Automatic EEG analysis during long-term monitoring in the ICU

Comparison of different algorithms based on TKEO for EMG change point detection

OBJECTIVE

A significant challenge in surface electromyography (EMG) is the accurate identification of onset and offset of muscle activation while maintaining high real-time performance. Teager-Kaiser energy operator (TKEO) is widely used in muscle activity monitoring systems because of its computational simplicity and strong real-time performance. However, in contrast to TKEO ontology, few studies have examined how well the energy operator variants from multiple fields perform in conditioning EMG signals. This paper aims to investigate the role of the energy operator and its variants in EMG change point detection by a threshold detector.

APPROACH

To compare the stability and accuracy of TKEO and its variants for EMG change point detection, the EMG data of extensor carpi radialis longus and flexor carpi radialis were acquired from twenty participants operating a controller under normal and disturbed conditions, and EMG change point detection was performed by four energy operators and their rectified versions.

MAIN RESULTS

Based on the "standard" change points collected by the controller, the detection results were evaluated by three evaluation indexes: detection rate, F1 Score, and accuracy. The experimental results show that the multiresolution energy operator (MTEO) and the TKEO with rectified (abs-TKEO) are more suitable for EMG change point detection.

SIGNIFICANCE

This paper compared the effect of the energy operator and its variants on a threshold-based EMG change point detector. The experimental results in this paper can provide a reference for the selection of EMG signal conditioning methods to improve the detection performance of the EMG change point detector.

Detection of EEG burst-suppression in neurocritical care patients using an unsupervised machine learning algorithm

OBJECTIVE

The burst suppression pattern in clinical electroencephalographic (EEG) recordings is an important diagnostic tool because of its association with comas of various etiologies, as with hypoxia, drug related intoxication or deep anesthesia. The detection of bursts and the calculation of burst/suppression ratio are often used to monitor the level of anesthesia during treatment of status epilepticus. However, manual counting of bursts is a laborious process open to inter-rater variation and motivates a need for automatic detection.

METHODS

We describe a novel unsupervised learning algorithm that detects bursts in EEG and generates burst-per-minute estimates for the purpose of monitoring sedation level in an intensive care unit (ICU). We validated the algorithm on 29 hours of burst annotated EEG data from 29 patients suffering from status epilepticus and hemorrhage.

RESULTS

We report competitive results in comparison to neural networks learned via supervised learning. The mean absolute error (SD) in bursts per minute was 0.93 (1.38).

CONCLUSION

We present a novel burst suppression detection algorithm that adapts to each patient individually, reports bursts-per-minute quickly, and does not require manual fine-tuning unlike previous approaches to burst-suppression pattern detection.

SIGNIFICANCE

Our algorithm for automatic burst suppression quantification can greatly reduce manual oversight in depth of sedation monitoring.

Machine learning model on heart rate variability metrics identifies asymptomatic toddlers exposed to zika virus during pregnancy

Objective. Although the Zika virus (ZIKV) seems to be prominently neurotropic, there are some reports of involvement of other organs, particularly the heart. Of special concern are those children exposed prenatally to ZIKV and born without microcephaly or other congenital anomalies. Electrocardiogram (ECG)-derived heart rate variability (HRV) metrics represent an attractive, low-cost, widely deployable tool for early identification of developmental functional alterations in exposed children born without such overt clinical symptoms. We hypothesized that HRV in such children would yield a biomarker of fetal ZIKV exposure. Our objective was to test this hypothesis in young children exposed to ZIKV during pregnancy.Approach. We investigated the HRV properties of 21 children aged 4-25 months from Brazil. The infants were divided into two groups, the ZIKV-exposed (n = 13) and controls (n = 8). Single-channel ECG was recorded in each child at ∼15 months of age and HRV was analyzed in 5 min segments to provide a comprehensive characterization of the degree of variability and complexity of the heart rate.Main results.Using a cubic support vector machine classifier we identified babies as Zika cases or controls with a negative predictive value of 92% and a positive predictive value of 86%. Our results show that a machine learning model derived from HRV metrics can help differentiate between ZIKV-affected, yet asymptomatic, and non-ZIKV-exposed babies. We identified the box count as the best HRV metric in this study allowing such differentiation, regardless of the presence of microcephaly.Significance.We show that it is feasible to measure HRV in infants and toddlers using a small non-invasive portable ECG device and that such an approach may uncover the memory ofin uteroexposure to ZIKV. We discuss putative mechanisms. This approach may be useful for future studies and low-cost screening tools involving this challenging to examine population.

