Knowledge-based approach to sleep EEG analysis--a feasibility study

A perspective on automated rapid eye movement sleep assessment

Rapid eye movement sleep is associated with distinct changes in various biomedical signals that can be easily captured during sleep, lending themselves to automated sleep staging using machine learning systems. Here, we provide a perspective on the critical characteristics of biomedical signals associated with rapid eye movement sleep and how they can be exploited for automated sleep assessment. We summarise key historical developments in automated sleep staging systems, having now achieved classification accuracy on par with human expert scorers and their role in the clinical setting. We also discuss rapid eye movement sleep assessment with consumer sleep trackers and its potential for unprecedented sleep assessment on a global scale. We conclude by providing a future outlook of computerised rapid eye movement sleep assessment and the role AI systems may play.

AI Empowered Virtual Reality Integrated Systems for Sleep Stage Classification and Quality Enhancement

Insomnia is a common public health problem and an open biomedical research topic. Insomnia results in various health problems, including memory decline, decreases concentration and weakens problem-solving ability. The insufficient sleep also leads to skin ageing, heart disease, high blood pressure, arrhythmia and stroke. While it remains as a global health concern, sleep quality improvement using modern technologies, such as machine learning, classification technologies, virtual reality (VR), becomes an open and hot research problem. These modern technologies offer new curing solutions under certain conditions. In this paper, we present a sleeping-aid system with a single-channel electroencephalogram (EEG) sleep stage classification algorithm to improve the sleep quality. The sleeping-aid system promotes machine learning integrated VR and multimedia technology for sleep improvement. Ninety participants were invited to test on three different systems with 3D VR, 2D video, and music only. An adequate stimulus of audio-vision can be a complement of the drug treatment. The experimental results showed that the proposed method demonstrated superior performance over existing methods.

Detection of k-complexes in EEG signals using a multi-domain feature extraction coupled with a least square support vector machine classifier

Sleep scoring is one of the primary tasks for the classification of sleep stages using electroencephalogram (EEG) signals. It is one of the most important diagnostic methods in sleep research and must be carried out with a high degree of accuracy because any errors in the scoring in the patient's sleep EEG recordings can cause serious problems. The aim of this research is to develop a new automatic method for detecting the most important characteristics in sleep stage 2 such as k-complexes based on multi-domain features. In this study, each EEG signal is divided into a set of segments using a sliding window technique. Based on extensive experiments during the training phase, the size of the sliding window is set to 0.5 s (s). Then a set of statistical, fractal, frequency and non-linear features are extracted from each epoch based on the time domain, Katz's algorithm, power spectrum density (PSD) and tunable Q-factor wavelet transform (TQWT). As a result, a vector of twenty-two features is obtained to represent each EEG segment. In order to detect k-complexes, the extracted features were analysed for their ability to detect the k-complex waveforms. Based on the analysis of the features, twelve out of twenty-two features are selected and forwarded to a least square support vector machine (LS-SVM) classifier to identify k-complexes in EEG signals. A set of various classification techniques of K-means and extreme learning machine classifiers are used to compare the obtained results and to evaluate the performance of the proposed method.The experimental results showed that the proposed method, based on multi-domain features, achieved better recognition results than other methods and classifiers. An average accuracy, sensitivity and specificity of 97.7 %, 97 %, and 94.2 % were obtained, respectively, with the CZ-A1 channel according to the R&K standard. The experimental results with high classification performance demonstrated that the technique can help doctors optimize the diagnosis and treatment of sleep disorders.

A transition-constrained discrete hidden Markov model for automatic sleep staging

BACKGROUND

Approximately one-third of the human lifespan is spent sleeping. To diagnose sleep problems, all-night polysomnographic (PSG) recordings including electroencephalograms (EEGs), electrooculograms (EOGs) and electromyograms (EMGs), are usually acquired from the patient and scored by a well-trained expert according to Rechtschaffen & Kales (R&K) rules. Visual sleep scoring is a time-consuming and subjective process. Therefore, the development of an automatic sleep scoring method is desirable.

METHOD

The EEG, EOG and EMG signals from twenty subjects were measured. In addition to selecting sleep characteristics based on the 1968 R&K rules, features utilized in other research were collected. Thirteen features were utilized including temporal and spectrum analyses of the EEG, EOG and EMG signals, and a total of 158  hours of sleep data were recorded. Ten subjects were used to train the Discrete Hidden Markov Model (DHMM), and the remaining ten were tested by the trained DHMM for recognition. Furthermore, the 2-fold cross validation was performed during this experiment.

