN. Typical sleep-hypnogram representation for an animal subjected to (a) acute exposure to high environmental heat and compared with (b) a control rat. Sleep-wake data were analyzed for four hours of sleep recordings. Analyses of results suggest increased sleep states following four hours of acute exposure to heat stress at 38
C. Keys: 1-AWA; 2-SWS and 3-REM sleep.

DISCUSSION

The effects of acute and chronic heat stress on similar animal model have been exclusively studied and established by the same group of researchers [4, 15, 17, 23]. The results of these studies suggested significant alterations in different pathophysiological variables due to both acute and chronic heat stress, which were found in accordance with other previously published reports [24–28]. In continuation, Sinha and Ray [4] have also demonstrated the effects of acute and chronic heat stress on various sleepwake parameters on three different age groups of rats and suggested a differential and age dependent variations in time and occurrence of different sleep-wake states.

Fig. 5. Backpropagation ANN based typical sleep-hypnogram representation for an animal subjected to (a) chronic heat stress and it’s comparison with (b) a control rat. Sleep-wake data were analyzed for four hours of sleep recordings. Analyses of results suggest fragmentized sleep with decreased sleep states following one hour daily for 21 days of chronic heat exposure at 38
C. Keys: 1-AWA; 2-SWS and 3-REM sleep.

Similar to earlier reports, the quantitative EEG analyses of acute exposed animals show that the power of the b2 activity was significantly decreased during SWS and REM sleep, whereas the reverse occurred in the AWA state. It has been assumed that acute heat stress altered the EEG frequencies that may have occurred owing to neuronal and non-neuronal changes in the central nervous system (CNS) [15, 29, 30]. On the other hand, the EEG changes following chronic heat exposure were found to be strikingly similar to the past reports on exercise as well as on chronic heat exposure [6, 10, 15]. Conversely, the analysis of sleep results revealed that acute heat stress is responsible for significant increase the TSWS and TST as well as decrease in TWT [1, 31]. This increase in SWS has been suggested to occur due to the active thermoregulatory response triggered to counter hyperthermia [32–34]. The

Sinha: EEG Power Spectrum and ANN-based Sleep-Hypnogram Analysis

results of this study confirm that chronic heat exposure results in adaptation of hypothalamo-pituitary-adrenal (HPA) system [35, 36]. The involvement of perifrontal lateral hypothalamic area (LHA) has also been implicated in sleep-wake control [37, 38]. Further, Alam et al. [38] have hypothesized that hypocretin neurons of LHA exhibit a wake related discharge patterns, with peak discharge in waking state. Thus, activation of hypocretin neurons of LHA following chronic exposure to hot environment may be responsible for the increase in AWA, with corresponding decrease in SWS, and finally results in decrease in TST in the present study. Artificial intelligence and expert systems have been tried for human sleep stage scoring [39] that reported a moderately good agreement with the manual scoring. Some other investigators have tried for automated sleep staging in animals by using variety of data collection and analyses methods [7, 11]. ANN pattern classifiers have also been used to evaluate changes in sleep-EEG power spectra in human [40] and rats [10, 16, 41]. The variations in the sensitivity of previous methods in comparison to the manual sleep staging and greater acceptability of ANNs in pattern recognition tasks projected its potentials of the effective use. Power spectrum by calculating FFT was one of the most popular approaches, which has largely been used as a data reduction tool for various long-term electrophysiologocal recordings including EEG [10, 16, 41–43]. Further, it has been known that the length of sleep-wake states, fragmentation in sleep and LSWS as well as REM sleep are highly sensitive to event the minute alterations in internal and external environmental variables of the subjects. The presented system also demonstrated that ANN based hypnogram preparation and its analysis can be used for the analysis of microsleep and number of sleep stage changes in variety of psychopathophysiological events. Considering the confirmed changes in EEG power spectra as well as with established pathophysiological and sleep-wake variations in similar animal model of acute and chronic heat stress, this work demonstrates that how effectively the ANN based sleephypnogram can be prepared and used for the study of effects of sleep-wake variables. The intension of the present investigation was not to calculate the percentage recognition rate for EEG patterns for SWS, REM sleep and AWA state. However, through the present approach, sleep-hypnogram has been prepared to identify the alterations in sleep-wake states due to acute as well as chronic heat stress. For the purpose, the misclassified data sets were considered similar as the previous correctly classified data. The present method has successfully used to create the sleep-hypnogram that also reveals the differential effects of acute and chronic heat stress on sleep-wake cycles. The results obtained from this study