Seizure Onset Detection Using Empirical Mode Decomposition and Common Spatial Pattern

Automatic seizure onset detection plays an important role in epilepsy diagnosis. In this paper, a novel seizure onset detection method is proposed by combining empirical mode decomposition (EMD) of long-term scalp electroencephalogram (EEG) with common spatial pattern (CSP). First, wavelet transform (WT) and EMD are employed on EEG recordings respectively for filtering pre-processing and time-frequency decomposition. Then CSP is applied to reduce the dimension of multi-channel time-frequency representation, and the variance is extracted as the only feature. Afterwards, a support vector machine (SVM) group consisting of ten SVMs is served as a robust classifier. Finally, the post-processing is adopted to acquire a higher recognition rate and reduce the false detection rate. The results obtained from CHB-MIT database of 977 h scalp EEG recordings reveal that the proposed system can achieve a segment-based sensitivity of 97.34% with a specificity of 97.50% and an event-based sensitivity of 98.47% with a false detection rate of 0.63/h. This proposed detection system was also validated on a clinical scalp EEG database from the Second Hospital of Shandong University, and the system yielded a sensitivity of 93.67% and a specificity of 96.06%. At the event-based level, a sensitivity of 99.39% and a false detection rate of 0.64/h were obtained. Furthermore, this work showed that the CSP spatial filter was helpful to identify EEG channels involved in seizure onsets. These satisfactory results indicate that the proposed system may provide a reference for seizure onset detection in clinical applications.

Motor Imagery Hand Movement Direction Decoding Using Brain Computer Interface to Aid Stroke Recovery and Rehabilitation

Motor Imagery (MI)-based Brain Computer Interface (BCI) system is a potential technology for active neurorehabilitation of stroke patients by complementing the conventional passive rehabilitation methods. Research to date mainly focused on classifying left vs. right hand/foot MI of stroke patients. Though a very few studies have reported decoding imagined hand movement directions using electroencephalogram (EEG)-based BCI, the experiments were conducted on healthy subjects. Our work analyzes MI-based brain cortical activity from EEG signals and decodes the imagined hand movement directions in stroke patients. The decoded direction (left vs. right) of hand movement imagination is used to provide control commands to a motorized arm support on which patient's affected (paralyzed) arm is placed. This enables the patient to move his/her stroke-affected hand towards the intended (imagined) direction that aids neuroplasticity in the brain. The synchronization measure called Phase Locking Value (PLV), extracted from EEG, is the neuronal signature used to decode the directional movement of the MI task. Event-related desynchronization/synchronization (ERD/ERS) analysis on Mu and Beta frequency bands of EEG is done to select the time bin corresponding to the MI task. The dissimilarities between the two directions of MI tasks are identified by selecting the most significant channel pairs that provided maximum difference in PLV features. The training protocol has an initial calibration session followed by a feedback session with 50 trials of MI task in each session. The feedback session extracts PLV features corresponding to most significant channel pairs which are identified in the calibration session and is used to predict the direction of MI task in left/right direction. An average MI direction classification accuracy of 74.44% is obtained in performing the training protocol and 68.63% from the prediction protocol during feedback session on 16 stroke patients.

First evidence that intrinsic fetal heart rate variability exists and is affected by hypoxic pregnancy

KEY POINTS

We introduce a technique to test whether intrinsic fetal heart rate variability (iFHRV) exists and we show the utility of the technique by testing the hypothesis that iFHRV is affected by chronic fetal hypoxia, one of the most common adverse outcomes of human pregnancy complicated by fetal growth restriction. Using an established late gestation ovine model of fetal development under chronic hypoxic conditions, we identify iFHRV in isolated fetal hearts and show that it is markedly affected by hypoxic pregnancy. Therefore, the isolated fetal heart has intrinsic variability and carries a memory of adverse intrauterine conditions experienced during the last third of pregnancy.

ABSTRACT

Fetal heart rate variability (FHRV) emerges from influences of the autonomic nervous system, fetal body and breathing movements, and from baroreflex and circadian processes. We tested whether intrinsic heart rate variability (iHRV), devoid of any external influences, exists in the fetal period and whether it is affected by chronic fetal hypoxia. Chronically catheterized ewes carrying male singleton fetuses were exposed to normoxia (n = 6) or hypoxia (10% inspired O₂ , n = 9) for the last third of gestation (105-138 days of gestation (dG); term ∼145 dG) in isobaric chambers. At 138 dG, isolated hearts were studied using a Langendorff preparation. We calculated basal intrinsic FHRV (iFHRV) indices reflecting iFHRV's variability, predictability, temporal symmetry, fractality and chaotic behaviour, from the systolic peaks within 15 min segments in each heart. Significance was assumed at P < 0.05. Hearts of fetuses isolated from hypoxic pregnancy showed approximately 4-fold increases in the Grid transformation as well as the AND similarity index (sgridAND) and a 4-fold reduction in the scale-dependent Lyapunov exponent slope. We also detected a 2-fold reduction in the Recurrence quantification analysis, percentage of laminarity (pL) and recurrences, maximum and average diagonal line (dlmax, dlmean) and the Multiscale time irreversibility asymmetry index. The iHRV measures dlmax, dlmean, pL and sgridAND correlated with left ventricular end-diastolic pressure across both groups (average R²  = 0.38 ± 0.03). This is the first evidence that iHRV originates in fetal life and that chronic fetal hypoxia significantly alters it. Isolated fetal hearts from hypoxic pregnancy exhibit a time scale-dependent higher complexity in iFHRV.