RESULTS

Overall agreement between the expert and the results presented is 85.29%. With the exception of S1, the sensitivities of each stage were more than 81%. The most accurate stage was SWS (94.9%), and the least-accurately classified stage was S1 (<34%). In the majority of cases, S1 was classified as Wake (21%), S2 (33%) or REM sleep (12%), consistent with previous studies. However, the total time of S1 in the 20 all-night sleep recordings was less than 4%.

CONCLUSION

The results of the experiments demonstrate that the proposed method significantly enhances the recognition rate when compared with prior studies.

A rule-based automatic sleep staging method

In this paper, a rule-based automatic sleep staging method was proposed. Twelve features including temporal and spectrum analyses of the EEG, EOG, and EMG signals were utilized. Normalization was applied to each feature to eliminating individual differences. A hierarchical decision tree with fourteen rules was constructed for sleep stage classification. Finally, a smoothing process considering the temporal contextual information was applied for the continuity. The overall agreement and kappa coefficient of the proposed method applied to the all night polysomnography (PSG) of seventeen healthy subjects compared with the manual scorings by R&K rules can reach 86.68% and 0.79, respectively. This method can integrate with portable PSG system for sleep evaluation at-home in the near future.

A rule-based automatic sleep staging method

In this paper, a rule-based automatic sleep staging method was proposed. Twelve features, including temporal and spectrum analyses of the EEG, EOG, and EMG signals, were utilized. Normalization was applied to each feature to reduce the effect of individual variability. A hierarchical decision tree, with fourteen rules, was constructed for sleep stage classification. Finally, a smoothing process considering the temporal contextual information was applied for the continuity. The average accuracy and kappa coefficient of the proposed method applied to the all night polysomnography (PSG) of twenty subjects compared with the manual scorings reached 86.5% and 0.78, respectively. This method can assist the clinical staff reduce the time required for sleep scoring in the future.

Artificial neural network and wavelet based automated detection of sleep spindles, REM sleep and wake states

Backpropagation artificial neural network (ANN) has been designed to classify sleep-wake stages. Four hours continuous three channel polygraphic signals such as EEG (electroencephalogram), EOG (electrooculogram) and EMG (electromyogram) from conscious subjects were digitally recorded and stored in computer. EOG and EMG signals were used for manual identification of sleep states before training and testing of ANN. The percentages power of the 2 s epochs of the digitized EEG signals from each of three sleep-wake patterns, sleep spindles (SS), rapid eye movement (REM) sleep and awake (AWA) sates, were calculated and analyzed to select the manually confirmed sleep-wake states for each epoch. Further, second order Daubechies mother wavelet has been used to get the wavelet coefficients for the selected EEG epochs. The wavelet coefficients for the EEG epochs (64 data) were selected as inputs for the training the network and to classify SS, REM sleep and AWA stages. The ANN architecture used (64-14-3) in present study shows overall very good agreement with manual sleep stage scoring with an average of 95.35% for all the 1,140 samples tested from SS, REM and AWA stages. This architecture of ANN was also found effectively differentiating the EEG power spectra from different sleep-wake states (96.84% in SS, 93.68% in REM sleep, 95.52% in AWA state). The high performance observed with the system based on wavelet coefficients along with the ANN, highlights the need of this computational tool into the field of sleep research.

Extracting Fuzzy Rules From Polysomnographic Recordings for Infant Sleep Classification

A neuro-fuzzy classifier (NFC) of sleep-wake states and stages has been developed for healthy infants of ages 6 mo and onward. The NFC takes five input patterns previously identified on 20-s epochs from polysomnographic recordings and assigns them to one out of five possible classes: Wakefulness, REM-Sleep, Non-REM Sleep Stage 1, Stage 2, and Stage 3-4. The definite criterion for a sleep state or stage to be established is duration of at least 1 min. The data set consisted of a total of 14 continuous recordings of naturally occurring naps (average duration: 143 +/- 39 min), corresponding to a total of 6021 epochs. They were divided in a training, a validation and a test set with 7, 2, and 5 recordings, respectively. During supervised training, the system determined the fuzzy concepts associated to the inputs and the rules required for performing the classification, extracting knowledge from the training set, and pruning nonrelevant rules. Results on an independent test set achieved 83.9 +/- 0.4% of expert agreement. The fuzzy rules obtained from the training examples without a priori information showed a high level of coincidence with the crisp rules stated by the experts, which are based on internationally accepted criteria. These results show that the NFC can be a valuable tool for implementing an automated sleep-wake classification system.

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.