267

strongly recommended that acute heat stress increases the sleep time and the duration of occurrence of SWS states, whereas chronic heat stress is responsible for reduction in sleep time and also with stage fragmentation. The results achieved from this method and ANN architecture was found very much analogous to the manual results as suggested by Sinha and Ray [4]. The success of the ANN in pattern classification involves the optimization of the network structure with the input parameters. Alongside, ANN learns to associate the given input pattern with the assigned output value and so, training and testing with more samples as well as by using different ANN architecture may improve the accuracy of identification. Based on several previous studies, one hidden layer was used, which suggested that one hidden layer resulted in the same performance as two or more hidden layers [22, 44, 45]. However, conflicting results were reported in the literature on the number of hidden nodes [46]. Therefore, the optimized number of hidden layer neurons were fixed and assigned by trial and error method. The presented method of ANN based sleep hypnogram preparation and analysis can allow the clinicians and sleep researchers in detecting micro-sleeps and frequent state transition with greater sensitivity. Furthermore, it can be said that ANN can provide an effective tool for recognition and determination of various sleep-wake parameters. The accuracy of sleep staging however, is sensitive to several parameters such as: the recording environment, the type of signals used, sample size, training method and the choice of network model. Preprocessing of signals has also critical impact upon the results. Thus, the pattern recognition by ANN and the clinical acumen, are however, not mutually exclusive and reinforce each other and it is believed that a human clinician must remain a necessary component of computerized diagnostic procedures to ensure a significant high level of diagnostic validity [9].

REFERENCES 1. Thomas TC, Kumar VM. Effect of ambient temperature on sleep-wakefulness in normal and medial preoptic area lesioned rats. Sleep Res Online 2000; 3: 141–145. 2. Thomas TC, Kumar VM. Effect of ambient temperature on brain temperature and sleep-wakefulness in medial preoptic area lesioned rats. Ind J Physiol Pharmacol 2002; 46: 287–297. 3. Libert JP, Di Nisi J, Fukuda H, Muzet A, Ehrhart J, Amoros C. Effect of continuous heat exposure on sleep stages in humans. Sleep 1988; 11: 195–209. 4. Sinha RK, Ray AK. Sleep-wake study in an animal model of acute and chronic heat stress. Physiol Behav 2006; 89: 364–372. 5. Karmanova IG. Evolution of sleep. Basel: Karger, 1982, pp. 1–146.

268

Journal of Clinical Monitoring and Computing

6. Sarbadhikari SN, Dey S, Ray AK. Chronic exercise alters EEG power spectra in an animal model of depression. Ind J Physiol Pharmacol 1996; 40: 47–57. 7. Agarwal R, Gotman J. Computer-assisted sleep staging. IEEE Trans BME 2001; 48: 1412–1423. 8. Sarbadhikari SN, Ray AK. Identifying EEG power spectra of depressed rats using a neural network. In: Reddy, DC, ed., Recent advances in biomedical engineering. Tata McGrowHill, New Delhi, 1994: 76–79. 9. Sarbadhikari SN. A Neural network confirms that physical exercise reverses EEG changes in depressed rats. Med Eng Phys 1995; 17: 579–582. 10. Sinha RK. Artificial neural network detects changes in electroencephalogram power spectrum of different sleep-wake states in an animal model of heat stress. Med Biol Eng Comput 2003; 41: 595–600. 11. Mamelak A, Quattrochi JJ, Hobson JA. A microcomputer based system for automated EEG collection and scoring of behavioural state in cats. Brain Res Bull 1988; 21: 843–849. 12. Mamelak A, Quattrochi JJ, Hobson JA. Automatic staging of sleep in cats using neural networks. Electroenceph Clin Neurophy 1991; 79: 52–61. 13. O’Boyle DJ, Choi EWK, Conroy G, Turega M. Learned classification of EEG power spectra using a neural network. J Physiol, Proc Physiol Society, Shaffield Meeting. 19–20 April 1991; 438: 345. 14. Shimada T, Shiina T. Detection of characteristic waves of sleep EEG by neural network analysis. IEEE Trans BME 2000; 47: 369–379. 15. Sinha RK. Electro-encephalogram disturbances in different sleep-wake states following exposure to high environmental heat. Med Biol Eng Comput 2004; 42: 282–287. 16. Sinha RK. An approach to estimate EEG power spectrum as an index of heat stress using backpropagation artificial neural network. Med Eng Phys 2007a; 29: 120–124. 17. Sinha RK, Ray AK. An assessment of changes in open-field and elevated plus-maze behavior following heat stress in rats. Iran Biomed J 2004; 8: 127–133. 18. Stearns SD, David RA. Signal processing algorithms, Printice Hall of India Inc: Englewood Cliffs, New Jersy, 1988. 19. Goel V, Brambrink AM, Baykal A, Koeler RC, Hanley DF, Thakor NV. Dominant frequency analysis of EEG reveals brain’s response during injury and recovery. IEEE Trans BME 1996; 43: 1083–1092. 20. Andersen ML, Tufik S. Altered sleep and behavioral patterns in arthritic rats. Sleep Res Online 2000; 3: 161–167. 21. Rumelhart DE, Hinton GE, Williams RJ. Learning representations by back-propagating errors. Nature 1986; 323: 533–536. 22. Hassoun HM. Fundamentals of artificial neural networks, Printice-Hall of India Private Limited: New Delhi, 1998, pp. 35–56. 23. Sinha RK. Study of changes in some pathophysiological stress markers in different age groups of an animal model of acute and chronic heat stress. Iran Biomed J 2007b; 11: 101–111. 24. Dey PK. Modification of dopamine receptor agonist mediated behavioral responses in rats following exposure to chronic heat stress. Biomedicine 1998; 18: 41–47. 25. Dey PK. Modification of functional responsiveness of central serotonin receptor subtype in rats following exposure to chronic heat stress. Biomedicine 1999; 19: 57–65. 26. Dey PK. Involvement of endogenous opiates in heat stress. Biomedicine 2000; 20: 143–148.