Rapid Annotation of Seizures and Interictal-ictal Continuum EEG Patterns

Seizures, status epilepticus, and seizure-like rhythmic or periodic activities are common, pathological, harmful states of brain electrical activity seen in the electroencephalogram (EEG) of patients during critical medical illnesses or acute brain injury. Accumulating evidence shows that these states, when prolonged, cause neurological injury. In this study we developed a valid method to automatically discover a small number of homogeneous pattern clusters, to facilitate efficient interactive labelling by EEG experts. 592 time domain and spectral features were extracted from continuous EEG (cEEG) data of 369 ICU (intensive care unit) patients. For each patient, feature dimensionality was reduced using principal component analysis (PCA), retaining 95% of the variance. K-medoids clustering was applied to learn a local dictionary from each patient, consisting of k=100 exemplars/words. Changepoint detection (CPD) was utilized to break each EEG into segments. A bag-of-words (BoW) representation was computed for each segment, specifically, a normalized histogram of the words found within each segment. Segments were further clustered using the BoW histograms by Affinity Propagation (AP) using a χ² distance to measure similarities between histograms. The resulting 30 50 clusters for each patient were scored by EEG experts through labeling only the cluster medoids. Embedding methods t-SNE (t-distributed stochastic neighbor embedding) and PCA were used to provide a 2D representation for visualization and exploration of the data. Our results illustrate that it takes approximately 3 minutes to annotate 24 hours of cEEG by experts, which is at least 60 times faster than unaided manual review.

Potential EEG biomarkers of sedation doses in intensive care patients unveiled by using a machine learning approach

OBJECTIVE

Sedation of neurocritically ill patients is one of the most challenging situation in ICUs. Quantitative knowledge on the sedation effect on brain activity in that complex scenario could help to uncover new markers for sedation assessment. Hence, we aim to evaluate the existence of changes of diverse EEG-derived measures in deeply-sedated (RASS-Richmond agitation-sedation scale  -4 and  -5) neurocritically ill patients, and also whether sedation doses are related with those eventual changes.

APPROACH

We performed an observational prospective cohort study in the intensive care unit of the Hospital de la Princesa. Twenty-six adult patients suffered from traumatic brain injury and subarachnoid hemorrhage were included in the present study. Long-term continuous electroencephalographic (EEG) recordings (2141 h) and hourly annotated information were used to determine the relationship between intravenous sedation infusion doses and network and spectral EEG measures. To do that, two different strategies were followed: assessment of the statistical dependence between both variables using the Spearman correlation rank and by performing an automatic classification method based on a machine learning algorithm.

MAIN RESULTS

More than 60% of patients presented a correlation greater than 0.5 in at least one of the calculated EEG measures with the sedation dose. The automatic classification method presented an accuracy of 84.3% in discriminating between different sedation doses. In both cases the nodes' degree was the most relevant measurement.

SIGNIFICANCE

The results presented here provide evidences of brain activity changes during deep sedation linked to sedation doses. Particularly, the capability of network EEG-derived measures in discriminating between different sedation doses could be the framework for the development of accurate methods for sedation levels assessment.

Extraction of features from sleep EEG for Bayesian assessment of brain development

Brain development can be evaluated by experts analysing age-related patterns in sleep electroencephalograms (EEG). Natural variations in the patterns, noise, and artefacts affect the evaluation accuracy as well as experts’ agreement. The knowledge of predictive posterior distribution allows experts to estimate confidence intervals within which decisions are distributed. Bayesian approach to probabilistic inference has provided accurate estimates of intervals of interest. In this paper we propose a new feature extraction technique for Bayesian assessment and estimation of predictive distribution in a case of newborn brain development assessment. The new EEG features are verified within the Bayesian framework on a large EEG data set including 1,100 recordings made from newborns in 10 age groups. The proposed features are highly correlated with brain maturation and their use increases the assessment accuracy.

Global EEG segmentation using singular value decomposition

In this paper, we propose a method based on singular value decomposition (SVD) for segmenting multichannel electroencephalography (EEG) data into temporal blocks during which the spatial distributions of the underlying active neuronal generators stay fixed. We locate segment boundaries by statistically comparing the residual error resulting from projecting the data under a reference window, on one hand, and a sliding window, on the other hand, onto a feature subspace. The basis of this subspace is the most significant left eigenvectors of the data block under the reference window. The statistical testing is performed using the Kolmogorov-Smirnov (K-S) test. To enhance the reliability of the K-S test, the consecutive K-S decisions are aggregated under a given decision window. Simulation results confirm that the proposed algorithm can successfully detect segment boundaries under a wide range of different conditions.

Automated analysis of multi-channel EEG in preterm infants

OBJECTIVE

To develop and validate two automatic methods for the detection of burst and interburst periods in preterm eight-channel electroencephalographs (EEG). To perform a detailed analysis of interobserver agreement on burst and interburst periods and use this as a benchmark for the performance of the automatic methods. To examine mathematical features of the EEG signal and their potential correlation with gestational age.

METHODS

Multi-channel EEG from 36 infants, born at less than 30 weeks gestation was utilised, with a 10 min artifact-free epoch selected for each subject. Three independent expert observers annotated all EEG activity bursts in the dataset. Two automatic algorithms for burst/interburst detection were applied to the EEG data and their performances were analysed and compared with interobserver agreement. A total of 12 mathematical features of the EEG signal were calculated and correlated with gestational age.