An artificial intelligence approach to classify and analyse EEG traces

We present a fully automatic system for the classification and analysis of adult electroencephalograms (EEGs). The system is based on an artificial neural network which classifies the single epochs of trace, and on an Expert System (ES) which studies the time and space correlation among the outputs of the neural network; compiling a final report. On the last 2000 EEGs representing different kinds of alterations according to clinical occurrences, the system was able to produce 80% good or very good final comments and 18% sufficient comments, which represent the documents delivered to the patient. In the remaining 2% the automatic comment needed some modifications prior to be presented to the patient. No clinical false-negative classifications did arise, i.e. no altered traces were classified as 'normal' by the neural network. The analysis method we describe is based on the interpretation of objective measures performed on the trace. It can improve the quality and reliability of the EEG exam and appears useful for the EEG medical reports although it cannot totally substitute the medical doctor who should now read the automatic EEG analysis in light of the patient's history and age.

Neural network for sleep EEG K-complex detection

The paper presents the development and application of an automatic system used to detect and classify the K-complexes aperiodic, waveforms found in electroencephalograms of patients during stage two sleep. The slow-wave transient K-complex is evoked by auditory or somatosensory stimulation being an event related potential. The analysis of this transitory waveform contributes to the assessment of sleep stages used by controlled learning during sleep. In our work we used a TMS320C30 DSP to implement an automatic detection procedure based on features extraction and classification using a feed-forward neural network.

Polysomnographic pattern recognition for automated classification of sleep-waking states in infants

A robust, automated pattern recognition system for polysomnography data targeted to the sleep-waking state and stage identification is presented. Five patterns were searched for: slow-delta and theta wave predominance in the background electro-encephalogram (EEG) activity; presence of sleep spindles in the EEG; presence of rapid eye movements in an electro-oculogram; and presence of muscle tone in an electromyogram. The performance of the automated system was measured indirectly by evaluating sleep staging, based on the experts' accepted methodology, to relate the detected patterns in infants over four months of post-term age. The set of sleep-waking classes included wakefulness, REM sleep and non-REM sleep stages I, II, and III-IV. Several noise and artifact rejection methods were implemented, including filters, fuzzy quality indices, windows of variable sizes and detectors of limb movements and wakefulness. Eleven polysomnographic recordings of healthy infants were studied. The ages of the subjects ranged from 6 to 13 months old. Six recordings counting 2665 epochs were included in the training set. Results on a test set (2,369 epochs from five recordings) show an overall agreement of 87.7% (kappa 0.840) between the automated system and the human expert. These results show significant improvements compared with previous work.

Automated sleep stage scoring using hybrid rule- and case-based reasoning

We propose an automated method for sleep stage scoring using hybrid rule- and case-based reasoning. The system first performs rule-based sleep stage scoring, according to the Rechtschaffen and Kale's sleep-scoring rule (1968), and then supplements the scoring with case-based reasoning. This method comprises signal processing unit, rule-based scoring unit, and case-based scoring unit. We applied this methodology to three recordings of normal sleep and three recordings of obstructive sleep apnea (OSA). Average agreement rate in normal recordings was 87.5% and case-based scoring enhanced the agreement rate by 5.6%. This architecture showed several advantages over the other analytical approaches in sleep scoring: high performance on sleep disordered recordings, the explanation facility, and the learning ability. The results suggest that combination of rule-based reasoning and case-based reasoning is promising for an automated sleep scoring and it is also considered to be a good model of the cognitive scoring process.

Context related artefact detection in prolonged EEG recordings

The need for reliable detection of artefacts in raw and processed EEG is widely acknowledged. Although different EEG analysis systems have been described, only few general applicable artefact recognition techniques have emerged. This paper tackles the problem of artefact detection in seven 24 h EEG recordings in the intensive care unit. ICU recordings have received less attention than, e.g. epilepsy monitoring, although recordings in this environment present an interesting application area. The EEG data used here was recorded during the difficult circumstances of an explorative ICU study. The data set includes a diverse set of EEG patterns, as well as EEG artefacts. The study investigates objective artefact detection methods based on statistical differences between signal parameters, using time-varying autoregressive modelling (AR) and Slope detection. In addition to matching the performance of artefact detection against two human observers, the study focuses on the optimal settings for context incorporation by testing the algorithms for different time windows and epoch lengths. Results indicate that a relatively short period (20-40 s) provides sufficient context information for the methods used. The combined AR and Slope detection parameters yielded good performance, detecting approximately 90% of the artefacts as indicated by the consensus score of the human observers.