27. Sharma HS. Heat-related deaths are largely due to brain damage. Ind J Med Res 2005; 121: 621–623. 28. Sharma HS. Hyperthermia induced brain oedema: current status & future perspectives. Ind J Med Res 2006; 123(5): 629–652. 29. Dubois M, Sato S, Lees DE, Bull JM, Smith R, White BG, Moore H, Macnamara TE. Electroencephalographic changes during whole body hyperthermia in humans. Electroenceph Clin Neurophysiol 1980; 50: 486–495. 30. Sharma HS, Westman J, Nyberg F. Pathophysiology of brain edema and cell changes following hyperthermic brain injury. In: Sharma, HS, Westman, J, eds, Progress in brain research. 115 Amsterdam, Elsevier, 1998: 351–412. 31. Parmeggiani PL, Rabini C. Sleep and environmental temperature. Arch Ital Biol 1970; 108: 369–387. 32. McGinty D, Szymusiak R. Keeping cool: a hypothesis about the mechanisms and functions of slow-wave sleep. Trends Neurosci 1990; 13: 480–487. 33. Morairty SR, Szymusiak R, Thomson D, McGinty DJ. Selective increases in non-rapid eye movement sleep following whole body heating in rats. Brain Res 1993; 617: 10–16. 34. Morrissette RN, Heller HC. Effects of temperature on sleep in the developing rat. Am J Physiol 1998; 274: R1087–R1093. 35. Rivier C, Vale W. Diminished responsiveness of the hypothalamic-pituitary-adrenal axis of the rat during exposure to prolonged stress: a pituitary-mediated mechanism. Endocrinology 1986; 121: 1320–1328. 36. Cure M. Plasma corticosterone response in continuous versus discontinuous chronic heat exposure in rat. J Physiol Behav 1989; 45: 1117–1122. 37. Sinnamon HM, Karvosky ME, Ilch CP. Locomotion and head scanning initiated by hypothalamic stimulation are inversely related. Behav Brain Res 1999; 99: 219–229. 38. Alam MN, Gong H, Alam T, Jaganath R, McGinty D, Szymusiak R. Sleep-waking discharge patterns of neurons recorded in the rat perifornical lateral hypothalamic area. J Physiol 2002; 538: 619–631. 39. Stanus E, Lacroix B, Kerkhofs M, Menlewiez J. Automated sleep scoring: a comparitive reliability of two algorithms. Electroenceph Clin Neurophy 1987; 66: 448–456. 40. Gevins AS, Stone RK, Ragsdale SD. Differentiating the effects of three benzodizepines on non REM sleep EEG spectra. A neural network pattern classification analysis. Neuropsychobiology 1988; 19: 108–115. 41. Sinha RK, Agrawal NK, Ray AK. A power spectrum based backpropagation artificial neural network model for classification of sleep-wake stages in rats. Online J Heal All Sci 2003; 1: 1. 42. Sirne RO, Isaacson SI, D’Attellis CE. A data-reduction process for long-term EEGs. IEEE Eng Med Biol 1999; 18: 56–61. 43. Sinha RK, Aggarwal Y, Das BN. Backpropagation artificial neural network detects changes in electro-encephalogram power spectra of syncopic patients. J Med Sys 2007; 31: 63–68. 44. Villiers JD, Barnard E. Backpropagation neural nets with one and two hidden layers. IEEE Trans Neural Network 1992; 4: 136–141. 45. Chen JDZ, Lin Z, Wu Q, McCallum RW. Non-invasive identification of gastric contractions from surface electrogastrogram using backpropagation neural networks. Med Eng Phys 1995; 17: 219–225. 46. Zurada JM. Introduction to artificial neural network systems, West Publishing Company: St. Paul, MN, 1997, pp. 163–250.