RESULTS

The mean interobserver agreement was found to be 77% while mean algorithm/observer agreement was 81%. Six of the mathematical features calculated (spectral entropy, Higuchi fractal dimension, spectral edge frequency, variance, extrema median and Hilberts transform amplitude) were found to have significant correlation with gestational age.

CONCLUSIONS

Automatic detection of burst/interburst periods has been performed in multi-channel EEG of 36 preterm infants. The algorithm agreement with expert observers is found to be on a par with interobserver agreement. Mathematical features of EEG have been calculated which show significant correlation with gestational age.

SIGNIFICANCE

Automatic analysis of preterm multi-channel EEG is possible. The methods described here have the potential to be incorporated into a fully automatic system to quantitatively assess brain maturity from preterm EEG.

Evaluation of the different sleep-disordered breathing patterns of the compressed tracheal sound

OBJECTIVE

Suitability of the compressed tracheal sound signal for screening different sleep-disordered breathing patterns was evaluated. The previous results suggest that the plain pattern in the compressed sound signal represents mostly normal, unobstructed breathing, the thick pattern consists of periodic apneas/hypopneas and during the thin pattern, flow limitation in the nasal cannula signal is abundant.

METHODS

Twenty-seven patients underwent a polysomnography with a tracheal sound and oesophageal pressure monitoring. The tracheal sound data was compressed and scored visually into three different breathing patterns. The percentage of oesophageal pressure values under -8cm H2O, the minimum pressure value and the average duration of the breathing cycles were extracted from 10-min episodes of those plain, thick and thin patterns. In addition, the spectral contents of the tracheal sound during the different breathing patterns were evaluated.

RESULTS

The percentage of time when the oesophageal pressure negativity increased was highest during the thin pattern and lowest during the plain pattern. In addition, the thin pattern presented most high frequency components in the 1001-2000Hz frequency band of the tracheal sound.

CONCLUSIONS

The results confirmed our previous findings that both the thick and thin patterns seem to consist of obstructed breathing, whereas during the plain pattern the breathing is normal, unobstructed.

SIGNIFICANCE

Most screening methods for sleep-disordered breathing reveal only periodic apneas/hypopneas, but with the compressed sound signal the sustained partial obstruction can be estimated as well.

Classification of normal and epileptic seizure EEG signals using wavelet transform, phase-space reconstruction, and Euclidean distance

This paper proposes new combined methods to classify normal and epileptic seizure EEG signals using wavelet transform (WT), phase-space reconstruction (PSR), and Euclidean distance (ED) based on a neural network with weighted fuzzy membership functions (NEWFM). WT, PSR, ED, and statistical methods that include frequency distributions and variation, were implemented to extract 24 initial features to use as inputs. Of the 24 initial features, 4 minimum features with the highest accuracy were selected using a non-overlap area distribution measurement method supported by the NEWFM. These 4 minimum features were used as inputs for the NEWFM and this resulted in performance sensitivity, specificity, and accuracy of 96.33%, 100%, and 98.17%, respectively. In addition, the area under Receiver Operating Characteristic (ROC) curve was used to measure the performances of NEWFM both without and with feature selections.

Automatic identification and classification of muscle spasms in long-term EMG recordings

Spinal cord injured (SCI) individuals may be afflicted by spasticity, a condition in which involuntary muscle spasms are common. EMG recordings can be analyzed to quantify this symptom of spasticity but manual identification and classification of spasms are time consuming. Here, an algorithm was created to find and classify spasm events automatically within 24-h recordings of EMG. The algorithm used expert rules and time-frequency techniques to classify spasm events as tonic, unit, or clonus spasms. A companion graphical user interface (GUI) program was also built to verify and correct the results of the automatic algorithm or manually defined events. Eight channel EMG recordings were made from seven different SCI subjects. The algorithm was able to correctly identify an average (±SD) of 94.5 ± 3.6% spasm events and correctly classify 91.6 ± 1.9% of spasm events, with an accuracy of 61.7 ± 16.2%. The accuracy improved to 85.5 ± 5.9% and the false positive rate decreased to 7.1 ± 7.3%, respectively, if noise events between spasms were removed. On average, the algorithm was more than 11 times faster than manual analysis. Use of both the algorithm and the GUI program provide a powerful tool for characterizing muscle spasms in 24-h EMG recordings, information which is important for clinical management of spasticity.

Methods of EEG signal features extraction using linear analysis in frequency and time-frequency domains

Technically, a feature represents a distinguishing property, a recognizable measurement, and a functional component obtained from a section of a pattern. Extracted features are meant to minimize the loss of important information embedded in the signal. In addition, they also simplify the amount of resources needed to describe a huge set of data accurately. This is necessary to minimize the complexity of implementation, to reduce the cost of information processing, and to cancel the potential need to compress the information. More recently, a variety of methods have been widely used to extract the features from EEG signals, among these methods are time frequency distributions (TFD), fast fourier transform (FFT), eigenvector methods (EM), wavelet transform (WT), and auto regressive method (ARM), and so on. In general, the analysis of EEG signal has been the subject of several studies, because of its ability to yield an objective mode of recording brain stimulation which is widely used in brain-computer interface researches with application in medical diagnosis and rehabilitation engineering. The purposes of this paper, therefore, shall be discussing some conventional methods of EEG feature extraction methods, comparing their performances for specific task, and finally, recommending the most suitable method for feature extraction based on performance.