The future of computer-assisted investigation of the polysomnogram: sleep microstructure

Previous attempts at automated analysis of sleep were mainly directed towards imitating the Rechtschaffen and Kales rules (RKR) in order to save scoring time and further objectify the procedure. RKR, however, do not take into consideration the sleep microstructure of REM, stage 2, and SWS. While the microstructure of stage 2 has been analyzed in the past decade, the microstructure of REM and SWS are virtually unknown. In stage 2 the amount and distribution of spindles, K complexes, and arousal reactions have been studied. At least two types of spindles (12/s and 14/s) with different dynamics and locations have been identified. Two different shapes for K complexes have been described: one related to external sensory stimuli with similarities to evoked potentials and another one more related to sinusoidal slow wave activity seen in SWS. These two different K complex shapes have different distributions and, obviously, different functions. The authors also suggest that one should differentiate between arousal reactions and true arousals. Recent investigations suggest two types of delta waves in SWS. The more sinusoidal 1-3/s delta waves with a frontal maximum are already seen with lower amplitude in late stage 2 and increase their amplitude and incidence towards stage 3 and Stage 4. The other delta-wave type is slower (< 1/s), polymorphic, and has varying amounts of theta and higher frequency waves superimposed. During REM sleep it seems to be important to separate phases with rapid eye movements from those with none (REM sine REM), and count the amount and distribution of sawtooth activity. Background activity during REM and REM sine REM, as well as intra- and interhemispheric coherence should be analyzed separately. Only if the microstructure of the sleep EEG can be analyzed automatically using newer techniques such as transformation into wavelets and pattern classification with neuronal networks, and only if we learn more about the importance of microstructure elements, can automated sleep analysis go beyond the limited information obtained from scoring according to RKR.

Semi-automatic analysis of electroencephalogram in sleep apnea syndromes

STUDY OBJECTIVE

The present study was undertaken to validate a system of condensed display of the sleep EEG spectral analysis with a view to simplifying sleep studies in patients affected by sleep apnea syndromes (SAS).

DESIGN

All subjects underwent polysomnography on two consecutive nights using a Respisomnograph. The 8-h EEG, electro-oculogram, and electromyogram were digitalized. A fast Fourier transform was done and successive spectrums were computed resulting in one spectrum for each 20-s period. Then the EEG spectrum was plotted as a Lofard color diagram.

SETTING

To validate this system we compared a visual analysis of EEG recordings by a trained neurophysiologist with 10-s epochs and a visual analysis of the Lofard diagram by a practitioner with no previous understanding in EEG after only 2 h of instruction. The 8-h recording was divided into 10-min blocks. Each block was staged in function of the amplitude and prevailing frequency of EEG as indicated by the color (wakefulness, light slow wave sleep, deep slow wave sleep, rapid eye movement sleep). The stage of each block was compared with the predominant stage found in this block by the neurophysiologist.

PATIENTS

Eight normal adults and 24 patients suffering from SAS were recruited.

RESULTS

When all the blocks of all patients were pooled (1,438 blocks), agreement was total in 81 percent, partial in 11 percent, and discordant in 8 percent.

CONCLUSION

This system of spectral analysis representation allows a fast evaluation of quality of sleep and of EEG.

Feature-based detection of the K-complex wave in the human electroencephalogram using neural networks

The main difficulties in reliable automated detection of the K-complex wave in EEG are its close similarity to other waves and the lack of specific characterization criteria. We present a feature-based detection approach using neural networks that provides good agreement with visual K-complex recognition: a sensitivity of 90% is obtained with about 8% false positives. The respective contribution of the features and that of the neural network is demonstrated by comparing the results to those obtained with i) raw EEG data presented to neural networks, and ii) features presented to Fisher's linear discriminant.

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.

Quantitative analysis of electroencephalograms: is there chaos in the future?

The history of quantitative, computerized electroencephalogram (EEG) analysis is reviewed. It is shown that, until very recently, the basic approach to EEG analysis involved the assumption that the EEG is stochastic. Consequently, statistical pattern recognition techniques, segmentation procedures, syntactic methods, knowledge-based approaches, and even artificial neural network methods have been developed with different levels of success. A fundamentally different approach to computerized EEG analysis, however, is making its way into the laboratories. The basic idea, inspired by recent advances in the area of non-linear dynamics, and especially the theory of chaos, is to view an EEG as the output of a deterministic system of relatively simple complexity, but containing non-linearities. This suggests that studying the geometrical dynamics of EEGs, and the development of neurophysiologically realistic models of EEG generation may produce more successful automated EEG analysis techniques than the classical, stochastic methods. Evidence supporting the non-linear dynamics paradigm is reviewed, and possible research paths are indicated.