A Cerebral Recovery Index (CRI) for early prognosis in patients after cardiac arrest

Introduction

Electroencephalogram (EEG) monitoring in patients treated with therapeutic hypothermia after cardiac arrest may assist in early outcome prediction. Quantitative EEG (qEEG) analysis can reduce the time needed to review long-term EEG and makes the analysis more objective. In this study, we evaluated the predictive value of qEEG analysis for neurologic outcome in postanoxic patients.

Methods

In total, 109 patients admitted to the ICU for therapeutic hypothermia after cardiac arrest were included, divided over a training and a test set. Continuous EEG was recorded during the first 5 days or until ICU discharge. Neurologic outcomes were based on the best achieved Cerebral Performance Category (CPC) score within 6 months. Of the training set, 27 of 56 patients (48%) and 26 of 53 patients (49%) of the test set achieved good outcome (CPC 1 to 2). In all patients, a 5 minute epoch was selected each hour, and five qEEG features were extracted. We introduced the Cerebral Recovery Index (CRI), which combines these features into a single number.

Results

At 24 hours after cardiac arrest, a CRI <0.29 was always associated with poor neurologic outcome, with a sensitivity of 0.55 (95% confidence interval (CI): 0.32 to 0.76) at a specificity of 1.00 (CI, 0.86 to 1.00) in the test set. This results in a positive predictive value (PPV) of 1.00 (CI, 0.73 to 1.00) and a negative predictive value (NPV) of 0.71 (CI, 0.53 to 0.85). At the same time, a CRI >0.69 predicted good outcome, with a sensitivity of 0.25 (CI, 0.10 to 0.14) at a specificity of 1.00 (CI, 0.85 to 1.00) in the test set, and a corresponding NPV of 1.00 (CI, 0.54 to 1.00) and a PPV of 0.55 (CI, 0.38 to 0.70).

Conclusions

We introduced a combination of qEEG measures expressed in a single number, the CRI, which can assist in prediction of both poor and good outcomes in postanoxic patients, within 24 hours after cardiac arrest.

A high-performance lossless compression scheme for EEG signals using wavelet transform and neural network predictors

Developments of new classes of efficient compression algorithms, software systems, and hardware for data intensive applications in today's digital health care systems provide timely and meaningful solutions in response to exponentially growing patient information data complexity and associated analysis requirements. Of the different 1D medical signals, electroencephalography (EEG) data is of great importance to the neurologist for detecting brain-related disorders. The volume of digitized EEG data generated and preserved for future reference exceeds the capacity of recent developments in digital storage and communication media and hence there is a need for an efficient compression system. This paper presents a new and efficient high performance lossless EEG compression using wavelet transform and neural network predictors. The coefficients generated from the EEG signal by integer wavelet transform are used to train the neural network predictors. The error residues are further encoded using a combinational entropy encoder, Lempel-Ziv-arithmetic encoder. Also a new context-based error modeling is also investigated to improve the compression efficiency. A compression ratio of 2.99 (with compression efficiency of 67%) is achieved with the proposed scheme with less encoding time thereby providing diagnostic reliability for lossless transmission as well as recovery of EEG signals for telemedicine applications.

COMPUTER DETECTION APPROACHES FOR IDENTIFICATION OF PHASIC ELECTROMYOGRAPHIC (EMG) ACTIVITY DURING HUMAN SLEEP

BACKGROUND: Examination of spontaneously occurring phasic muscle activity from the human polysomnogram may have considerable clinical importance for patient care, yet most attempts to quantify the detection of such activity have relied upon laborious and intensive visual analyses. We describe in this study innovative signal processing approaches to this issue. METHODS: We examined multiple features of surface electromyographic signals based on 16,200 individual 1-second intervals of low impedance sleep recordings. We validated which of those features most closely mirrored the careful judgments of trained human observers in making discriminations of the presence of short-lived (100-500 msec) phasic activity, and also examined which features provided maximal differences across 1-second intervals and which features were least susceptible to residual levels of amplifier noise. RESULTS: Our data suggested particularly promising and novel features (e.g., Non-linear energy, 95(th) percentile of Spectral Edge Frequency) for developing automated systems for quantifying muscle activity during human sleep. CONCLUSIONS: The EMG signals recorded from surface electrodes during sleep can be processed with techniques that reflect the visually based analyses of the human scorer but also offer potential for discerning far more subtle effects, Future studies will explore both the clinical utility of these techniques and their relative susceptibility to and/or independence from signal artifacts.

Model-based seizure detection for intracranial EEG recordings

This paper presents a novel model-based patient-specific method for automatic detection of seizures in the intracranial EEG recordings. The proposed method overcomes the complexities in the practical implementation of the patient-specific approach of seizure detection. The method builds a seizure model (set of basis functions) for a priori known seizure (the template seizure pattern), and uses the statistically optimal null filters as a building block for the detection of similar seizures. The process of modeling the template seizure is fully automatic. Overall, the detection method involves the segmentation of the template seizure pattern, rejection of the redundant and noisy segments, extraction of features from the segments to generate a set of models, selection of the best seizure model, and training of the classifier. The trained classifier is used to detect similar seizures in the remaining data. The resulting seizure detection method was evaluated on a total of 304 h of single-channel depth EEG recordings from 14 patients. The system performance is further compared to the Qu-Gotman patient-specific system using the same data. A significant improvement in the proposed system, in terms of specificity, is observed over the compared method.

Rapid identification of epileptogenic sites in the intracranial EEG

The paper presents a novel computationally simple, easy-to-interpret compressed EEG display for multichannel intracranial EEG recordings. The compressed display is based on the level of sharp activity (relative sharpness index (RSI)) in the EEG, which profoundly increases during paroxysmal activities. RSI is graphically presented as a color-intensity plot that allows compressing several hours of EEG into a single display page. RSI display is a bird's-eye-view of the EEG that may reveal seizure evolution ('build-up'), seizure precursors, or sites associated with the seizures. We present examples from two patients to illustrate the method's ability to identify epileptogenic sites that may be difficult to observe in the conventional review process. RSI is compared with the color density spectral array (CDSA) and amplitude integrated EEG (aEEG) display. Examples demonstrate the RSI display to be simple, easy to interpret, computationally light and fast enough for online application.

Review and classification of variability analysis techniques with clinical applications

Analysis of patterns of variation of time-series, termed variability analysis, represents a rapidly evolving discipline with increasing applications in different fields of science. In medicine and in particular critical care, efforts have focussed on evaluating the clinical utility of variability. However, the growth and complexity of techniques applicable to this field have made interpretation and understanding of variability more challenging. Our objective is to provide an updated review of variability analysis techniques suitable for clinical applications. We review more than 70 variability techniques, providing for each technique a brief description of the underlying theory and assumptions, together with a summary of clinical applications. We propose a revised classification for the domains of variability techniques, which include statistical, geometric, energetic, informational, and invariant. We discuss the process of calculation, often necessitating a mathematical transform of the time-series. Our aims are to summarize a broad literature, promote a shared vocabulary that would improve the exchange of ideas, and the analyses of the results between different studies. We conclude with challenges for the evolving science of variability analysis.

Modeling electroencephalography waveforms with semi-supervised deep belief nets: fast classification and anomaly measurement

Clinical electroencephalography (EEG) records vast amounts of human complex data yet is still reviewed primarily by human readers. Deep belief nets (DBNs) are a relatively new type of multi-layer neural network commonly tested on two-dimensional image data but are rarely applied to times-series data such as EEG. We apply DBNs in a semi-supervised paradigm to model EEG waveforms for classification and anomaly detection. DBN performance was comparable to standard classifiers on our EEG dataset, and classification time was found to be 1.7-103.7 times faster than the other high-performing classifiers. We demonstrate how the unsupervised step of DBN learning produces an autoencoder that can naturally be used in anomaly measurement. We compare the use of raw, unprocessed data--a rarity in automated physiological waveform analysis--with hand-chosen features and find that raw data produce comparable classification and better anomaly measurement performance. These results indicate that DBNs and raw data inputs may be more effective for online automated EEG waveform recognition than other common techniques.

Creating symbolic representations of electroencephalographic signals: an investigation of alternate methodologies on intracranial data

The electroencephalogram (EEG) is widely used in the investigation of neurological disorders. Continuous long-term EEG data offers the opportunity to assess patient health over long periods of time, and to discover previously unknown physiological phenomena. However, the sheer volume of information generated by long-term EEG monitoring also poses a serious challenge for both analysis and visualization. Symbolization has been successful in addressing information overload in many disciplines. In this paper, we present different approaches to transform EEG signals into symbolic sequences. This discrete symbolic representation reduces the amount of EEG data by several orders of magnitude and makes the task of discovering and visualizing interesting activity more manageable. We describe alternate methodologies to symbolize EEG data from patients with epilepsy. When evaluated on long-term intracranial data from 10 patients, our symbolization produced results that were consistent with clinical labels of seizures (for 97% of the seizures and 68% of the seizure segments), and often produced finer-grained distinctions.

Optimization of an NLEO-based algorithm for automated detection of spontaneous activity transients in early preterm EEG

We propose here a simple algorithm for automated detection of spontaneous activity transients (SATs) in early preterm electroencephalography (EEG). The parameters of the algorithm were optimized by supervised learning using a gold standard created from visual classification data obtained from three human raters. The generalization performance of the algorithm was estimated by leave-one-out cross-validation. The mean sensitivity of the optimized algorithm was 97% (range 91-100%) and specificity 95% (76-100%). The optimized algorithm makes it possible to systematically study brain state fluctuations of preterm infants.

Continuous electroencephalogram monitoring in the intensive care unit

Because of recent technical advances, it is now possible to record and monitor the continuous digital electroencephalogram (EEG) of many critically ill patients simultaneously. Continuous EEG monitoring (cEEG) provides dynamic information about brain function that permits early detection of changes in neurologic status, which is especially useful when the clinical examination is limited. Nonconvulsive seizures are common in comatose critically ill patients and can have multiple negative effects on the injured brain. The majority of seizures in these patients cannot be detected without cEEG. cEEG monitoring is most commonly used to detect and guide treatment of nonconvulsive seizures, including after convulsive status epilepticus. In addition, cEEG is used to guide management of pharmacological coma for treatment of increased intracranial pressure. An emerging application for cEEG is to detect new or worsening brain ischemia in patients at high risk, especially those with subarachnoid hemorrhage. Improving quantitative EEG software is helping to make it feasible for cEEG (using full scalp coverage) to provide continuous information about changes in brain function in real time at the bedside and to alert clinicians to any acute brain event, including seizures, ischemia, increasing intracranial pressure, hemorrhage, and even systemic abnormalities affecting the brain, such as hypoxia, hypotension, acidosis, and others. Monitoring using only a few electrodes or using full scalp coverage, but without expert review of the raw EEG, must be done with extreme caution as false positives and false negatives are common. Intracranial EEG recording is being performed in a few centers to better detect seizures, ischemia, and peri-injury depolarizations, all of which may contribute to secondary injury. When cEEG is combined with individualized, physiologically driven decision making via multimodality brain monitoring, intensivists can identify when the brain is at risk for injury or when neuronal injury is already occurring and intervene before there is permanent damage. The exact role and cost-effectiveness of cEEG at the current time remains unclear, but we believe it has significant potential to improve neurologic outcomes in a variety of settings.

EEG spectral changes and onset of burst suppression pattern in propofol/remifentanil anesthesia

This paper studies how remifentanil, a commonly used intraoperative opioid, affects the relation of the onset of burst suppression pattern (BSP) and the spectral changes of EEG during anesthesia. The onsets of BSP were detected using both manual and the automatic method proposed from the EEGs of twenty-seven patients who had received different amount of remifentanil with the anesthetic. The spectral changes were determined by calculating the frequency progression patterns of the EEGs. The results showed that remifentanil significantly affects the relation of EEG spectral changes and the onset of BSP. The finding is important since the current EEG-based assessment of the depth of anesthesia basically relies on the analysis of the spectral features and BSP.

An adaptive error modeling scheme for the lossless compression of EEG signals

Lossless compression of EEG signal is of great importance for the neurological diagnosis as the specialists consider the exact reconstruction of the signal as a primary requirement. This paper discusses a lossless compression scheme for EEG signals that involves a predictor and an adaptive error modeling technique. The prediction residues are arranged based on the error count through an histogram computation. Two optimal regions are identified in the histogram plot through a heuristic search such that the bit requirement for encoding the two regions is minimum. Further improvement in the compression is achieved by removing the statistical redundancy that is present in the residue signal by using a context-based bias cancellation scheme. Three neural network predictors, namely, single-layer perceptron, multilayer perceptron, and Elman network and two linear predictors, namely, autoregressive model and finite impulse response filter are considered. Experiments are conducted using EEG signals recorded under different physiological conditions and the performances of the proposed methods are evaluated in terms of the compression ratio. It is shown that the proposed adaptive error modeling schemes yield better compression results compared to other known compression methods.

Continuous EEG monitoring in the ICU

Continuous EEG (cEEG) monitoring is one of many available techniques to assess cerebral function in critically ill patients. Detection and treatment of nonconvulsive seizures (NCSZ) and nonconvulsive status epilepticus (NCSE) are the main clinical applications of cEEG. These patterns are common and associated with poor outcome after severe brain injury. Quantitative EEG parameters can be used for early detection of NCSZ and ischemia caused by vasospasm after subarachnoid hemorrhage. Early and aggressive treatment of such complications may prevent secondary brain injury and avoid irreversible damage. Periodic epileptiform discharges (PEDs) are also seen frequently after acute brain injury and may be associated with poor outcome. However, to date, it is uncertain whether NCSZ, NCSE or PEDs cause additional injury or if they are epiphenomena of brain damage. Currently, there are many limitations to the widespread use of cEEG, particularly the lack of high quality studies. In the future, the role of cEEG as part of multimodality neuromonitoring should be further investigated to determine if optimization of neuronal activity, brain metabolism, oxygenation and perfusion profiles can prevent further damage to the brain and thereby improve outcome.

Automatic detection of burst suppression

Burst suppression pattern (BSP) as a common diffuse abnormal electroencephalographic (EEG) pattern requires close monitoring in the intensive care unit (ICU) environments. Automatic detection of individual BS events has a clinical and practical importance for brain function monitoring in the neurological ICUs (NICUs) using Continuous EEG (CEEG). In this paper, we present a novel method to automatically detect burst suppression events. The method is based on segmentation and detection of the suppression component of the BS event using integrated EEG signal across the channels of interest. Decisional rules are then applied to the suppression segments to identify the actual BS events. Additionally, algorithms were developed to identify EEG containing loose electrodes as well as those with EMG and large amplitude contaminations. The overall BS event detection sensitivity is greater than 92% with a specificity of 83% on data from 4 ICU recordings.

Analysis of EEG signals by combining eigenvector methods and multiclass support vector machines

A new approach based on the implementation of multiclass support vector machine (SVM) with the error correcting output codes (ECOC) is presented for classification of electroencephalogram (EEG) signals. In practical applications of pattern recognition, there are often diverse features extracted from raw data which needs recognizing. Decision making was performed in two stages: feature extraction by eigenvector methods and classification using the classifiers trained on the extracted features. The aim of the study is classification of the EEG signals by the combination of eigenvector methods and multiclass SVM. The purpose is to determine an optimum classification scheme for this problem and also to infer clues about the extracted features. The present research demonstrated that the eigenvector methods are the features which well represent the EEG signals and the multiclass SVM trained on these features achieved high classification accuracies.

Compression of long-term EEG using power spectral density

In this paper, we propose to use the features based on power spectral density as a descriptor of the EEG in the compression of the long-term intensive care unit EEG to obtain the temporal evolution of the recurrent patterns. Sleep EEG is used as a baseline since the sleep stages can be mapped to recurrent patterns in the background EEG. Our results indicate that the spectral features provide a better classification of the sleep EEG and assist in a better formation of homogenous clusters compared to the results obtained with the previously used features. The average overall agreement compared against manual scoring of seven sleep EEG records is 68.5%. It is an improvement compared to 62.7% obtained with the previously used features. Although our results for computer classification use only the EEG information from one frontal and one occipital channel, they are similar to the manual classification of sleep EEG, which is based on additional information.

On evaluation of spectrum estimators for EEG

In the search for how neonatal EEG is affected by asphyxia it is of importance to find reliable estimates of EEG power spectra. Several spectral estimation methods do exist, but since the true spectra are unknown it is hard to tell how well the estimators perform. Therefore a model to generate simulated EEG with known spectrum is proposed and the model is used to evaluate performance of several parametric and Fourier based spectral estimators.

DESIGN AND IMPLEMENTATION OF WIRELESS MULTI-CHANNEL EEG RECORDING SYSTEM AND STUDY OF EEG CLUSTERING METHOD

A multi-channel wireless EEG (electroencephalogram) acquisition and recording system is developed in this work. The system includes an EEG sensing and transmission unit and a digital processing circuit. The former is composed of pre-amplifiers, filters, and gain amplifiers. The kernel of the later digital processing circuit is a micro-controller unit (MCU, TI-MSP430), which is utilized to convert the EEG signals into digital signals and fulfill the digital filtering. By means of Bluetooth communication module, the digitized signals are sent to the back-end such as PC or PDA. Thus, the patient's EEG signal can be observed and stored without any long cables such that the analogue distortion caused by long distance transmission can be reduced significantly. Furthermore, an integrated classification method, consisting of non-linear energy operator (NLEO), autoregressive (AR) model, and bisecting k-means algorithm, is also proposed to perform EEG off-line clustering at the back-end. First, the NLEO algorithm is utilized to divide the EEG signals into many small signal segments according to the features of the amplitude and frequency of EEG signals. The AR model is then applied to extract two characteristic values, i.e., frequency and amplitude (peak to peak value), of each segment and to form characteristic matrix for each segment of EEG signal. Finally, the improved modified k-means algorithm is utilized to assort similar EEG segments into better data classification, which allows accessing the long-term EEG signals more quickly.

Quantitative topographic differentiation of the neonatal EEG

OBJECTIVE

To test the discriminatory topographic potential of a new method of the automatic EEG analysis in neonates. A quantitative description of the neonatal EEG can contribute to the objective assessment of the functional state of the brain, and may improve the precision of diagnosing cerebral dysfunctions manifested by 'disorganization', 'dysrhythmia' or 'dysmaturity'.

METHODS

21 healthy, full-term newborns were examined polygraphically during sleep (EEG-8 referential derivations, respiration, ECG, EOG, EMG). From each EEG record, two 5-min samples (one from the middle of quiet sleep, the other from the middle of active sleep) were subject to subsequent automatic analysis and were described by 13 variables: spectral features and features describing shape and variability of the signal. The data from individual infants were averaged and the number of variables was reduced by factor analysis.

RESULTS

All factors identified by factor analysis were statistically significantly influenced by the location of derivation. A large number of statistically significant differences were also established when comparing the effects of individual derivations on each of the 13 measured variables. Both spectral features and features describing shape and variability of the signal are largely accountable for the topographic differentiation of the neonatal EEG.

CONCLUSIONS

The presented method of the automatic EEG analysis is capable to assess the topographic characteristics of the neonatal EEG, and it is adequately sensitive and describes the neonatal electroencephalogram with sufficient precision.

SIGNIFICANCE

The discriminatory capability of the used method represents a promise for their application in the clinical practice.